Editor-in-Chief Zeev Zalevsky and the journal’s team of senior editors introduce an updated slate of research topic categories appropriate for JEI.

The Editor-in-Chief and the publisher have retracted this article, which was submitted as part of a guest-edited special section. An investigation uncovered evidence of systematic manipulation of the publication process, including compromised peer review. The Editor and publisher no longer have confidence in the results and conclusions of the article.

RD, KKH, BPD, and AKV either did not respond directly or could not be reached.

Aiming at the low efficiency of information processing in a robot vision system, our work studies how to extract fixed points during tasks so that significant and interesting target areas in the scene can be located. First, the architecture of a gaze extraction model of bionic vision is proposed, which consists of a spatial visual saliency model and a task-driven fixed point extraction model. The task-driven fixed point extraction process is represented as a closed-loop control problem. To couple exploration with the sampling process, a Q learning algorithm based on a random strategy is adopted, which is a top-down task-driven gaze extraction in the time dimension. It is fused with the spatial-based visual saliency model to form a spatiotemporal hybrid gaze extraction model to determine the fixed point in the final image. Finally, our qualitative image visualization experiment indicates that the fixed point extraction results of the model in two consecutive frames are close to the truth value obtained by an eye tracker. The quantitative area under the curve and average angular error data confirms the effectiveness of the model in predicting and extracting fixed points.

The Editor-in-Chief and the publisher have retracted this article, which was submitted as part of a guest-edited special section. An investigation uncovered evidence of systematic manipulation of the publication process, including compromised peer review. The Editor and publisher no longer have confidence in the results and conclusions of the article.

QX, GZ, and HC agree with the retraction.

The Editor-in-Chief and the publisher have retracted this article, which was submitted as part of a guest-edited special section. An investigation uncovered evidence of systematic manipulation of the publication process, including compromised peer review. The Editor and publisher no longer have confidence in the results and conclusions of the article.

CZ, JL, and FZ either did not respond directly or could not be reached.

The Editor-in-Chief and the publisher have retracted this article, which was submitted as part of a guest-edited special section. An investigation uncovered evidence of systematic manipulation of the publication process, including compromised peer review. The Editor and publisher no longer have confidence in the results and conclusions of the article.

ZL and JH either did not respond directly or could not be reached.

The Editor-in-Chief and the publisher have retracted this article, which was submitted as part of a guest-edited special section. An investigation uncovered evidence of systematic manipulation of the publication process, including compromised peer review. The Editor and publisher no longer have confidence in the results and conclusions of the article.

XY either did not respond directly or could not be reached.

The Editor-in-Chief and the publisher have retracted this article, which was submitted as part of a guest-edited special section. An investigation uncovered evidence of systematic manipulation of the publication process, including compromised peer review. The Editor and publisher no longer have confidence in the results and conclusions of the article.

LZ agreed with retraction. YS did not agree with the retraction. YL either did not respond directly or could not be reached.

The Editor-in-Chief and the publisher have retracted this article, which was submitted as part of a guest-edited special section. An investigation uncovered evidence of systematic manipulation of the publication process, including compromised peer review. The Editor and publisher no longer have confidence in the results and conclusions of the article.

ZC and LZ agreed with retraction.

The Editor-in-Chief and the publisher have retracted this article, which was submitted as part of a guest-edited special section. An investigation uncovered evidence of systematic manipulation of the publication process, including compromised peer review. The Editor and publisher no longer have confidence in the results and conclusions of the article.

CC, HH, JS, and SL either did not respond directly or could not be reached.

Many people around the world are affected by pulmonary disease as well as asthma, pneumonia, lung cancer, and tuberculosis. All of these conditions have one thing in common: airway obstruction. Lung illnesses are a worldwide issue in which chronic obstructive pulmonary disease, pneumonia, asthma, tuberculosis, and fibrosis are few of the most common. An accurate diagnosis of a lung problem is essential, and it has been pursued by many researchers using image processing and machine learning models. Multiple deep learning methods are used to predict lung disease; these include convolutional neural networks, vanilla neural networks, visual geometry group-based neural networks, and the capsule network. Medical data are scarce, so diagnosing pulmonary diseases using chest x-ray pictures from datasets with less than 1000 samples is considered to address the problem. Three deep learning multiple neural networks (DLMNNs) were generated using the chicken swarm optimization (CSO) approach, for which transfer learning was applied to evaluate the performance of each. First, we created an algorithm for segmenting CXR images, and then we compared it with other classification systems. Our results were compared with those of other methods using publicly available data from the Shenzhen and Montgomery lung datasets. However, our technique has a lower number of trainable parameters compared with the best-performing models trained on the Montgomery dataset. The DLMNN-CSO virtually matched the best performance on the Shenzhen dataset, although it was computationally less expensive than the other models. DLMNN-CSO’s validation loss was 0.4, whereas the validation loss for CNN, VDSNet, VGG, and DBN was 0.9, 0.8, 0.6, and 0.5, respectively.

