Object detection in SAR images has always been a topic of great interest in the field of deep learning. Early works commonly focus on improving performance on convolutional neural network frameworks. More recent works continue this path and introduce the attention mechanisms of Transformers for better semantic interpretation. However, these methods fail to treat the Transformer itself as a detection framework and, therefore, lack the development of various details that contribute to the state-of-the-art performance of Transformers. In this work, we first base our work on a fully multi-scale Transformer-based detection framework, DETR (DEtection TRansformer) to utilize its superior detection performance. Secondly, to acquire rotation-related attributes for better representation of SAR objects, an Orientation Enhancement Module (OEM) is proposed to facilitate the enhancement of rotation characteristics. Then, to enable learning of more effective and discriminative representations of foreground objects and background noises, a contrastive-loss-based GRC Loss is proposed to preserve patterns of both categories. Moreover, to not restrict comparisons exclusively to maritime objects, we have also developed an open-source labeled vehicle dataset. Finally, we evaluate both detection performance and generalization ability on two well-known ship datasets and our vehicle dataset. We demonstrated our method’s superior performance and generalization ability on both datasets.

Monitoring and identification of ships in remote sensing images is of great significance for port management, marine traffic, marine security, etc. However, due to small size and complex background, ship detection in remote sensing images is still a challenging task. Currently, deep-learning-based detection models need a lot of data and manual annotation, while training data containing ships in remote sensing images may be in limited quantities. To solve this problem, in this paper, we propose a few-shot multi-class ship detection algorithm with attention feature map and multi-relation detector (AFMR) for remote sensing images. We use the basic framework of You Only Look Once (YOLO), and use the attention feature map module to enhance the features of the target. In addition, the multi-relation head module is also used to optimize the detection head of YOLO. Extensive experiments on publicly available HRSC2016 dataset and self-constructed REMEX-FSSD dataset validate that our method achieves a good detection performance.

Recent years have witnessed rapid development and remarkable achievements on deep learning object detection in remote sensing (RS) images. The growing improvement of the accuracy is inseparable from the increasingly complex deep convolutional neural network and the huge amount of sample data. However, the under-fitting neural network will damage the detection performance facing the difficulty of sample acquisition. Thus, it evolves into few-shot object detection (FSOD). In this article, we first briefly introduce the object detection task and its algorithms, to better understand the basic detection frameworks followed by FSOD. Then, FSOD design methods in RS images for three important aspects, such as sample, model, and learning strategy, are respectively discussed. In addition, some valuable research results of FSOD in computer vision field are also included. We advocate a wide research technique route, and some advice about feature enhancement and multi-modal fusion, semantics extraction and cross-domain mapping, fine-tune and meta-learning strategies, and so on, are provided. Based on our stated research route, a novel few-shot detector that focuses on contextual information is proposed. At the end of the paper, we summarize accuracy performance on experimental datasets to illustrate the achievements and shortcomings of the stated algorithms, and highlight the future opportunities and challenges of FSOD in RS image interpretation, in the hope of providing insights into future research.

Relative pose estimation of a satellite is an essential task for aerospace missions, such as on-orbit servicing and close proximity formation flying. However, the changeable situation makes precise relative pose estimation difficult. This paper introduces a deep-learning-based satellite relative pose estimation method for monocular optical images. The method is geared towards uncooperative target satellites with known 3D models. This paper proposes a novel convolutional neural network combined with 3D prior knowledge expressed by the 3D model in the form of the point cloud. The method utilizes point cloud convolution to extract features from the point cloud. To make the result more precise, a loss function that is more suitable for satellite pose estimation tasks is designed. For training and testing the proposed method, large amounts of data are required. This paper constructs a satellite pose estimation dataset BUAA-SID-POSE 1.0 by simulation. The proposed method is applied to the dataset and shows desirable performance on the pose estimation task. The proposed technique can be used to accomplish monocular vision-based relative pose estimation tasks in space-borne applications.