Although deep learning techniques have achieved noticeable success in aircraft detection, the scale heterogeneity, position difference, complex background interference, and speckle noise keep aircraft detection in large-scale synthetic aperture radar (SAR) images challenging. To solve these problems, we propose the geospatial transformer framework and implement it as a three-step target detection neural network, namely, the image decomposition, the multiscale geospatial contextual attention network (MGCAN), and result recomposition. First, the given large-scale SAR image is decomposed into slices via sliding windows according to the image characteristics of the aircraft. Second, slices are input into the MGCAN network for feature extraction, and the cluster distance nonmaximum suppression (CD-NMS) is utilized to determine the bounding boxes of aircraft. Finally, the detection results are produced via recomposition. Two innovative geospatial attention modules are proposed within MGCAN, namely, the efficient pyramid convolution attention fusion (EPCAF) module and the parallel residual spatial attention (PRSA) module, to extract multiscale features of the aircraft and suppress background noise. In the experiment, four large-scale SAR images with 1-m resolution from the Gaofen-3 system are tested, which are not included in the dataset. The results indicate that the detection performance of our geospatial transformer is better than Faster R-CNN, SSD, Efficientdet-D0, and YOLOV5s. The geospatial transformer integrates deep learning with SAR target characteristics to fully capture the multiscale contextual information and geospatial information of aircraft, effectively reduces complex background interference, and tackles the position difference of targets. It greatly improves the detection performance of aircraft and offers an effective approach to merge SAR domain knowledge with deep learning techniques.

The automatic detection of aircrafts from SAR images is widely applied in both military and civil fields, but there are still considerable challenges. To address the high variety of aircraft sizes and complex background information in SAR images, a new fast detection framework based on convolution neural networks is proposed, which achieves automatic and rapid detection of aircraft with high accuracy. First, the airport runway areas are detected to generate the airport runway mask and rectangular contour of the whole airport are generated. Then, a new deep neural network proposed in this paper, named Efficient Weighted Feature Fusion and Attention Network (EWFAN), is used to detect aircrafts. EWFAN integrates the weighted feature fusion module, the spatial attention mechanism, and the CIF loss function. EWFAN can effectively reduce the interference of negative samples and enhance feature extraction, thereby significantly improving the detection accuracy. Finally, the airport runway mask is applied to the detected results to reduce false alarms and produce the final aircraft detection results. To evaluate the performance of the proposed framework, large-scale Gaofen-3 SAR images with 1 m resolution are utilized in the experiment. The detection rate and false alarm rate of our EWFAN algorithm are 95.4% and 3.3%, respectively, which outperforms Efficientdet and YOLOv4. In addition, the average test time with the proposed framework is only 15.40 s, indicating satisfying efficiency of automatic aircraft detection.

Bridge detection from Synthetic Aperture Radar (SAR) images has very important strategic significance and practical value, but there are still many challenges in end-to-end bridge detection. In this paper, a new deep learning-based network is proposed to identify bridges from SAR images, namely, multi-resolution attention and balance network (MABN). It mainly includes three parts, the attention and balanced feature pyramid (ABFP) network, the region proposal network (RPN), and the classification and regression. First, the ABFP network extracts various features from SAR images, which integrates the ResNeXt backbone network, balanced feature pyramid, and the attention mechanism. Second, extracted features are used by RPN to generate candidate boxes of different resolutions and fused. Furthermore, the candidate boxes are combined with the features extracted by the ABFP network through the region of interest (ROI) pooling strategy. Finally, the detection results of the bridges are produced by the classification and regression module. In addition, intersection over union (IOU) balanced sampling and balanced L1 loss functions are introduced for optimal training of the classification and regression network. In the experiment, TerraSAR data with 3-m resolution and Gaofen-3 data with 1-m resolution are used, and the results are compared with faster R-CNN and SSD. The proposed network has achieved the highest detection precision (P) and average precision (AP) among the three networks, as 0.877 and 0.896, respectively, with the recall rate (RR) as 0.917. Compared with the other two networks, the false alarm targets and missed targets of the proposed network in this paper are greatly reduced, so the precision is greatly improved.

The water and shadow areas in SAR images contain rich information for various applications, which cannot be extracted automatically and precisely at present. To handle this problem, a new framework called Multi-Resolution Dense Encoder and Decoder (MRDED) network is proposed, which integrates Convolutional Neural Network (CNN), Residual Network (ResNet), Dense Convolutional Network (DenseNet), Global Convolutional Network (GCN), and Convolutional Long Short-Term Memory (ConvLSTM). MRDED contains three parts: the Gray Level Gradient Co-occurrence Matrix (GLGCM), the Encoder network, and the Decoder network. GLGCM is used to extract low-level features, which are further processed by the Encoder. The Encoder network employs ResNet to extract features at different resolutions. There are two components of the Decoder network, namely, the Multi-level Features Extraction and Fusion (MFEF) and Score maps Fusion (SF). We implement two versions of MFEF, named MFEF1 and MFEF2, which generate separate score maps. The difference between them lies in that the Chained Residual Pooling (CRP) module is utilized in MFEF2, while ConvLSTM is adopted in MFEF1 to form the Improved Chained Residual Pooling (ICRP) module as the replacement. The two separate score maps generated by MFEF1 and MFEF2 are fused with different weights to produce the fused score map, which is further handled by the Softmax function to generate the final extraction results for water and shadow areas. To evaluate the proposed framework, MRDED is trained and tested with large SAR images. To further assess the classification performance, a total of eight different classification frameworks are compared with our proposed framework. MRDED outperformed by reaching 80.12% in Pixel Accuracy (PA) and 73.88% in Intersection of Union (IoU) for water, 88% in PA and 77.11% in IoU for shadow, and 95.16% in PA and 90.49% in IoU for background classification, respectively.

In aircraft detection from synthetic aperture radar (SAR) images, there are several major challenges: the shattered features of the aircraft, the size heterogeneity and the interference of a complex background. To address these problems, an Efficient Bidirectional Path Aggregation Attention Network (EBPA2N) is proposed. In EBPA2N, YOLOv5s is used as the base network and then the Involution Enhanced Path Aggregation (IEPA) module and Effective Residual Shuffle Attention (ERSA) module are proposed and systematically integrated to improve the detection accuracy of the aircraft. The IEPA module aims to effectively extract advanced semantic and spatial information to better capture multi-scale scattering features of aircraft. Then, the lightweight ERSA module further enhances the extracted features to overcome the interference of complex background and speckle noise, so as to reduce false alarms. To verify the effectiveness of the proposed network, Gaofen-3 airports SAR data with 1 m resolution are utilized in the experiment. The detection rate and false alarm rate of our EBPA2N algorithm are 93.05% and 4.49%, respectively, which is superior to the latest networks of EfficientDet-D0 and YOLOv5s, and it also has an advantage of detection speed.