The widely adopted video stabilization techniques in imaging devices have become a bottleneck that limits the performance of source camera identification (SCI) based on photo-response non-uniform (PRNU) noise. Affine transformations in the stabilized video, including scaling, translation, and rotation, introduce varying degrees of distortion to the PRNU pattern across frames, which lead to a degradation of accuracy and prolonged matching time in SCI methods. To address these issues, we propose a stabilized video source identification method that combines the undecimated dual-tree complex wavelet transform (UDTCWT) with the Bayesian adaptive direct search (BADS) algorithm. First, decoded frames are obtained by skipping the loop filter in the video codec to preserve more PRNU. Then, a PRNU extraction algorithm based on the UDTCWT is utilized to filter out more reliable PRNU from the decoded frames. Finally, the distorted PRNU is optimized and recovered by the affine transformation model with the BADS. Experiments on public datasets show that the proposed method outperforms state-of-the-art methods in terms of matching speed and accuracy.

Detecting pavement cracks from images is a complex computer vision task due to their varying shapes, backgrounds, and sizes. We propose CrackF-Net, an end-to-end convolutional neural network for automatic crack detection in road images. We construct the CrackF-Net network using an encoder–decoder architecture to extract image features in convolutional blocks with residuals and fuse the multiscale convolutional features produced by the decoder. Convolutional blocks with residuals are used to capture the strong semantic features of cracks, and an adaptive filter fusion module is proposed to assist the network make a selection of filter fusion features on the channels. CrackF-Net fuses the multiscale features in decoder to improve crack detection performance. The proposed CrackF-Net is compared to other advanced crack detection methods using three public datasets. The experimental results show that CrackF-Net achieves state-of-the-art performance, which obtains F-measures of 0.866, 0.737, and 0.852 on the three datasets.

Deep-learning-based semantic segmentation is the research focus for unmanned aerospace vehicle (UAV) aerial images analysis. However, there are problems in segmenting small and narrow objects and boundary regions, due to the large size differences between objects and the unbalanced class data in aerial images. A network named SEC-BRNet is proposed for the boundary refinement problem. First, the semantic embedding connections and progressive upsampling decoder are used to obtain spatial details for generating fused feature maps, which are then concatenated in decoding process level by level for recovering the boundary details. Second, a multiloss training strategy is developed for data imbalance and boundary roughness problems, including cross-entropy loss, Dice loss, and active boundary loss. After extensive experiments, our network could achieve 84.8% mIoU and 89.04% Boundary IoU on the AeroScapes dataset and achieve 62.81% mIoU and 90.78% Boundary IoU on the Semantic Drone Dataset. The experimental results indicate that the proposed SEC-BRNet performs well in semantic segmentation task for UAV aerial images.

Color constancy (CC) refers to the ability to reproduce the veritable color of objects by eliminating the influence of environmental illumination. The performance of CC has been greatly boosted by advanced learning-based methods, such as deep neural networks. However, most current deep learning methods suffer from large model sizes and large-scale training datasets for promising performance. We propose least square support vector machine-CC (LSSVM-CC), an efficient CC method using LSSVM, which combines preferably unitary statistics-based methods to achieve better illuminant estimation performance at low computation efforts with small training samples. The core of our LSSVM-CC is to cluster training images based on illumination-preferred features and then train an ensemble model using LSSVM for each cluster, which can effectively improve the overall performance. Once trained, our LSSVM-CC selects adaptive models for different input images to perform accurate illuminant estimation in the testing phase. Experiments on typical benchmark datasets show that our method significantly outperforms most traditional statistics-based methods and is also comparable to some sophisticated and computationally demanding learning-based methods.

Autoencoder (AE) is widely used in image fusion. However, AE-based fusion methods usually use the same encoder to extract the features of images from different sensors/modalities without considering the differences between them. In addition, these methods cannot fuse the images in real time. To solve these problems, an end-to-end fusion network is proposed for fast infrared image and visible image fusion. We design an end-to-end W-shaped network (W-Net), which consists of two independent encoders, one shared decoder and skip connections. The two encoders extract the representative features of images from different sources respectively, and the decoder combines the hierarchical features from corresponding layers and reconstructs the fused image without using additional fusion layer or any handcrafted fusion rules. Skip connections are added to help retain the details and salient features in the fused image. Specifically, W-Net is lightweight, with fewer parameters than the existing AE-based methods. The experimental results show that our fusion network performs well in terms of subjective and objective visual assessments compared with other state-of-the-art fusion methods. It can fuse the images very fast (e.g., the fusion time of 20 pairs of images in the TNO dataset is 0.871 to 1.081 ms), operating above real-time speed.

Powerful image editing software makes the process of image manipulation easy, which increases security risks. Therefore, it is urgent to locate the tampered region to uncover the processing history. However, previous research has mainly focused on feature extraction, with little discussion on classifiers for classifying original and tampered regions. We improve the splicing forgery localization method from a statistical perspective. The refined color filter array feature provides sufficient data for statistical analysis, and the geometric mean is used to eliminate anomalous data. Subsequently, a classifier that combines the expectation–maximization algorithm and Bayesian theory is proposed to binarize the original and tampered regions. The two steps of feature extraction and feature classification are associated from a statistical perspective, which ultimately improve the performance of the method effectively. Extensive experimental results demonstrate that the refined feature used for classification has several advantages, and the proposed classifier is appropriate for handling complex image manipulation across different statistical distributions. The proposed method outperforms the reference methods in both the Columbia and Korus datasets.

