For the recognition of armored targets in complex battlefield environments, how to reduce missed and false alarms while achieving real-time is an urgent issue. To this end, the GCD-YOLOv5 algorithm is innovatively proposed. Firstly, array lidar is used to acquire the armor target data. Secondly, the armor target data is expanded with an improved GAN(Generative Adversarial Network) to increase the diversity of training data. Afterward, the expanded dataset is fed into the GCD-YOLv5(You Only Look Once) for training. And the GCD-YOLOv5 is reflected in the following aspects. Firstly, the CBAM(Convolutional Block Attention Module) and the multi-scale feature fusion are added to improve the feature extraction capability and detection efficiency, increasing the recognition capability of small and obscured targets. Secondly, combining with DETR(Detection Transformer) to lighten YOLOv5 to achieve the real-time requirement. Thirdly, the YOLOv5 loss function and prediction box filtering method are improved to increase the detection accuracy and the confidence of the detection boxes. The experimental results show that the GCD-YOLOv5 algorithm has higher accuracy and real-time, the mAP(mean Average Precision) can reach 99.7%, and fps is 68.56% higher compared to YOLOv5, which significantly improves the recognition capability of armored targets in complex battlefield environments.

A snake robot is a type of highly redundant mobile robot that significantly differs from a tracked robot, wheeled robot and legged robot. To address the issue of a snake robot performing self-localization in the application environment without assistant orientation, an autonomous navigation method is proposed based on the snake robot’s motion characteristic constraints. The method realized the autonomous navigation of the snake robot with non-nodes and an external assistant using its own Micro-Electromechanical-Systems (MEMS) Inertial-Measurement-Unit (IMU). First, it studies the snake robot’s motion characteristics, builds the kinematics model, and then analyses the motion constraint characteristics and motion error propagation properties. Second, it explores the snake robot’s navigation layout, proposes a constraint criterion and the fixed relationship, and makes zero-state constraints based on the motion features and control modes of a snake robot. Finally, it realizes autonomous navigation positioning based on the Extended-Kalman-Filter (EKF) position estimation method under the constraints of its motion characteristics. With the self-developed snake robot, the test verifies the proposed method, and the position error is less than 5% of Total-Traveled-Distance (TDD). In a short-distance environment, this method is able to meet the requirements of a snake robot in order to perform autonomous navigation and positioning in traditional applications and can be extended to other familiar multi-link robots.

In simultaneous localization and mapping (SLAM) technique, the presence of cumulative errors in laser odometry positions leads to localization bias or even mapbuilding failure. To address this problem, a laser SLAM method called PA-SLAM based on ground plane constraints and twostep accurate identification of loopback frames is proposed. In the point cloud preprocessing process, the Patchwork++ method is introduced to obtain the ground plane constraints to reduce the localization error caused by invalid feature point matching. Then, accurate six-degree-of-freedom bit positions of the carrier are obtained from Lidar Orientation and Map Construction (LOAM), candidate loopback frames are obtained by searching within a certain radius according to the k-dimensional (K-D) tree data structure, and a Lidar Iris descriptor is introduced to calculate the similarity scores among the candidate frames. Finally, the Fast Generalized Iterative Closest Point (FasterGICP) algorithm is introduced to obtain the loopback position constraints to correct the global pose, so as to obtain more accurate localization information. The proposed method is compared with LeGO-LOAM and SC-LeGO-LOAM algorithms on the KITTI dataset. The experimental results show that the absolute position root-mean-square error (RMSE) of this method is reduced by 45.7% and 48.3% on average relative to the above two algorithms, which plays a significant role in eliminating the cumulative error of the position, and realizes a more robust and accurate localization and 3D map construction.

Unexploded ordnance (UXO) presents a significant risk to both the natural environment and human safety. Current deep learning detection mechanisms are characterized by limited interpretability, resulting in a persistent absence of detection methods that are simultaneously efficient, secure, and precise. To address these challenges, a dual-mode detection method based on interpretation-promoted YOLOv5 is proposed in this paper. A comprehensive dataset has been constructed to focus on the scarcity of low-altitude UXO target datasets, incorporating both visible light and infrared imagery. Dataset augmentation has been implemented through the application of generative adversarial networks for image super-resolution reconstruction, thereby enhancing the robustness of the dataset. To tackle issues of low detection accuracy and inadequate background discrimination associated with a single information source, an integration of visible light and infrared data has been proposed to enhance the interpretability of the YOLOv5 algorithm, leading to improved detection performance. Extensive low-altitude detection experiments were conducted in field environments using UAVs. Experimental results demonstrate that the proposed method achieves a remarkable detection accuracy of up to 97.1% and an impressive detection speed of up to 60.3 frames per second.

The target tracking of unmanned aerial vehicles (UAVs) has attracted significant attention recently with the increasing application of UAVs, yet few studies have made breakthroughs in dynamic dim target detection. How to efficiently and accurately identify dynamic dim targets in complex contexts poses a challenge. To address this issue, we proposed an improved lightweight Siamese network (ILSN) with an optimized feature-extraction network design and similarity measurement.As for the feature-extraction network, we built a position-wise attention module to obtain the target feature’s position information, which enhanced the network’s ability to extract weak targets while reducing the model parameters, thus ensuring the network is lightweight. For the similarity-measurement module, the tracking accuracy was expected to be improved by deeply mining the localization information of the shallow features and the semantic information of the deep features in the feature networks. To evaluate the performance of the proposed method, we established a simulated experimental environment and a physical experimental platform and then carried out comparison experiments on the attention modules, tracking accuracy, and efficiency performance. The experimental results showed that, compared with the five introduced comparison algorithms, the ILSN had apparent advantages in tracking accuracy: the tracking speed reached 108 frames per second, which met the real-time requirements while improving the success rate.