The object detection of unmanned aerial vehicle (UAV) images has widespread applications in numerous fields; however, the complex background, diverse scales, and uneven distribution of objects in UAV images make object detection a challenging task. This study proposes a convolution neural network (CNN)-transformer hybrid model to achieve efficient object detection in UAV images, which has three advantages that contribute to improving the object detection performance. First, the efficient and effective cross-shaped window (CSWin) transformer can be used as a backbone to obtain image features at different levels, and the obtained features can be input into the feature pyramid network (FPN) to achieve multi-scale representation, which will contribute to multi-scale object detection. Second, a hybrid patch embedding module is constructed to extract and utilize low-level information such as the edges and corners of the image. Finally, a slicing-based inference method is constructed to fuse the inference results of the original image and sliced images, which will improve the small object detection accuracy without modifying the original network. Experimental results on public datasets illustrate that the proposed method can improve the performance more effectively than several popular and state-of-the-art object detection methods.

Heterogeneous images acquired from various platforms and sensors provide complementary information. However, to use that information in applications such as image fusion and change detection, accurate image matching is essential to further process and analyze these heterogeneous images, especially if they have significant differences in radiation and geometric characteristics. Therefore, matching heterogeneous remote sensing images is challenging. To address this issue, we propose a feature point matching method named Cross and Self Attentional Matcher (CSAM) based on Attention mechanisms (algorithms) that have been extensively used in various computer vision-based applications. Specifically, CSAM alternatively uses self-Attention and cross-Attention on the two matching images to exploit feature point location and context information. Then, the feature descriptor is further aggregated to assist CSAM in creating matching point pairs while removing the false matching points. To further improve the training efficiency of CSAM, this paper establishes a new training dataset of heterogeneous images, including 1,000,000 generated image pairs. Extensive experiments indicate that CSAM outperforms the existing feature extraction and matching methods, including SIFT, RIFT, CFOG, NNDR, FSC, GMS, OA-Net, and Superglue, attaining an average precision and processing time of 81.29% and 0.13 s. In addition to higher matching performance and computational efficiency, CSAM has better generalization ability for multimodal image matching and registration tasks.

To effectively guarantee the safety and efficiency of space activities, and to meet the needs of different levels of users for space object situation (SOS) knowledge, formalized and standardized SOS knowledge is needed to assist cognition. A knowledge graph (KG), as a suitable technology that can formally express knowledge and construct a standardized knowledge base, can provide knowledge generation, representation and intelligent service of SOS. In this paper, an SOS information service based on the KG (SOSKG) was constructed. Aimed at different scales of components in SOS, a multi-granularity knowledge structure was constructed, and a formalized expression of SOSKG was realized. To incorporate the complex relations into the SOSKG, a multi-level semantic relation parsing model and formalized representation were proposed from the three aspects of basic relations, spatial relations, and temporal relations. Additionally, a multi-element knowledge construction model was constructed to address the dynamic characteristics of various relations in SOSKG under a complex spatio-temporal environment. This study constructed different cases and scenarios to test the service, and the results show that the proposed service can effectively organize and express SOS knowledge in a complex spatio-temporal environment, provide an intermediate bridge between users and SOS knowledge, and promote users' cognition of SOS knowledge.

The implementation of increased space exploration missions reduces the distance between human beings and outer space. Although it is impossible for everyone to enter the remote outer space, virtual environments could provide computer-based digital spaces that we can observe, participate in, and experience. In this study, Sino-InSpace, a digital simulation platform, was developed to support the construction of virtual space environments. The input data are divided into two types, the environment element and the entity object, that are then supported by the unified time-space datum. The platform adopted the pyramid model and octree index to preprocess the geographic and space environment data, which ensured the efficiency of data loading and browsing. To describe objects perfectly, they were abstracted and modeled based on four aspects including attributes, ephemeris, geometry, and behavior. Then, the platform performed the organization of a visual scenario based on logical modeling and data modeling; in addition, it ensured smooth and flexible visual scenario displays using efficient data and rendering engines. Multilevel modes (application directly, visualization development, and scientific analysis) were designed to support multilevel applications for users from different grades and fields. Each mode provided representative case studies, which also demonstrated the capabilities of the platform for data integration, visualization, process deduction, and auxiliary analysis. Finally, a user study with human participants was conducted from multiple views (usability, user acceptance, presence, and software design). The results indicate that Sino-InSpace performs well in simulation for virtual space environments, while a virtual reality setup is beneficial for promoting the experience.

Compared with general object detection, the scale variations, arbitrary orientations, and complex backgrounds of objects in remote sensing images make it more challenging to detect oriented objects. Especially for oriented objects that have large aspect ratios, it is more difficult to accurately detect their boundary. Many methods show excellent performance on oriented object detection, most of which are anchor-based algorithms. To mitigate the performance gap between anchor-free algorithms and anchor-based algorithms, this paper proposes an anchor-free algorithm called dual-resolution and deformable multi-head network (DDMNet) for oriented object detection. Specifically, the dual-resolution network with bilateral fusion is adopted to extract high-resolution feature maps which contain both spatial details and multi-scale contextual information. Then, the deformable convolution is incorporated into the network to alleviate the misalignment problem of oriented object detection. And a dilated feature fusion module is performed on the deformable feature maps to expand their receptive fields. Finally, box boundary-aware vectors instead of the angle are leveraged to represent the oriented bounding box and the multi-head network is designed to get robust predictions. DDMNet is a single-stage oriented object detection method without using anchors and exhibits promising performance on the public challenging benchmarks. DDMNet obtains 90.49%, 93.25%, and 78.66% mean average precision (mAP) on the HRSC2016, FGSD2021, and DOTA datasets. In particular, DDMNet achieves 79.86% at mAP ₇₅ and 53.85% at mAP ₈₅ on the HRSC2016 dataset, respectively, outperforming the current state-of-the-art methods.

Visual place recognition (VPR) is a task that aims to predict the location of an image based on the existing images. Because image data can often be massive, extracting features efficiently is critical. To solve the problems of model redundancy and poor time efficiency in feature extraction, this study proposes a fast dual-feature extraction method based on a tightly coupled lightweight network. The tightly coupled network extracts local and global features in a unified model which has a lightweight backbone. Learned step size quantization is then performed to reduce the computational overhead in the inference stage. Additionally, an efficient channel attention module ensures feature representation ability. Efficiency and performance experiments on different hardware platforms showed that the proposed algorithm incurred significant runtime savings for feature extraction, and the inference was 2.9–4.0 times faster than that in the general model. The experimental results confirmed that the proposed method can significantly improve VPR efficiency while ensuring accuracy.