Due to the small size, high density, and background noise associated with strip surface defects, the current object detection model commonly faces limitations in performance. To address this issue, we propose a spatial-to-depth feature-enhanced detection method called STD-Detector. The method consists of two types STD-Conv-A and STD-Conv-B. First, the STD-Conv-A module is used in the backbone feature extraction network to expand the field of perception and enable the model to learn a wider range of background information. Then, the STD-Conv-B module is used for feature fusion networks to improve the expression of output features. In addition, we incorporate the convolutional block attention module to mitigate background interference and enhance the performance of the model. Finally, experimental results on the NEU-DET dataset show that our method achieves a mean average precision of 82.9%, which represents a 3.9% improvement over the baseline. Compared with the state-of-the-art model, our method exhibits greater competitiveness in detecting strip surface defects. Moreover, experimental results on the road surface defect dataset show that our method has good robustness.

Image inpainting is promising but challenging in computer vision tasks; it aims to fill in missing regions of corrupted images with semantically sensible content. By utilizing generative adversarial networks (GAN), state-of-the-art methods have achieved great improvements, but the ordinary GAN generally suffers from difficulties in training and unstable gradients, leading to unsatisfactory inpainting results. Image-level predictive filtering is a widely used restoration method that adaptively predicts the weights of pixels around a target pixel and then linearly combines these pixels to generate the image, but it cannot fill larger missing regions. Thus, we extend image-level predictive filtering to the deep feature level through an encoder–decoder network and embed adaptive channel attention and spatial attention modules in the encoder network. We use Wasserstein GAN instead of normal GAN due to its superior properties and then combine it with image-level predictive filtering and deep feature-level predictive filtering, which ultimately leads to a significant improvement in image inpainting. We validate our method on two public datasets: CelebA-HQ and Places2. Our method demonstrates good performance across four metrics: peak signal-to-noise ratio, L1, structural similarity index measure, and learned perceptual image patch similarity.

When conducting image classification, traditional and deep-learning based methods require extracting target features in different ways and then classifying targets based on the features. Therefore, extracting more informative and effective features is very important, and this problem also exists in plant leaf classification. To address the issue for traditional leaf classification, this article carefully designs two shape features [area ratio-based shape descriptor (ARSD) and corner-based shape descriptor (CSD)] and one leaf vein feature (LVF) extraction method based on statistical characteristics. Then, according to the extraction of ARSD, CSD, and LVF, a general feature extraction framework is formed, which first normalizes direction, segments region of interest, then divides it into subregions, counts the number of pixels in each sub region, and finally forms feature vectors. Finally, to visually verify the effectiveness of these features, they were applied to the classic leaf classification framework, which is adjusted to form a two-stage classification method from coarse classification of shapes and fine classification of species. In the first stage (coarse classification), leaves in three universal datasets [Flavia dataset, Swedish dataset and Intelligent Computing Laboratory (ICL) dataset] and a self-built LZU dataset are respectively divided into several shape categories only by the ARSD. Experiment results show that the ARSD has obvious advantages in shape-based rough classification. Then in the second stage (fine classification), shape, texture, vein, and color features, including ARSD, CSD, local binary patterns, gray-level co-occurrence matrix, the developed LVF, and color moments, are combined to subdivide finer species in each shape category. Extensive experiment results show that the classification accuracy on the Flavia, Swedish, ICL, and LZU datasets have reached 99.3%, 98.9%, 91.5%, and 95.3%, respectively, comparable to deep learning-based methods.

The addition of objects to the original image in splice forgery results in a change in the semantics of the original image, and distribution of these spliced images may bring negative impacts. To solve this issue, many forgery detection methods based on convolutional neural networks are presented. However, they tend to extract deep features, but ignore the importance of shallow semantics. In addition, the complexity of the forgery localization task leads to insufficient accuracy in detecting smaller forged regions. Based on the above problems, an end-to-end image splicing localization network based on multi-scale features and a residual refinement module (RRM) is proposed in this work. Our approach can be roughly divided into two modules: the detection module and the RRM. First, shallow and deep features are extracted from a backbone network, and deep features are then processed by the deeper atrous spatial pyramid pooling (ASPP) module to extract multi-scale features. The deeper ASPP module uses a smaller dilation rate and light convolution, which is more suitable for detecting complex counterfeit images. Second, the shallow features are fused with the multi-scale features to complement the shallow semantic information, such as texture and edges, which further improve the robustness of the model. Finally, the detection network generates coarse prediction maps that are fed to the RRM, and the RRM optimizes these masks by smoothing the boundaries, filling small gaps, and enhancing the edge details, which improves the pixel-level segmentation results for forgery detection. Extensive experimental results on several public datasets show that this method outperforms other state-of-the-art methods in image forgery localization.

