In the field of remote sensing, cloud and cloud shadow will result in optical remote sensing image contamination, particularly high cloud cover, which will result in the complete loss of certain ground object information. The presence of thick cloud severely limits the use of optical images in production and scientific research, so it is critical to conduct further research into removing the thick cloud occlusion in optical images to improve the utilization rate of optical images. The state-of-the-art cloud removal methods proposed are largely based on convolutional neural network (CNN). However, due to CNN’s inability to gather global content information, those cloud removal approaches cannot be improved further. Inspired by the transformer and multisource image fusion cloud removal method, we propose a transformer-based cloud removal method (Former-CR), which directly reconstructs cloudless images from SAR images and cloudy optical images. The transformer-based model can efficiently extract and fuse global and local context information in SAR and optical images, generating high-quality cloudless images with higher global consistency. In order to enhance the global structure, local details, and visual effect of the reconstructed image, we design a new loss function to guide the image reconstruction. A comparison with several SAR-based cloud removal methods through qualitative and quantitative experimental evaluation on the SEN12MS-CR dataset demonstrates that our proposed method is effective and superior.

Autonomous parking in an indoor parking lot without human intervention is one of the most demanded and challenging tasks of autonomous driving systems. The key to this task is precise real-time indoor localization. However, state-of-the-art low-level visual feature-based simultaneous localization and mapping systems (VSLAM) suffer in monotonous or texture-less scenes and under poor illumination or dynamic conditions. Additionally, low-level feature-based mapping results are hard for human beings to use directly. In this paper, we propose a semantic landmark-based robust VSLAM for real-time localization of autonomous vehicles in indoor parking lots. The parking slots are extracted as meaningful landmarks and enriched with confidence levels. We then propose a robust optimization framework to solve the aliasing problem of semantic landmarks by dynamically eliminating suboptimal constraints in the pose graph and correcting erroneous parking slots associations. As a result, a semantic map of the parking lot, which can be used by both autonomous driving systems and human beings, is established automatically and robustly. We evaluated the real-time localization performance using multiple autonomous vehicles, and an repeatability of 0.3 m track tracing was achieved at a 10 kph of autonomous driving.

Ship detection and recognition in Synthetic Aperture Radar (SAR) images are crucial for maritime surveillance and traffic management. Limited availability of high-quality datasets hinders in-depth exploration of ship features in complex SAR images. While most existing SAR ship research is primarily based on Convolutional Neural Networks (CNNs), and although deep learning advances SAR image interpretation, it often prioritizes recognition over computational efficiency and underutilizes SAR image prior information. Therefore, this paper proposes YOLOv5s-based ship detection in SAR images. Firstly, for comprehensive detection enhancement, we employ the lightweight YOLOv5s model as the baseline. Secondly, we introduce a sub-net into YOLOv5s, learning traditional features to augment ship feature representation of Constant False Alarm Rate (CFAR). Additionally, we attempt to incorporate frequency-domain information into the channel attention mechanism to further improve detection. Extensive experiments on the Ship Recognition and Detection Dataset (SRSDDv1.0) in complex SAR scenarios confirm our method’s 68.04% detection accuracy and 60.25% recall, with a compact 18.51 M model size. Our network surpasses peers in mAP, F1 score, model size, and inference speed, displaying robustness across diverse complex scenes.

Global localization is a key problem that needs to be solved for single-line LiDAR based robot navigation since it will directly affect the estimation accuracy of the robot's initial pose and the success rate of recovering the robot's state when it loses its local pose. Existing studies to deal with this problem usually extract feature points from laser beams and then resort to fast retrieval and registration methods to further determine the robot's pose. Although these methods have achieved good results in specific scenes, they often fail to perform well when the robot is far away from the map-building trajectory. It is therefore highly desirable to develop more robust techniques for this problem. In this work, we propose a novel solution which is based on “Dense 2D Signature and 1D Registration”, D2S1R for short. By establishing a dense signature database for 2D locations and combining with the fast retrieval technology, the 2D search space is extremely compressed. Furthermore, fast yaw angle determination is achieved by converting scan points to 1D space and measuring the difference of scan contours based on relative entropy. Experimental results on several complex indoor scenes show that D2S1R can complete global localization within 0.03s on an ordinary CPU in an area of nearly 4,000m ² . Besides, on the premise of achieving a location accuracy of 0.08±0.04m and an orientation accuracy of 0.72±0.60°, it can achieve an average success rate of 95% on all test datasets.

In the realm of intelligent transportation systems, Automatic License Plate Recognition (ALPR) is integral, yet is often hindered by heavy computational loads and massive memory footprints. Existing methods either enhance recognition accuracy through complex models or minimize resource requirements by utilizing low-performance models and low-resolution inputs, both of which pose challenges in real-world applications. To address these issues, this paper introduces a novel ultra-lightweight ALPR system specifically designed for deployment on microcontrollers, offering a cost-effective and energy-efficient solution for large-scale applications. We simplify the process by decoupling ALPR into three sub-tasks: detection, localization, and recognition. For each sub-task, specific improvement strategies such as Residual Log-likelihood Estimation (RLE)-based keypoint localization, adaptive data augmentation, and multi-voting recognition mechanism are designed. Our proposed system naturally mitigates the issues of large perspective transformations frequently encountered in real-world settings. As a supplementary contribution, we introduce and release the TJLP dataset. Practical application and superior performance of our method have been demonstrated on resource-limited devices like RV1106 (ARM Cortex-A7 with a 0.5 TOPS NPU), achieving an inference speed of 22 FPS with a power consumption of 629 mW. Notably, our fully optimized model exhibits a recognition accuracy of 99.23% in real-world environments.

Human activity recognition (HAR) in complex scenarios, particularly in through-wall environments, represents a significant frontier in sensing technology research. The implementation of reliable HAR systems faces substantial challenges due to severe noise interference, limited sensor penetration capability, and unpredictable human motion patterns. This work presents two main contributions to address these challenges. First, we develop an integrated HAR system that combines advanced radar signal processing with deep learning approaches. The system features two key innovations: the usage of compact multi-input multi-output (MIMO) ultra-wideband (UWB) radar for capturing multidimensional echo sequences, and a proposed MobRNet architecture that enables efficient feature extraction through integrated radar modules. Second, to systematically evaluate wall obstruction effects on UWB radar data, we establish a comprehensive radar dataset where wall presence serves as the controlled environmental variable with different materials and thickness. The dataset comprises 29,900 radar feature images collected from real indoor environments with various wall materials and thicknesses, accompanied by corresponding video data for validation. Experimental results demonstrate that the proposed system achieves a recognition accuracy of 99% on the total dataset and maintains robust performance with previously unseen subjects, representing an 8% improvement in comprehensive recognition performance compared to state-of-the-art methods.