We study the problem of object detection in remote sensing images. As a simple but effective feature extractor, Feature Pyramid Network (FPN) has been widely used in several generic vision tasks. However, it still faces some challenges when used for remote sensing object detection, as the objects in remote sensing images usually exhibit variable shapes, orientations, and sizes. To this end, we propose a dedicated object detector based on the FPN architecture to achieve accurate object detection in remote sensing images. Specifically, considering the variable shapes and orientations of remote sensing objects, we first replace the original lateral connections of FPN with Deformable Convolution Lateral Connection Modules (DCLCMs), each of which includes a $3\times 3$ deformable convolution to generate feature maps with deformable receptive fields. Additionally, we further introduce several Attention-based Multi-Level Feature Fusion Modules (A-MLFFMs) to integrate the multi-level outputs of FPN adaptively, further enabling multi-scale object detection. Extensive experimental results on the DIOR dataset demonstrated the state-of-the-art performance achieved by the proposed method, with the highest mean Average Precision (mAP) of 73.6%.

Recently, convolutional neural networks (CNNs) have attracted enormous attention in pattern recognition and demonstrated excellent performance in hyperspectral image (HSI) classification. However, high-dimensional HSI dataset versus limited training samples is easy to cause the over-fitting phenomenon in deep neural networks. Additionally, the intraclass distance of the embedding features extracted through the softmax-based CNNs may be greater than that of the interclass, which makes it difficult to further improve the classification accuracy. To address these issues, this paper proposes a deep prototypical network with hybrid residual attention (DPN-HRA), which can effectively investigate the spectral-spatial information in HSI. Specifically, in order to improve the generalization capability of the model, feature extraction with a hybrid residual attention module is presented to enhance the critical spectral-spatial features and suppress the useless ones in the classification task. Furthermore, a novel discriminant distance-based cross-entropy loss (D²CEL) is proposed to increase the intraclass compactness, so as to obtain more superior results. Extensive experiments on three benchmark datasets are carried out to convincingly evaluate the proposed framework. With the generation of optimal prototypes representing each class and more discriminative embedding features, encouraging classification results are achieved compared with state-of-the-art methods.

The typical probability based point pattern matching method is coherent point drift (CPD) algorithm, which treats one point set as centroids of a Gaussian mixture model, and then fits it to the other. It uses the expectation maximization framework, where the point correspondences and transformation parameters are updated alternately. However, the anti-outlier performance of CPD is not robust enough as outliers have always been involved in the operation until the CPD converges. Hence, an automatic outlier suppression (AOS) mechanism is proposed. First, outliers are judged by a matching probability matrix. Then, transformation parameters are fitted using accurate matching point sets. Finally, the Gaussian centroids are forced to move coherently by this transformation model. AOS-CPD can efficiently improve the anti-outlier performance of rigid CPD. Furthermore, CPD is applied to image matching. A new local changing information descriptor-relative phase histogram (RPH) is designed and RPH-AOS-CPD is proposed to embed RPH measurement into AOS-CPD as a constraint condition. RPH-AOS-CPD makes full use of grayscale information besides having an excellent anti-outlier performance. The experimental results based on both synthetic and real data indicate that compared with other algorithms, AOS-CPD is more robust to outliers and RPH-AOS-CPD offers a good practicability and accuracy in image matching applications.

For infrared imaging systems with high sampling width applying to the traditional display device or real-time processing system with 8-bit data width, this paper presents a new high dynamic range compression and detail enhancement (DRCDDE) algorithm for infrared images. First, a bilateral filter is adopted to separate the original image into two parts: the base component that contains large-scale signal variations, and the detail component that contains high-frequency information. Then, the operator model for DRC with local-contrast preservation is established, along with a new proposed nonlinear intensity transfer function (ITF) to implement adaptive DRC of the base component. For the detail component, depending on the local statistical characteristics, we set up suitable intensity level extension criteria to enhance the low-contrast details and suppress noise. Finally, the results of the two components are recombined with a weighted coefficient. Experiment results by real infrared data, and quantitative comparison with other well-established methods, show the better performance of the proposed algorithm. Furthermore, the technique could effectively project a dim target while suppressing noise, which is beneficial to image display and target detection.

基于相对相位直方图的数字表面模型数据与遥感图像配准(Registration of DSM data and remote sensing image based on relative phase histogram)