Many working-age people suffer from diabetic retinopathy (DR), one of the chronic retinal conditions caused by diabetes that eventually results in blindness. The classification of the DR severity level has been an arduous task due to the complexity of the lesion features. An efficient detection technique is needed for the screening procedure to categorize the retina’s subtle pathologies. Deep neural networks are essential for identifying eye diseases and enabling ophthalmologists to treat patients immediately. To categorize the severity of DR utilizing images from the general IDRiD, MESSIDOR, and KAGGLE datasets, our research provides an effective hybrid optimized deep-learning network (HODLNet) model. The quality of the input fundus image is enhanced by performing preprocessing initially with noise removal and contrast enhancement. After that, to segment the blood vessels and optic disc, a modified ResUNet model is used. The features are then extracted after the lesion region has been found using the Gabor filter banks. An improved ShuffleNet v2 is used for the final classification process. According to the experimental results, the proposed HODLNet model produces higher results with a 99.84% accuracy for multiclass classification than the existing deep-learning models for MESSIDOR, IDRiD, and KAGGLE datasets.

With the increasing global concern regarding public health, accurate diagnosis and treatment of diseases have become critical. In the context of liver computed tomography (CT) image diagnosis, obtaining precise liver segmentation output samples can save consultation time and reduce the risk of misdiagnosis. We propose a full-scale connected attention-aware segmentation network, called Attention UNet3+. To fully leverage semantic information at different scales, we redesign the depth supervised decoder and adopt a full-scale skip connection, which can effectively extract features from different layers thus increasing accuracy. The proposed Attention UNet3+ model uses an attention gate connection instead of the skip connection, which effectively suppresses irrelevant regions and highlights salient features of specific local regions during feature extraction, therefore, improving the segmentation accuracy. Additionally, the classification-guided module enhances the liver boundary and reduces over-segmentation of non-liver regions, obtaining accurate segmentation results. Our experimental evaluation on the medical image computing and computer assisted intervention Liver Tumor Segmentation Challenge 2017 dataset shows that the proposed Attention UNet3+ outperforms other improved UNet algorithms for liver image segmentation by a minimum of 2.9% in intersection over union and a minimum of 1.1% in Dice.

Although learned image compression methods have achieved competitive rate-distortion performances, learned video compression remains challenging. The current mainstream learned video compression frameworks usually improve the motion prediction module to reduce the redundancy in video sequences. Although these methods can achieve a great compression ratio, they often ignore the improvement in the entropy model, which does not fully exploit the temporal and spatial characteristics in the video. Meanwhile, these methods always suffer from the error propagation problem. To solve those problems, we propose a learned video compression framework with channel-wise autoregressive entropy model. Our framework captures spatial–temporal dependencies through a powerful entropy model to reduce the redundancy in video sequences. In particular, we do not directly compress the frames in the pixel domain, with no need for the motion prediction module, avoiding the error propagation problem. To better utilize the temporal contexts of the previous frame, we propose the window temporal prior module. Experiments show that our proposed video compression framework achieves promising compression effects in terms of peak signal-to-noise ratio and multiscale structural similarity.

Existing methods for fusing infrared and visible images prioritize the fusion effect at the expense of the model size and inevitably tend to be more oriented toward infrared images during fusion, which results in fused images that can lack the texture detail information of visible images. Therefore, a new feature gradient attention block is designed in our model to extract the texture detail of the image, in which the module extracts the gradient information of the original image while extracting features, then uses depthwise separable convolution to optimize and enhance the information with rich edge features. To preserve the original image information, we also use short links to reference previous features. Since the important features are strengthened in the feature extraction stage, we design an adaptive weight energy attention network based on the energy fusion strategy in the fusion stage to further preserve the thermal radiation area of the infrared image and the spatial details of the visible image. The proposed method is experimentally verified on the public visible-infrared paired dataset for low-light vision and the TNO dataset, and six objective evaluation indicators are used to prove that our model is better to the existing fusion algorithms. In addition, we further verify the effectiveness of the proposed method for high-level vision task models by object detection and semantic segmentation models.

Traditional rain models typically focus on modeling images based on the physical characteristics and generation of rain streaks, which often obtain good results in the synthetic dataset. However, in real-world scenarios, the transformation of the camera depth of field (DOF) and the focusing position lead to different degrees of blur in different areas of the image. This variability poses challenges in extracting and processing rain streaks far from the camera DOF. Furthermore, the current image rain removal model has grown increasingly complex, which makes the image rain removal task difficult to make real time for practical applications. To address this challenge, we take a novel approach by considering the camera imaging principle. We explore the potential depth information concealed within regions of image blurriness, introducing a raindrop model and proposing the progressive depth information network (PDIN). To be specific, PDIN concurrently learns both the direction characteristics and depth information of the rain streaks and gradually removes the rain streaks in a cyclic way. Our method innovatively considers the blur problem of rainfall images from the perspective of camera shooting. Importantly, our model demonstrates excellent rain removal performance with a relatively low time complexity. A large number of experiments on synthetic and real datasets show that PDIN has better performance in image rain removal, which also indicates the rationality of our new model.

