ABSTRACT The Chinese Space Station Telescope (CSST) is scheduled to launch soon, which is expected to provide a vast amount of image potentially containing low-surface brightness galaxies (LSBGs). However, detecting and characterizing LSBGs is known to be challenging due to their faint surface brightness, posing a significant hurdle for traditional detection methods. In this paper, we propose LSBGnet, a deep neural network specifically designed for automatic detection of LSBGs. We established LSBGnet-SDSS model using data set from the Sloan Digital Sky Survey (SDSS). The results demonstrate a significant improvement compared to our previous work, achieving a recall of 97.22 per cent and a precision of 97.27 per cent on the SDSS test set. Furthermore, we use the LSBGnet-SDSS model as a pre-training model, employing transfer learning to retrain the model with LSBGs from Dark Energy Survey (DES), and establish the LSBGnet-DES model. Remarkably, after retraining the model on a small DES sample, it achieves over 90 per cent precision and recall. To validate the model’s capabilities, we utilize the trained LSBGnet-DES model to detect LSBG candidates within a selected 5 sq. deg area in the DES footprint. Our analysis reveals the detection of 204 LSBG candidates, characterized by a mean surface brightness range of $23.5\ \mathrm{ mag}\ \mathrm{ arcsec}^{-2}\le \bar{\mu }_{\text{eff}}(g)\le 26.8\ \mathrm{ mag}\ \mathrm{ arcsec}^{-2}$ and a half-light radius range of 1.4 arcsec ≤ r1/2 ≤ 8.3 arcsec. Notably, 116 LSBG candidates exhibit a half-light radius ≥2.5 arcsec. These results affirm the remarkable performance of our model in detecting LSBGs, making it a promising tool for the upcoming CSST.

Low-surface-brightness (LSB) galaxies play a crucial role in our understanding of galaxy evolution and dark matter cosmology. However, efficiently detecting them in large-scale surveys is challenging, due to their dim appearance. In this study, we propose a two-step detection method based on deep learning to address this issue. First, an object detection model called GalCenterNet was designed to detect LSB galaxy candidates in astronomical images. The model was trained using a data set of 665 Sloan Digital Sky Survey (SDSS) images, which contained 667 LSB galaxies. On the test set, the model achieved an accuracy of 95.05% and a recall of 96.00%. Next, an anomaly detection technique known as Deep Support Vector Data Description was applied to identify abnormal sources, thus refining the LSB candidates. By applying the two-step detection method to SDSS images, we have obtained a sample of 37,536 LSB galaxy candidates. This wide-area sample contains diverse and abundant LSB galaxies, which are valuable for studying the properties of LSB galaxies and the role that the environment plays in their evolution. The proposed detection method enables end-to-end detection from the SDSS images to the final detection results. This approach will be further employed to efficiently identify objects in the upcoming Chinese Survey Space Telescope sky survey.

Cardiac magnetic resonance imaging analysis has been a useful tool in screening patients for heart disease. Early, timely and accurate diagnosis of diseases of the heart series is the key to effective treatment. MRI provides important material for the diagnosis of cardiac diseases. The rise of deep learning has transformed computer-aided diagnostic systems, especially in the field of medical imaging. Existing work on cardiac structure segmentation models based on MRI imaging mainly relies on convolutional neural networks (CNNs), which lack model diversity and limit the prediction performance. This paper introduces Visual Transformer with Feature Recombination and Feature Distillation(ViT-FRD), a novel learning pipeline that combines a visual transformer (ViT) and a CNN through knowledge refinement. The training procedure allows the student model, i.e., ViT, to learn from the teacher model, i.e., CNN, by optimizing distillation losses. Meanwhile, ViT-FRD provides two performance boosters to increase the efficacy and efficiency of training. The proposed method is validated on two cardiac MRI image datasets. The findings demonstrate that ViT-FRD achieves SOTA and outperforms the widely used baseline model.

