13 October 2020 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Remote sensing image scene classification is one of the important means for remote sensing image understanding. Convolutional neural networks have been successfully applied to remote sensing image scene classifications and have demonstrated remarkable performance. However, with improvements in image resolution, remote sensing image categories are becoming increasingly diverse, and problems such as high intraclass diversity and high interclass similarity have arisen. The performance of ordinary convolutional neural networks at distinguishing increasingly complex remote sensing images is still limited. Therefore, we propose a feature fusion framework based on hierarchical attention and bilinear pooling called HABFNet for scene classification of remote sensing images. First, the deep convolutional neural network ResNet50 is used to extract the deep features from different layers of the image, and these features are fused to boost the robustness and effectiveness of the features. Second, we design an improved channel attention scheme to enhance the features from different layers. Finally, the enhanced features are cross-layer bilinearly pooled and fused, and the fused features are used for classification. Extensive experiments were conducted on three publicly available remote sensing image benchmarks. Comparisons with the state-of-the-art methods demonstrated that the proposed HABFNet achieved competitive classification performance.
30 October 2019 in IEEE Geoscience and Remote Sensing Letters
Point cloud classification of airborne light detection and ranging (LiDAR) data is essential to extract geoinformation. Although deep learning provides a new approach for classification, the time-consuming training process and data dependence prevent its widespread application to point clouds. To solve these problems and leverage the potential of high-performing neural networks, we propose an airborne LiDAR point cloud classification method based on transfer learning. A strategy to generate feature images considering the point cloud spatial distribution is first introduced for applying traditional convolutional neural networks to point clouds. Then, transfer learning is used to extract multiscale and multiview deep features. A simple neural network classifier is designed to reduce dimensionality, fuse and learn high-level features, and postprocessing considering contextual information further improves the classification accuracy. We verified the performance of the proposed method through experiments on two airborne LiDAR data sets with different characteristics and containing eight classes. The results demonstrate that the proposed method can achieve a satisfactory classification accuracy with relatively short training time and less training samples than if using conventional methods.
15 August 2023 in IEEE Transactions on Geoscience and Remote Sensing
The detection and recognition of oriented objects in remote sensing images is a challenging task due to their complex backgrounds, various sizes, diverse aspect ratios, and especially arbitrary orientations. Many oriented object detection algorithms need to obtain accurate angles or adopt anchors to predict the oriented bounding boxes. When directly predicting the angles of objects’ oriented bounding boxes, the loss of angle is discontinuous during training, which makes it difficult to obtain accurate boundary of oriented objects. And the anchors also aggravate the problems of class imbalance and computational cost. To address the above problems, this paper proposes an anchor-free and angle-free detector called AF 2 Det. AF 2 Det adopts the information of the bounding box projection instead of the angle to represent and reconstruct the object’s oriented bounding boxes, which could avoid the problem of boundary discontinuity. To predict the information of the bounding box projection, an anchor-free architecture is built to predict objects as points based on a simple but strong U-shaped architecture. And the deformable convolution and the bottom-up feature fusion method are integrated effectively to enhance AF 2 Det’ s capacity for objects’ shapes, orientations, and scales. The extensive experiments are conducted on multiple datasets, i.e., HRSC2016, FGSD2021, DOTA, and RSDD-SAR to validate the effectiveness of our method. The experimental results demonstrate that the proposed AF 2 Det outperforms other anchor-free algorithms and obtains competitive results on oriented object detection.