Extracting dense rotating objects accurately from remote sensing images is an emerging task in object detection. To increase the applicability of existing algorithms in the above tasks, an optimized anchor-free network optimized by a dual attention mechanism (DAM) and gate multiscale feature fusion (GMFF) is designed. The DAM module is composed of two attention mechanisms with different functions. This part can enhance the backbone network’s ability to extract and model information at different levels and reduce the accuracy loss caused by object density changes in the image. The GMFF module uses the gating structure to realize adaptive transmission and integration of multiscale information. Through this module, the useless information in features will be filtered, and the key information will be retained. Several experiments are designed to verify the feasibility of the algorithm. Compared with the baseline model, adding DAM and GMFF to the dense rotating object extraction task in remote sensing images improves the model accuracy by 3.5% and 2.1%, respectively, while adding two modules simultaneously, and the accuracy increases from 79.1% to 84.3%. In conventional object extraction tasks, such as dataset for object detection in aerial images and HRSC2016, our method has the highest accuracy compared to other similar algorithms, with 76.5% and 90.3%, respectively.

With the expanding scale of underground cable pipelines, the stable operation of underground power grid is essential for the orderly development of human production and life. However, once there are foreign objects, defects, or other anomalies in pipelines, it will lead to a series of problems, such as electric discharge, trip, and fire, seriously threatening the life and property safety of surrounding residents. Thus the research on anomaly detection of underground cable pipelines plays an important role. To address the challenges of limited anomaly samples and complex detection, an anomaly detection method for underground cable pipelines is proposed based on reconstruction error. Prior to training, a pseudorandom anomaly generation module is employed to create random anomaly shapes and pair them with random anomaly textures. The anomaly images and corresponding detection result maps produced by this module facilitate self-supervised training and improve the generalization performance of anomaly detection, effectively addressing the dynamic nature of anomalies within the pipeline. This model primarily consists of two encoder–decoder architecture networks, with the first network incorporating the convolutional block attention module in its middle to enhance the feature extraction capability during image reconstruction. The experiments on public dataset MVTec AD and self-build dataset HHU_UP verify the effectiveness and robustness of the proposed method compared with some current existing anomaly detection methods.

With the remarkable advances of unmanned aerial vehicles (UAVs) and machine vision, aerial tracking has attracted wide attention from scholars. Previous tracking methods were mostly implemented in clean and well-lit environments, making it challenging to track camouflaged people rapidly and accurately in woodlands. We develop a framework for camouflaged people aerial tracking (CPAT) based on transformer. Specifically, a camouflaged people discovery strategy is proposed to rapidly generate training samples from the unlabeled videos captured by the UAV. Dynamic programming is also employed to filter noises to generate smooth candidate frames. To exploit multilevel feature information, a transformer fusion framework is designed to integrate shallow spatial information and in-depth semantic features. For reducing computing consumption, the spatial attention reduction mechanism is embedded in the multihead attention for fast tracking. Further, we build a dataset for evaluating the effect of camouflaged people tracking called Cam235, which consists of 85 manually labeled test sequences and more than 100k frames of the unlabeled training set. Exhaustive experiments on Cam235-test and popular tracking datasets prove that the CPAT is superior to other trackers for practical application. Under the most challenging condition of camouflaged people tracking, the CPAT achieves the precision of 67.9%, surpassing the state-of-the-art trackers by large margins.

Aiming at the detection difficulties in camouflage target detection, such as the high similarity between the target and its background, serious damage to the edge, and strong concealment of the target, a camouflage target detection algorithm YOLO of camouflage object detection based on strong semantic information and feature fusion is proposed. First, the attention mechanism convolutional block attention module (CBAM) is constructed to highlight the important channel features and target spatial locations to further aggregate rich semantic information from the high-level feature map. Then the atrous spatial pyramid pooling module is constructed to repeatedly sample the multiscale feature maps to expand the receptive field of the neural network, reduce feature sparsity in the process of convolution, and ensure dense features and multiscale contextual semantic information enter the feature fusion module. Finally, the attention skip-connections are constructed based on the CBAM module for fusing the original feature maps extracted by the backbone network to the corresponding detection outputs so as to eliminate the redundant features as well as enrich the target information of the network outputs. In order to fully verify the performance of the proposed algorithm, a camouflage target detection dataset named strong camouflage efficiency target dataset (SCETD) is constructed. Experimental results on SCETD show that the precision and recall of the proposed algorithm achieve 96.1% and 87.1%, respectively. The AP_0.5 and AP_{0.5 : 0.95} achieve 92.3% and 54.4%, respectively. The experimental results prove the effectiveness of the proposed method in detecting camouflage targets.

Abnormal event detection in computer vision addresses the task of identifying events that deviate from expected behavior in video scenes. Issues, such as occlusion in crowded scenes, the powerful generalization capabilities of deep neural networks, and the heavy reliance on contextual information, make this task particularly challenging. To address these issues, we propose a cascaded form of abnormal detection framework that combines the paradigms of reconstruction and prediction in this paper. First, stochastic masking techniques are employed for image reconstruction to alleviate the overgeneralization of neural networks under abnormal conditions. Second, an innovative motion characterization of frame-difference streak streams is introduced to better characterize the motion of video frames in crowded scenes. Finally, a dual-channel autoencoder-based prediction network is introduced to jointly learn appearance and motion features. This network captures contextual information to better generate predictive features. Meanwhile, adversarial learning is introduced for abnormal inference to improve the detection performance. Experimental results on several benchmark datasets validate the effectiveness of our approach.