Low-surface-brightness galaxies (LSBGs), fainter members of the galaxy population, are thought to be numerous. However, due to their low surface brightness, the search for a wide-area sample of LSBGs is difficult, which in turn limits our ability to fully understand the formation and evolution of galaxies as well as galaxy relationships. Edge-on LSBGs, due to their unique orientation, offer an excellent opportunity to study galaxy structure and galaxy components. In this work, we utilize the You Only Look Once object detection algorithm to construct an edge-on LSBG detection model by training on 281 edge-on LSBGs in Sloan Digital Sky Survey (SDSS) gri -band composite images. This model achieved a recall of 94.64% and a purity of 95.38% on the test set. We searched across 938,046 gri -band images from SDSS Data Release 16 and found 52,293 candidate LSBGs. To enhance the purity of the candidate LSBGs and reduce contamination, we employed the Deep Support Vector Data Description algorithm to identify anomalies within the candidate samples. Ultimately, we compiled a catalog containing 40,759 edge-on LSBG candidates. This sample has similar characteristics to the training data set, mainly composed of blue edge-on LSBG candidates. The catalog is available online at https://github.com/worldoutside/Edge-on_LSBG .

We present a data-driven approach to automatically detect L dwarfs from Sloan Digital Sky Survey (SDSS) images using an improved Faster R-CNN framework based on deep learning. The established L-dwarf automatic detection (LDAD) model distinguishes L dwarfs from other celestial objects and backgrounds in SDSS field images by learning the features of 387 SDSS images containing L dwarfs. Applying the LDAD model to the SDSS images containing 93 labeled L dwarfs in the test set, we successfully detected 83 known L dwarfs with a recall rate of 89.25% for known L dwarfs. Several techniques are implemented in the LDAD model to improve its detection performance for L dwarfs, including the deep residual network and the feature pyramid network. As a result, the LDAD model outperforms the model of the original Faster R-CNN, whose recall rate of known L dwarfs is 80.65% for the same test set. The LDAD model was applied to detect L dwarfs from a larger validation set including 843 labeled L dwarfs, resulting in a recall rate of 94.42% for known L dwarfs. The newly identified candidates include L dwarfs, late M and T dwarfs, which were estimated from color ( i − z ) and spectral type relation. The contamination rates for the test candidates and validation candidates are 8.60% and 9.27%, respectively. The detection results indicate that our model is effective to search for L dwarfs from astronomical images.

Ship classification in synthetic aperture radar (SAR) images is a fundamental and significant step in ocean surveillance. Recently, with the rise of deep learning (DL), modern abstract features from convolutional neural networks (CNNs) have hugely improved SAR ship classification accuracy. However, most existing CNN-based SAR ship classifiers overly rely on abstract features, but uncritically abandon traditional mature hand-crafted features, which may incur some challenges for further improving accuracy. Hence, this article proposes a novel DL network with histogram of oriented gradient (HOG) feature fusion (HOG-ShipCLSNet) for preferable SAR ship classification. In HOG-ShipCLSNet, four mechanisms are proposed to ensure superior classification accuracy, that is, 1) a multiscale classification mechanism (MS-CLS-Mechanism); 2) a global self-attention mechanism (GS-ATT-Mechanism); 3) a fully connected balance mechanism (FC-BAL-Mechanism); and 4) an HOG feature fusion mechanism (HOG-FF-Mechanism). We perform sufficient ablation studies to confirm the effectiveness of these four mechanisms. Finally, our experimental results on two open SAR ship datasets (OpenSARShip and FUSAR-Ship) jointly reveal that HOG-ShipCLSNet dramatically outperforms both modern CNN-based methods and traditional hand-crafted feature methods.

Huge imbalance of different scenes' sample numbers seriously reduces synthetic aperture radar (SAR) ship detection accuracy. Thus, to solve this problem, this letter proposes a balance scene learning mechanism (BSLM) for offshore and inshore ship detection in SAR images. BSLM involves three steps: 1) based on unsupervised representation learning, a generative adversarial network (GAN) is used to extract the scene features of SAR images; 2) using these features, a scene binary cluster (offshore/inshore) is conducted by K-means; and 3) finally, the small cluster's samples (inshore) are augmented via replication, rotation transformation or noise addition to balance another big cluster (offshore), so as to eliminate scene learning bias and obtain balanced learning representation ability that can enhance learning benefits and improve detection accuracy. This letter applies BSLM to four widely used and open-sourced deep learning detectors, i.e., faster regions-convolutional neural network (Faster R-CNN), Cascade R-CNN, single shot multibox detector (SSD), and RetinaNet, to verify its effectiveness. Experimental results on the open SAR ship detection data set (SSDD) reveal that BSLM can greatly improve detection accuracy, especially for more complex inshore scenes.

