Ship detection is an important and challenging task in the field of synthetic aperture radar (SAR) image processing. Recently, deep learning technologies have yielded superior performance for object detection in remote sensing images. However, it is difficult to obtain the labels of SAR images, which limits the application of deep learning in ship detection from SAR images. To break the limitation of label information, we propose a self-supervised framework based on self-distillation for ship detection from SAR images in this paper. The framework consists of three core components: a self-supervised learning paradigm utilizing knowledge distillation, a deep residual shrinkage network (SAR-DRSN) model, and an oriented bounding boxes progressive generation model. The core of our method is a self-supervised variant of knowledge distillation, which propels the deep learning process in the absence of labeled data. The SAR-DRSN model excels in generating high-quality feature maps, significantly reducing the speckle noise. In addition, we introduce an iterative strategy for the accurate and precise delineation of ships, involving continuous refinement of oriented bounding boxes to optimize size and rotation angle for precise ship localization. Our experiments, obtained on two SAR datasets, demonstrate that the proposed method can achieve a satisfactory performance in ship detection without requiring any label information.
In recent years, methods based on convolutional neural networks (CNNs) have achieved significant results in the problem of target classification of synthetic aperture radar (SAR) images. However, the challenges of SAR image data labeling and the characteristics of CNNs relying on a large amount of labeled data for training have seriously limited the further development of this field. In this work, we propose an approach based on attention mechanism and feature complementary fusion (AFCF-CNN) to address these challenges. First, we design and construct a feature complementary module for extracting and fusing multi-layer features, making full use of limited data and utilizing contextual information between different layers to capture more robust feature representations. Then, the attention mechanism reduces the interference of redundant background information, while it highlights the weight information of key targets in the image to further enhance the key local feature representations. Finally, experiments conducted on the moving and stationary target acquisition and recognition dataset show that our model significantly outperforms other state-of-the-art methods despite severe shortages of training data.
Remote sensing scene classification has received more and more attention as important fundamental research in recent years. However, the redundant background information and complex spatial scale variability of remote sensing scene images make the existing convolutional neural network models, which mainly concentrate on global features, perform poorly. To effectively alleviate these problems, we proposed an MSRes-SplitNet model based on multiscale features and attention mechanisms for remote sensing scene image classification. First, MSRes blocks are constructed for the extraction of multi-scale features. Then, the multi-channel local features are fused by the Split-Attention block. Finally, the global and local feature information is aggregated by convolution, thus obtaining multi-scale features while alleviating the small-sample learning problem. Experiments are conducted on three publicly available datasets and compared with other state-of-the-art methods, showing that the proposed method MSRes-SplitNet has better performance while effectively reducing a large number of parameters.
It is small footprint, simplicity, and inexpensive that the direct modulation of a semiconductor laser to generate optical frequency combs (OFCs). However, their OFCs spectral characteristics are heavily dependent on the modulation signal waveform and the current parameters of the laser. It is essential that the research of more useful modulation signal and driving conditions to optimize the OFC performance. We evaluated the performance of the OFCs under gain-switched distributed feedback (DFB) lasers with several common modulation waveforms. It is showed that the relaxation oscillation frequency (ROF) was a restricted condition on the coherence of the OFC under the gain-switching (GS) mode when modulation frequency lower than ROF. We demonstrated that OFCs can be realized in a gain-switched DFB laser with a narrow pulse-width modulation signal under a lower modulation frequency than the ROF of the laser. The OFCs was realized with a comb spacing of 100 MHz, spectral width of 2 GHz, and carrier-to-noise ratio of 29.01 dB by a narrow pulse modulation signal with modulation frequency of 50 MHz at ION = 38.72 mA, IOFF = 0.12 mA, pulse width of 1 ns. The performance of the OFCs under the narrow pulse and sinc modulation signals with different modulation frequency parameters and current driving conditions is further evaluated. This method of tuning the pulse modulation signals to achieve optimization of OFC performance has the advantages of simple operation, flexible settings, and inexpensive. This will have greater application value in high-resolution spectroscopy detection, especially a dual comb spectrum technology.
Hollow waveguide (HWG) as an absorption cell for gas spectral sensing has the advantages of low transmission loss, fast response speed, and high path length to volume ratio. However, the transmission of laser beams in HWGs relies on the multiple reflections by the inner wall, and it makes the path length of laser beams emitted from HWGs unequal, thereby limiting the high-precision measurement of gas concentration. Two mathematical models were established based on geometric optics to characterize the effective path length ratio (EPLR) distribution of the laser beam emitted from a straight HWG and a bent HWG, respectively. The effects of HWG parameters and incident conditions on EPLR distribution was investigated, and quantitative analysis was carried out. Experimental verification was performed by basing heterodyne interferometry. A formula to calculate the equivalent path length ratio of a laser beam propagating in an HWG was given, which simplifies the complicated calculation caused by the path length not unique and demonstrates 1-6% reduction in measurement error. The proposed method has guiding significance for high-precision measurement of absorption spectroscopy, and can extend to both substrate-integrated HWG (iHWG) and liquid waveguide capillary cell (LWCC) based optical sensors.
Deep capsule networks have more capsule layers, which makes their performance better on complex images. However, with the increase of layers, overfitting will become more serious. Image reconstruction is an effective regularization method for capsule networks. To improve it, we propose an adversarial decoder that introduces the generative adversarial network framework into the reconstruction process to implement learnable reconstruction losses. This architecture consists of three parts: a deep capsule network, a decoder, and a discriminator. The deep capsule network extracts feature capsules from input images, which are then reconstructed by the decoder. The discriminator is the learnable reconstruction loss function that evaluates the similarity between reconstructed images and input images. Minimizing this learnable reconstruction loss and mean square error of images provides a regularization effect for the deep capsule network. Experimental results show that our models have a competitive performance of regularization on CIFAR10, CIFAR100, and FMNIST.
Synthetic aperture radar (SAR) image segmentation is the key to SAR image automatic interpretation. However, speckle noise, intensity inhomogeneity, and irregular shaped objects with changing edge often make the SAR image segmentation very difficult, and existing algorithms have high computational complexity. We propose a region-based level set method using the local image intensity information. To represent the statistical characteristics of speckle noise, we first use a gamma statistical distribution to model every segmented SAR image. We then apply a modified region mean estimation formula to efficiently segment SAR images with inhomogeneity. Finally, Gaussian filtering is employed to regularize the level set function, which can avoid reinitialization. The experimental results on synthetic and real-world SAR images demonstrate that the proposed method has less computation cost, faster convergence rate, and more accurate segmentation results.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.