Ensuring reliable and steady operation of power equipment is paramount to safeguard the livelihoods and labor of the populace. However, traditional detection techniques face obstacles when adapting to the intricate background of transmission lines, leading to innumerable incorrect and missed detections. To resolve these issues, an improved YOLOv7 transmission line insulator defect recognition algorithm has been introduced. An anchor frame, matching the size of the transmission line fault, has been constructed using K-means++, followed by a convolutional block attention module to enhance the ability to extract features. Next, the wise-IoU loss function has been incorporated, providing a gradient gain allocation strategy, thereby enhancing the positioning performance and speed of the model. Finally, the SiLU activation function has been replaced with meta-ACON to enhance the adaptive feature capability of the YOLOv7 network. Experimental results have shown that the proposed method has an average accuracy (mAP) of 91.8%, and the detection accuracy for transmission lines can be improved to 98.8%. This model resolves the persisting issues of erroneous and missing detections by addressing the traditional technical difficulties of detecting complex backgrounds and defects with insufficient accuracy.

Recent studies have shown that deep neural networks are sensitive to visual blur. This vulnerability raises concerns about the robustness of learning-based methods in safety-critical applications, such as autonomous driving and industrial inspection. Although many methods have demonstrated the potential to enhance the robustness of vision tasks in blurry scenes, a standard benchmark for comparing and validating their effectiveness is still lacking. To address this, we propose a new benchmark for the comprehensive evaluation of visual recognition tasks in blurry scenes. The benchmark dataset includes a large number of scenes with different blur types, enabling multi-dimensional evaluation and comparison of existing methods. The existing methods are integrated into a unified evaluation framework for comparison, and the mechanism of robustness improvement is roughly categorized as reducing image-level shift and distribution-level shift. Through this unified framework, we identify the strengths and weaknesses of current methods in visual recognition tasks under blurry conditions. Furthermore, we propose a prior knowledge-based regularization module that can be easily incorporated into existing methods, which consistently boosts performance on visual recognition tasks in blurry scenes.

Vehicle detection is a fundamental problem in object detection and plays a significant role in intelligent transportation and smart driving. To enhance the accuracy of vehicle detection and the robustness of the model in detecting occluded vehicles, we propose an improved vehicle detection method for you only look once v5s (YOLOv5s). First, we introduce the coordinate attention module into the backbone of the model. This module guides the model to improve its attention toward the location information of vehicles and channel features under occlusion conditions. Second, the feature fusion component of the model is improved by incorporating bidirectional scale connections and weighted feature fusion. Finally, the prediction head of YOLOv5s is decoupled and the regression and classification tasks are assigned to two separate branches. Experimental results show that our proposed method has 2% and 2.5% higher average precision than YOLOv5s for the common objects in context vehicle dataset and University at Albany Detection and Tracking, respectively.

Partial least squares regression (PLSR) is a linear regression model suitable for handling data with high dimensions and small samples. However, a large amount of nonlinear structure data exists in real life, and PLSR may be not good at handling this type of data. Motivated by this, we propose a parameter-free, nonlinear PLSR model called estimating optimal transformations PLSR. First, we use the EOT model to transform a nonlinear data structure into a linear data structure without additional parameters by maximizing the correlation between data samples. In addition, the traditional PLSR model often uses a greedy algorithm, leading to suboptimal solutions. To obtain more accurate numerical solutions, we propose using manifold optimization for the PLSR model. Compared to the other existing PLSR models, our proposed model achieves lower classification error rates on several different types of datasets. The experimental results further confirm the importance of our proposed nonlinear data transformation and manifold optimization for feature extraction with nonlinear structure data.

The rapid development of Deepfake technology has posed significant challenges in detecting fake videos. In response to the existing problems in reference frame selection, spatial–temporal feature mining, and fusion in face-swapping video detection techniques, we propose a face-swapping video detection model based on spatial–temporal feature fusion. First, key frame sequences are selected using interframe facial edge region differences. Then, the key frame sequences are separately input into the spatial branch to extract hidden artifacts and the temporal branch to extract inconsistent information. Finally, the spatial–temporal features are fused using a self-attention mechanism and input into a classifier to achieve detection results. To validate the effectiveness of the proposed model, we conducted experiments on the Faceforensics++ and Celeb-DF open-source Deepfake datasets. The experimental results demonstrate that the proposed model achieves better detection accuracy and higher-ranking generalization performance than state-of-the-art competitors.