SAR Ship Detection Dataset (SSDD) is the first open dataset that is widely used to research state-of-the-art technology of ship detection from Synthetic Aperture Radar (SAR) imagery based on deep learning (DL). According to our investigation, up to 46.59% of the total 161 public reports confidently select SSDD to study DL-based SAR ship detection. Undoubtedly, this situation reveals the popularity and great influence of SSDD in the SAR remote sensing community. Nevertheless, the coarse annotations and ambiguous standards of use of its initial version both hinder fair methodological comparisons and effective academic exchanges. Additionally, its single-function horizontal-vertical rectangle bounding box (BBox) labels can no longer satisfy the current research needs of the rotatable bounding box (RBox) task and the pixel-level polygon segmentation task. Therefore, to address the above two dilemmas, in this review, advocated by the publisher of SSDD, we will make an official release of SSDD based on its initial version. SSDD’s official release version will cover three types: (1) a bounding box SSDD (BBox-SSDD), (2) a rotatable bounding box SSDD (RBox-SSDD), and (3) a polygon segmentation SSDD (PSeg-SSDD). We relabel ships in SSDD more carefully and finely, and then explicitly formulate some strict using standards, e.g., (1) the training-test division determination, (2) the inshore-offshore protocol, (3) the ship-size reasonable definition, (4) the determination of the densely distributed small ship samples, and (5) the determination of the densely parallel berthing at ports ship samples. These using standards are all formulated objectively based on the using differences of existing 75 (161 × 46.59%) public reports. They will be beneficial for fair method comparison and effective academic exchanges in the future. Most notably, we conduct a comprehensive data analysis on BBox-SSDD, RBox-SSDD, and PSeg-SSDD. Our analysis results can provide some valuable suggestions for possible future scholars to further elaborately design DL-based SAR ship detectors with higher accuracy and stronger robustness when using SSDD.

Synthetic aperture radar (SAR) is playing an important role in maritime domain awareness. As a fundamental ocean mission, SAR ship detection can offer high-quality services for marine surveillance. It receives widespread concern in the SAR remote sensing community. Recently, deep learning (DL) has greatly improved SAR ship detection accuracy. However, there are still some unperceived imbalance problems that seriously hinder further accuracy improvements among current DL-based SAR ship detectors. Therefore, we propose a novel concept of balance learning (BL) for high-quality SAR ship detection. We first point out the four unperceived imbalance problems, i.e., image sample scene imbalance, positive negative sample imbalance, ship scale feature imbalance, and classification regression task imbalance. Then, we offer some profound insights into these imbalances. Immediately, we propose four effective solutions to handle the above four imbalances correspondingly, i.e., balance scene learning mechanism (BSLM), balance interval sampling mechanism (BISM), balance feature pyramid network (BFPN), and balance classification regression network (BCRN). Finally, combined with the four solutions, a novel balance learning network (BL-Net) is proposed. Ablation studies can confirm each solution's effectiveness. Experimental results on five open datasets (SSDD, Gaofen-SSDD, Sentinel-SSDD, SAR-Ship-Dataset, and HRSID) reveal BL-Net's state-of-the-art SAR ship detection performance compared to the other 30 DL-based SAR ship detectors. Specifically, in contrast to the current most competitive method, the accuracy increase of BLNet is 2.98% on SSDD, 1.97% on Gaofen-SSDD, 1.49% on Sentinel-SSDD, 0.55% on SAR-Ship-Dataset, and 4.95% on HRSID. Last but not least, the satisfactory ship detection results on another two large-scene Sentinal-1 SAR images confirm BL-Net's strong migration capability. This indicates BL-Net's potential value in marine surveillance.

