As an active radar system, synthetic aperture radar (SAR) is often affected by different types of strong, complex, and variable electromagnetic interferences, which severely degrades the final imaging performance. Thus, how to effectively detect and suppress complex electromagnetic interferences is a crucial challenge currently. In this paper, we propose a self-supervised learning interference suppression method based on deep learning, including interference localization filtering and radar signal recovery. First, we construct a novel convolutional Autoencoder deep learning model —LocNet via the proposed optimization criterion, which is utilized to detect and locate the interference for subsequent filtration. Aiming at the issue of signal loss in the filtering process that is generally ignored in the current literature, we then reconstruct a novel U-Net neural network model—RecNet for the low-loss recovery of signal. Compared with the traditional parametric/non-parametric anti-interference methods, the most significant advantage of our method is that it overcomes the requirement for interference priori information, which is more consistent with the actual situation, and effectively solves the target information loss. Furthermore, since no interference information is involved in the training process (self-supervised training), our method applies to multiple types of interference rather than a specific one. Moreover, with our method, interference detection and suppression can be achieved simultaneously instead of separating the two steps as in existing literature. Measured and simulated SAR interference-contaminated data test results validate the effectiveness and robustness of the proposed method.

Accurately recognizing the type of complex electromagnetic jamming is the essential prerequisite for synthetic aperture radar (SAR) anti-jamming. However, current convolutional neural network (CNN)-based SAR jamming recognition methods require balanced training samples, which contradicts the varying difficulty of acquiring various jamming types, drastically reducing the recognition accuracy and generalization ability. This article proposes a discriminative feature distance metric model, JRSNet, for jamming recognition under imbalanced training samples, by refining the jamming modulation differences in the time-frequency (TF) domain into discriminative features. Novel feature discriminative distance metric (FD2M) loss function and discriminative feature constraint module (DFCM) are put forward to guarantee JRSNet learns embedding expression paradigm from jamming TF spectrograms to discriminative features, thus eliminating the influence of imbalanced training samples. Moreover, new spatial and channel attention modules are incorporated into JRSNet to capture jamming modulation information from multiple dimensions, consequently further improving recognition accuracy. Precisely because of the captured modulation regions in feature maps by spatial attention, the proposed approach can achieve jamming suppression synchronously. Experimental results show that under imbalanced training samples, JRSNet can accurately identify multiple jamming types both within and outside the training dataset with high generalizability. Compared with the existing jamming recognition methods, JRSNet performs superior recognition while taking into account good jamming suppression performance.

Matching infrared (IR) facial probes against a gallery of visible light faces remains a challenge, especially when combined with cross-distance due to deteriorated quality of the IR data. In this paper, we study the scenario where visible light faces are acquired at a short standoff, while IR faces are long-range data. To address the issue of quality imbalance between the heterogeneous imagery, we propose to compensate it by upgrading the lower-quality IR faces. Specifically, this is realized through cascaded face enhancement that combines an existing denoising algorithm (BM3D) with a new deep-learning-based deblurring model we propose (named SVDFace). Different IR bands, short-wave infrared (SWIR) and near-infrared (NIR), as well as different standoffs, are involved in the experiments. Results show that, in all cases, our proposed approach for quality balancing yields improved recognition performance, which is especially effective when involving SWIR images at a longer standoff. Our approach outperforms another easy and straightforward downgrading approach. The cascaded face enhancement structure is also shown to be beneficial and necessary. Finally, inspired by the singular value decomposition (SVD) theory, the proposed deblurring model of SVDFace is succinct, efficient and interpretable in structure. It is proven to be advantageous over traditional deblurring algorithms as well as state-of-the-art deep-learning-based deblurring algorithms.

Face recognition has been well studied under visible light and infrared (IR) in both intra-spectral and cross-spectral cases. However, how to fuse different light bands for face recognition, i.e., hyperspectral face recognition, is still an open research problem, which has the advantages of richer information retention and all-weather functionality over single-band face recognition. Thus, in this research, we revisit the hyperspectral recognition problem and provide a deep learning-based approach. A new fusion model (named HyperFace) is proposed to address this problem. The proposed model features a pre-fusion scheme, a Siamese encoder with bi-scope residual dense learning, a feedback-style decoder, and a recognition-oriented composite loss function. Experiments demonstrate that our method yields a much higher recognition rate than face recognition using only visible light or IR data. Moreover, our fusion model is shown to be superior to other general-purpose image fusion methods that are either traditional or deep learning-based, including state-of-the-art methods, in terms of both image quality and recognition performance.

Target detection and recognition of bistatic SAR image has been widely studied in recently. However, how to accurately detect and recognize targets with low-resolution and small size in the image is still a problem. Image super-resolution reconstruction technology can increase image resolution and expand target size to improve detection and recognition performance. Thus, we in this paper bring deep learning onto the topic of bistatic SAR image super-resolution reconstruction and propose a novel super-resolution reconstruction model (named FSRCNN) for bistatic SAR images. The proposed model is characterized by a feature extractor with different structures, a feedback feature enhancement block, and a feature fusion module. During the experiments of bistatic SAR measured data, our method is proved to perform clearer visual effects than other image super-resolution reconstruction method. Moreover, our model achieves the best value on peak signal to noise ratio and structural similarity, which indicates that our model can effectively super-resolution reconstruct bistatic SAR images.

Bistatic Synthetic Aperture Radar (SAR) provides enhanced angle-of-view information compared to monostatic SAR, offering improved target detection capabilities. Nevertheless, the variance of the imaging geometric configuration will change the imaging parameters in the matched filter, causing image defocusing. To address this issue, we proposed a frequency Domain Back projection (FBP) imaging algorithm for generally bistatic SAR. First, a local cartesian coordinate is established to reduce the spatial variance of imaging parameters. Then, a discriminant criterion is introduced to classify subregions of the imaging region. Next, a novel FBP method is proposed to solve the spatial variance of the imaging parameters. Compared to other frequency domain algorithms, the FBP imaging algorithm can directly obtain ground (SAR) image which have same resolution as back projection algorithm (BPA). The results of simulated data and flight experiment demonstrate the effectiveness of the proposed FBP algorithm.