The depth image based rendering (DIBR) technique plays a crucial role in free viewpoint video. However, the DIBR technology is not yet mature and still needs to be improved. In addition, depth image acquisition is inaccurate and prone to distortion, which affects human visual perception. The DIBR-synthesized image quality assessment (IQA) method, which simulates the human visual perception system, can be used as feedback to improve the DIBR technique and depth video quality. The existing IQA algorithms are insensitive to the diversity of geometric distortion produced by the DIBR technique and do not consider the impact of geometric distortion on the image quality from multiple perspectives. In this work, a no-reference IQA method for DIBR-synthesized images based on multi-feature fusion is proposed. Different from the existing IQA algorithms, this paper focuses on the diversity of geometric distortions in DIBR-synthesized images, measures the distortions of the edge region and the hole region, and integrates the global sharpness and the natural mixed distortion. Specifically, the global sharpness is first measured, and the image is divided into blocks to further improve the accuracy of the results. Secondly, the geometric distortions of the edge and hole regions are measured comprehensively. In addition, considering the impact of the natural mixed distortion on the quality of the DIBR-synthesized image, a Gaussian mixed statistical model is used to extract the natural mixed distortion of the DIBR-synthesized image. Finally, the quality score of the DIBR-synthesized image is generated by pooling regression. Experimental results on two synthesized image databases demonstrate that the method proposed in this paper outperforms classical and state-of-the-art IQA methods and is more aligned with the human visual perception system.

Authentication systems are now an important part of our daily life. Human biological, behavioral, and physical characteristics are usually applied in authenticating a person in various applications. Unimodal biometric systems have a number of limitations, such as noise sensitivity, population coverage, intra-class variations, non-universality, and vulnerability to spoofing. Multimodal biometric systems overcome these limitations and are being widely used in many real-world applications. In this work, to construct a three-dimensional (3-D) shell, we use fingerprint and iris. First, features are extracted from the fingerprint. Then, using a user key set, a two-dimensional spiral curve is generated from fingerprint features. Next, iris features are extracted using a pre-trained VGG-16 model, then feature vector-based random projection is applied to generate an iris feature vector. This generated feature vector is combined with the fingerprint shell to construct a secured 3-D shell. Finally, these fused 3-D templates are saved in the database and are used for matching. Our proposed technique has been evaluated on the three publicly available datasets, showing that it can preserve user privacy while maintaining the accuracy of the system with an equal error rate of 0.09%, 0.032%, and 0.015%.

Defect detection based on deep learning has been applied in many fields and has made great progress in recent years. However, the problems of high false detection rate in complex scenes and weak multiscale detection capability have not been solved well. To solve the above problems, a damper defect detection network (DDDNet) is proposed, which includes a cognitive preprocessing network (CogniPrepNet) and a heterogeneous feature pyramid network (HeteroFPN). CogniPrepNet enhances targets and mitigates the impact of complex backgrounds through the ghost-shuffle module. HeteroFPN obtains feature maps with a wider range of scales through a many-to-one mapping relationship to achieve high-quality multiscale feature fusion. Moreover, a fast proportional intersection over union is designed to improve the sensitivity of geometric factors during the bounding box regression. The experimental results on damper-DET dataset show that the DDDNet can achieve 98.25% mAP and 34 FPS detection speed in complex background and multiscale target scene.

Light field image (LFI) processing is a hot topic of current study; however, light field (LF) cameras’ poor sensor resolution forces a trade-off between spatial resolution and angular resolution. We propose an angular reconstruction network based on decoupled fusion and angular attention to achieve a dense LF with high spatial and angular resolution simultaneously. The angular self-attention module of the network is designed to capture the correspondence between different views and deeply refine the effective angular information. The network may obtain the related angular information non-locally and rebuild the dense LF by decoupling fusion of spatial and angular information. Furthermore, the feature information of the LFI in the spatial, angular, and epipolar plane image dimensions is entirely dissociated and merged, resulting in a higher perceptual quality of the reconstructed view. The design refinement module utilizes the correlation between spatial and angular information to recover the texture details of the LFI. Experimental results show that the method outperforms several current state-of-the-art angular reconstruction methods on a variety of LF datasets.

Contrastive learning for unpaired image-to-image translation utilizes adversarial loss to ensure the realism of generated images in the target domain and incorporates pixel-wise contrastive loss to maximize the correlation between them. However, existing methods only contrast pixel-wise features, ignoring higher-level features, and the pixel-wise contrast is imperfect, which leads to poorer perceptual and visual results. In order to alleviate these problems, we propose an effective multilevel contrast strategy for unpaired image-to-image translation (MLCUT), which contrasts features at three levels to generate more harmonious and realistic images. Specifically, we strengthen the pixel-wise level contrast and introduce the contrasts of plane and voxel-wise levels. On the one hand, MLCUT enhances training effectiveness by picking over hard negative keys for each query at the pixel-wise level. On the other hand, we strengthen the learning preferences of generators on features of objects rather than backgrounds by contrasting the plane-wise discriminative matrices in adversarial loss. Furthermore, by contrasting voxel-wise global semantic vectors, MLCUT effectively improves the realism of generated images and avoids mode collapse. Qualitative and quantitative experiments demonstrate that our method effectively improves performance in perception and vision.