Ship detection in synthetic aperture radar (SAR) images is becoming a research hotspot. In recent years, as the rise of artificial intelligence, deep learning has almost dominated SAR ship detection community for its higher accuracy, faster speed, less human intervention, etc. However, today, there is still a lack of a reliable deep learning SAR ship detection dataset that can meet the practical migration application of ship detection in large-scene space-borne SAR images. Thus, to solve this problem, this paper releases a Large-Scale SAR Ship Detection Dataset-v1.0 (LS-SSDD-v1.0) from Sentinel-1, for small ship detection under large-scale backgrounds. LS-SSDD-v1.0 contains 15 large-scale SAR images whose ground truths are correctly labeled by SAR experts by drawing support from the Automatic Identification System (AIS) and Google Earth. To facilitate network training, the large-scale images are directly cut into 9000 sub-images without bells and whistles, providing convenience for subsequent detection result presentation in large-scale SAR images. Notably, LS-SSDD-v1.0 has five advantages: (1) large-scale backgrounds, (2) small ship detection, (3) abundant pure backgrounds, (4) fully automatic detection flow, and (5) numerous and standardized research baselines. Last but not least, combined with the advantage of abundant pure backgrounds, we also propose a Pure Background Hybrid Training mechanism (PBHT-mechanism) to suppress false alarms of land in large-scale SAR images. Experimental results of ablation study can verify the effectiveness of the PBHT-mechanism. LS-SSDD-v1.0 can inspire related scholars to make extensive research into SAR ship detection methods with engineering application value, which is conducive to the progress of SAR intelligent interpretation technology.

Instance segmentation in high-resolution (HR) remote sensing imagery is one of the most challenging tasks and is more difficult than object detection and semantic segmentation tasks. It aims to predict class labels and pixel-wise instance masks to locate instances in an image. However, there are rare methods currently suitable for instance segmentation in the HR remote sensing images. Meanwhile, it is more difficult to implement instance segmentation due to the complex background of remote sensing images. In this article, a novel instance segmentation approach of HR remote sensing imagery based on Cascade Mask R-CNN is proposed, which is called a high-quality instance segmentation network (HQ-ISNet). In this scheme, the HQ-ISNet exploits a HR feature pyramid network (HRFPN) to fully utilize multi-level feature maps and maintain HR feature maps for remote sensing images’ instance segmentation. Next, to refine mask information flow between mask branches, the instance segmentation network version 2 (ISNetV2) is proposed to promote further improvements in mask prediction accuracy. Then, we construct a new, more challenging dataset based on the synthetic aperture radar (SAR) ship detection dataset (SSDD) and the Northwestern Polytechnical University very-high-resolution 10-class geospatial object detection dataset (NWPU VHR-10) for remote sensing images instance segmentation which can be used as a benchmark for evaluating instance segmentation algorithms in the high-resolution remote sensing images. Finally, extensive experimental analyses and comparisons on the SSDD and the NWPU VHR-10 dataset show that (1) the HRFPN makes the predicted instance masks more accurate, which can effectively enhance the instance segmentation performance of the high-resolution remote sensing imagery; (2) the ISNetV2 is effective and promotes further improvements in mask prediction accuracy; (3) our proposed framework HQ-ISNet is effective and more accurate for instance segmentation in the remote sensing imagery than the existing algorithms.

In this paper, a novel recognition method based on the squeeze-and-excitation networks (SE-Nets) is proposed in order to recognize intra-pulse modulation signals at varying noise levels automatically. Firstly, different signal transforms including time domain, frequency domain and time–frequency domain are used to convert seven different intra-pulse modulation signals into images. Then, since the SE-Net has great advantages in image processing, the images are classified by the squeeze-and-excitation networks and output the results in each domain. Lastly, the decisions of different domains are combined to obtain the final results. The simulation results demonstrate that the recognition accuracies are all more than 95% except BPSK signals which are still more than 90% at the case of − 8 dB. Compared with several other neural networks and traditional support vector machine method, the accuracy of SE-Net method has increased more than 2% and performs better under different SNR conditions. The measured signals results show that the accuracies of the SE-Net method are higher than those of several other neural networks, especially for BASK and BFSK signals.