Homogeneous data fusion plays a crucial role in multiview human action recognition (MvHAR). However, most existing methods tend to employ the concatenate strategy, ignoring the underlying correlation among views and degrading the recognition performance. To this end, we propose a practical MvHAR framework based on a deep discriminant analysis network that excavates multiview video features to obtain a more discrimination representation, from which the view correlation can be explored. Specifically, the spatial–temporal features of multiview data are extracted by the convolution network, and then a deep multiview feature fusion network is proposed to project these features into an advanced subspace for efficient fusion. Traditional methods can lead to class overlap problems, but, to avoid this problem, we improve the separation between two classes by using pairwise between-class scatter. Experiments on five benchmark datasets indicate the efficiency of our framework compared with the advanced algorithms under four metrics.

With similar textures, dark backgrounds, and complex scenes, RGB images are usually unable to provide discriminative information for model training, which often leads to inaccurate prediction results. Compared with RGB salient object detection (SOD) methods, RGB-T SOD have thermal infrared (TIR) information as an informational supplement. As a result, RGB-T SOD can adapt to more complex environments and achieve better results. However, existing methods do not efficiently integrate features between different modalities and do not fully exploit spatial information from the shallow-level features. Accordingly, we propose an EDGE-Net. First, we propose an edge extraction module to capture the edge information in the shallow-level features and use it to guide the subsequent decoding. The original edge features are then weighted after channel attention processing. Second, to additionally suppress the noise of the shallow-level features, we design a global information extraction module. In this module, multiple convolutions are used instead of single convolutions to reduce the computational effort, and convolutions with different dilation rates are used to obtain different receptive fields. We conduct extensive experiments on the RGB-T dataset and show that the proposed method achieves superior performance compared to several state-of-the-art algorithms. The code and results of our method are available in a Github repository available at: https://github.com/BorreloadD/EDGE-Net.git.

Most traditional loop closure detection (LCD) methods rely on manual feature design, which is sensitive to environmental conditions. Convolutional neural networks (CNNs) cope better with illumination changes by extracting hierarchical features and ignoring the local spatial characteristics of images. We propose an LCD algorithm that combines VGG16, NetVLAD, and image pyramids to enhance its accuracy and robustness. In particular, a three-level image pyramid was constructed via downsampling, and then a feature pyramid (FP) layer was obtained by extracting features through VGG16 network on different image resolutions. The obtained FPs were then passed into the VLAD model, and this model outputted VLAD vectors by performing residual summation with L2 normalization. Finally, a triplet loss function was employed for training. Experimental results on two benchmark datasets and a real scenario dataset demonstrated that this algorithm outperforms the NetVLAD baseline and the VGG16 network, exhibiting superior feature-learning capabilities and achieving a higher LCD accuracy. Further, it maintained real-time performance with only a 2% increase in processing time. The results indicate that the proposed method detects loop closures even in complex environments with varying conditions and perspectives. Hence, the approach can be used for large-scale visual simultaneous localization and mapping applications, such as autonomous driving, where LCD plays a crucial role in mapping.

Underwater trash detection faces many problems. For example, the resolution of underwater trash datasets is generally low, and trash may show irregular shapes, sizes, and scales, especially small objects that are difficult to detect. To solve the above problems, we propose UTD-YOLO, an improved YOLOv5s trash detection model. This model proposes a cross-layer aggregation spatial dimensionality reduction module (CASDRM) and channel-PAN-FPN with improved channel information. CASDRM preserves the information of the channel dimension and utilizes the receptive field between different layers to enhance the feature extraction ability while strengthening the ability to adapt to the geometric changes of objects. Channel-PAN-FPN realizes two-way information transmission and improves the recognition accuracy of the model. UTD-YOLO can achieve good recognition results for smaller-scale targets in low-resolution datasets. The experimental results show that the UTD-YOLO model improves the mAP50 to 73.2% on the Trashcan dataset, which is higher than the original YOLOv5 algorithm.

In computer vision, image recognition is one of the classic tasks. Currently, with the foundation of big data and advanced hardware, deep learning has achieved high accuracy. However, deep learning often fails to perform well when faced with a small number of samples. Therefore, few-shot learning has become a key technology to solve this problem. The learning paradigm of few-shot learning is different from that of deep learning. It aims to learn a universal representation from multiple training categories, used for recognition in new categories. Each few-shot learning training instance consists of a group of images and an unlabeled sample. The goal is to enable the model to perform well in recognizing new categories. To achieve this, the model needs to extract representative and highly generalizable features that enable the correct recognition of new category samples. To address the problem of small sample space being unable to describe enough dataset’s semantic features, we propose the attention mechanism and earth mover’s distance for few-shot learning (AMEMD-FSL) method. First, we fuse the attention mechanism (AM) to deep learning to help the model extract more semantically rich features. Then we use the earth mover’s distance (EMD) metric method to calculate the distance between samples, enabling better classification. Finally, we combine the deep-learning residual network and AMEMD to perform few-shot learning. We validate our algorithm on the Caltech-UCSD Birds-200-2011 dataset and the few-shot public dataset mini-ImageNet, which comes from the DeepMind team. The experimental results demonstrate that we have proposed an end-to-end and effective method in the field of few-shot image classification.

Erratum corrects affiliation errors.