Ship detection in high-resolution synthetic aperture radar (SAR) imagery is a challenging problem in the case of complex environments, especially inshore and offshore scenes. Nowadays, the existing methods of SAR ship detection mainly use low-resolution representations obtained by classification networks or recover high-resolution representations from low-resolution representations in SAR images. As the representation learning is characterized by low resolution and the huge loss of resolution makes it difficult to obtain accurate prediction results in spatial accuracy; therefore, these networks are not suitable to ship detection of region-level. In this paper, a novel ship detection method based on a high-resolution ship detection network (HR-SDNet) for high-resolution SAR imagery is proposed. The HR-SDNet adopts a novel high-resolution feature pyramid network (HRFPN) to take full advantage of the feature maps of high-resolution and low-resolution convolutions for SAR image ship detection. In this scheme, the HRFPN connects high-to-low resolution subnetworks in parallel and can maintain high resolution. Next, the Soft Non-Maximum Suppression (Soft-NMS) is used to improve the performance of the NMS, thereby improving the detection performance of the dense ships. Then, we introduce the Microsoft Common Objects in Context (COCO) evaluation metrics, which provides not only the higher quality evaluation metrics average precision (AP) for more accurate bounding box regression, but also the evaluation metrics for small, medium and large targets, so as to precisely evaluate the detection performance of our method. Finally, the experimental results on the SAR ship detection dataset (SSDD) and TerraSAR-X high-resolution images reveal that (1) our approach based on the HRFPN has superior detection performance for both inshore and offshore scenes of the high-resolution SAR imagery, which achieves nearly 4.3% performance gains compared to feature pyramid network (FPN) in inshore scenes, thus proving its effectiveness; (2) compared with the existing algorithms, our approach is more accurate and robust for ship detection of high-resolution SAR imagery, especially inshore and offshore scenes; (3) with the Soft-NMS algorithm, our network performs better, which achieves nearly 1% performance gains in terms of AP; (4) the COCO evaluation metrics are effective for SAR image ship detection; (5) the displayed thresholds within a certain range have a significant impact on the robustness of ship detectors.

Compressive Sensing(CS) has been widely utilized in Inverse synthetic aperture radar(ISAR) imaging since real ISAR data is easier to be non-completed, and CS-based methods can obtain high-quality imaging results using under-sampled data. However, traditional CS-based methods need pre-defined parameters, sparse transforms and iterative reconstruction processes. Optimal parameters as well as transforms are tough to be hand-crafted, and iterative reconstruction consumes plenty of time, which limit practical applications in ISAR imaging. Given that Convolution Neural Network(CNN) has great power to learn rapidly, we compose CNN with traditional Iterative Shrinkage-Thresholding Algorithm(ISTA) to propose CNN-ISTA(CIST)-based ISAR imaging method. CIST is capable of learning optimal parameters and transforms throughout the training (i.e. the optimization process is interpretable) instead of manually defined. Compared with traditional state-of-the-art CS imaging methods, the experimental results demonstrate that our proposed CIST-based imaging method is superior in both imaging quality and computational efficiency.

SAR Ship Detection Dataset (SSDD) has been widely used by researchers because it can achieve a good tradeoff between model performance and time consumption of experiments. However, there is still a lack of performance benchmark for state-of-the-art methods on SSDD. On the one hand, this makes it difficult for researchers to grasp the performance of mainstream methods in general; on the other hand, research results of SAR ship detection based on deep learning cannot be compared horizontally in a unified experiment setup. To solve the lack of this solid work in the field of ship detection in SAR images based on deep learning, this paper benchmarks state-of-the-art detection methods on SSDD to provide an overview for future research work.

Due to the rapid development of chip technology and deep learning revolution, many ship detection frameworks for synthetic aperture radar (SAR) imagery based on convolutional neural networks (CNNs) have been proposed and achieved great success. However, there are problems hampering their development: 1) For the SAR ship detection task, it is uneconomic to apply heavy backbone network to extract features because it results in heavy computing load and prolongs the inference time cost; 2) The anchor-based methods usually have massive hyper-parameters, which typically need to be tuned carefully and easily lead to weak detection performance. To alleviate the problems, an efficient low-cost ship detection network for SAR imagery is proposed in this paper. Firstly, a simplified U-Net as the backbone to extract features is proposed. It only contains ~0.47 million learnable weights, which is 2.37%, 0.76%, 0.34%, 1.01%, 0.55% and 1.07% of DarkNet-19, DarkNet-53, VGG-16, ResNet-50, ResNet-101 and ResNext-101, respectively. Secondly, an anchor-free SAR ship detection framework consisting of a bounding boxes regression sub-net and a score map regression sub-net based on simplified U-Net is proposed. To evaluate the effectiveness of our method, extensive experiments have been conducted and a more comprehensive set of evaluation metrics have been applied. Results demonstrate that the proposed network achieves 68.1% average precision and 67.6% average recall on the SAR ship detection dataset (SSDD), respectively. Compared with the state-of-the-art works, our proposed network achieves very competitive detection performance and extreme lightweight (~0.93 million learnable weights in total).

With the development of satellite technology, up to date imaging mode of synthetic aperture radar (SAR) satellite can provide higher resolution SAR imageries, which benefits ship detection and instance segmentation. Meanwhile, object detectors based on convolutional neural network (CNN) show high performance on SAR ship detection even without land-ocean segmentation; but with respective shortcomings, such as the relatively small size of SAR images for ship detection, limited SAR training samples, and inappropriate annotations, in existing SAR ship datasets, related research is hampered. To promote the development of CNN based ship detection and instance segmentation, we have constructed a High-Resolution SAR Images Dataset (HRSID). In addition to object detection, instance segmentation can also be implemented on HRSID. As for dataset construction, under the overlapped ratio of 25%, 136 panoramic SAR imageries with ranging resolution from 1m to 5m are cropped to 800 × 800 pixels SAR images. To reduce wrong annotation and missing annotation, optical remote sensing imageries are applied to reduce the interferes from harbor constructions. There are 5604 cropped SAR images and 16951 ships in HRSID, and we have divided HRSID into a training set (65% SAR images) and test set (35% SAR images) with the format of Microsoft Common Objects in Context (MS COCO). 8 state-of-the-art detectors are experimented on HRSID to build the baseline; MS COCO evaluation metrics are applicated for comprehensive evaluation. Experimental results reveal that ship detection and instance segmentation can be well implemented on HRSID.

Automatic modulation classification of radar signals, which plays a significant role in both civilian and military applications, is researched in this study through a deep learning network. In this study, a novel network combined a shallow convolution neural network (CNN), long short-term memory (LSTM) network and deep neural network (DNN) is proposed to recognise six types of radar signals with different signal-to-noise ratio (SNR) levels from -14 to 20 dB. First, raw signal sequences in the time domain, frequency domain and autocorrelation domain are as input for a shallow CNN. Then the features extracted by CNN will be the input of LSTM network. Finally, DNNs will output the signal modulation types directly. The simulation results demonstrate that the accuracies in autocorrelation domain are all more than 90% at -6 dB and close to 100% when SNR > -2 dB. The recognition performances of the three domains are compared. Compared with other recognition methods, the proposed method has higher average accuracy and better performance under low SNR condition. The measured results show that the proposed method has achieved high accuracies of common four kinds of measured radar signals.

Ship detection in high-resolution synthetic aperture radar (SAR) imagery is a fundamental and challenging problem due to the complex environments. In this paper, a RetinaNet-Plus method is presented for ship detection in high-resolution SAR imagery based on RetinaNet network modified. In this approach, instead of setting the score for neighboring region proposals to zero as in Non-Maximum Suppression (NMS), Soft-NMS decreases the detection scores as an increasing function of overlap to avoid loss of precision. In addition, focal loss is used to address the class imbalance and to increase the importance of the hard examples during training. The experiments on SAR ship SSDD dataset and TerraSAR-X image from Barcelona port, show that our method is more accurate than the existing algorithms and is effective for ship detection of high-resolution SAR imagery.

Autofocus method is a vital technology for high resolution and wide swath airborne Synthetic Aperture Radar (SAR) imaging. The autofocus algorithms via phase errors estimation and traditional Antenna Phase Centers (APC) errors estimation cannot completely compensate for the phase error of each pixel for the large scene ignoring the spatial variance, which results in corrupted SAR imagery for some part of the scene. In this paper, an autofocus algorithm through precise APC errors estimation based on joint multiregion is proposed to compensate motion error for the whole scene greatly. We established an image intensity model for strong point targets in multi-region with the weight coefficient to estimate APC errors. Moreover, the partial derivative of image intensity is simplified which can easily derive higher order criterion like image sharpness and Conjugate Gradient (CG) is utilized to solve the optimization problem. The simulation and experimental examples verify the effectiveness of the proposed method compared with traditional methods.