The control problem of leaderless multi-UAV formation in noisy environment is studied, and an event-triggering based strategy is adopted to reduce the number of communications and resource waste. Considering the effects of communication noise and random time delay, a Bernoulli random distribution is adopted to characterize the randomness of the occurrence of time delay. By constructing Lyapunov function and using the analysis method of linear matrix inequality, the stability conditions of multi-UAV system are given to ensure that UAVs can realize formation control under the influence of noise environment and random delay. By using the distributed control input and the neighbor information between UAVs, the system can respond and adjust quickly to form the expected formation. The algorithm is applied to the quadrotor UAV system, and the validity and correctness of the proposed algorithm are verified by theoretical analysis and simulation experiments.

Aiming at the problems of low accuracy, missing detection and false detection in the identification of small targets in UAV aerial images by existing target detection methods, a small target detection method based on improved YOLOv7 is proposed. Firstly, the network detection head is improved, the 20×20 detection head originally used for detecting large objects is removed, a 160×160 detection head for small targets is added, and the network feature fusion path is modified accordingly. Secondly, the SPPCSPC structure is reconstructed, the convolutional layer is clipped and the pooling structure is changed to reduce module complexity and speed up network convergence. Then, the original upsampling structure is replaced by the Content-Aware ReAssembly of FEatures (CARAFE) operator to reduce the loss of image information during upsampling and maximize the preservation of local and corner information of the input image. Finally, the ELAN module is improved to lighten the backbone while improving the sensitivity of the network to small-scale targets. Experiments are carried out on the public dataset VisDrone2019, and the mAP50 of the improved model reached 56.6%, which is 2.9 percentage points higher than that of the original YOLOv7 model, and the parameters are reduced by 33%.

To solve the trajectory tracking problem of Unmanned Surface Vessels (USV) in complex external environment, a control algorithm combining Adaptive Unscented Kalman Filter (AUKF) with Nonlinear Model Predictive Control (NMPC) is proposed. Firstly, the mathematical model of the 3-DOF fully-driven USV is established, and a nonlinear model predictive controller with dynamic weight adjustment based on current tracking error is designed. Secondly, according to the characteristics of complex environment and uncertain sensor noise, an AUKF is proposed to estimate the state by correlating noise changes with measurement parameters. Finally, by adding time-varying random disturbance and measurement noise to simulate the real external environment for the numerical simulation experiment, the certainty and effectiveness of the algorithm are verified.

In order to solve the problems of low target detection accuracy and difficulty in small target detection caused by blurred images and lack of texture features in the existing mainstream algorithms for ship target detection in SAR images, and considering that the real-time performance of the network will be affected by introducing too many parameters, an improved YOLOv7 ship target detection method based on coordinate attention mechanism and Normalized Wasserstein Distance (NWD) metric is proposed. Firstly, the maximum pooling and residual structure are introduced into the coordinate attention mechanism to improve the model feature extraction ability. Secondly, combining dense connection with lightweight convolution, SPPCSPC-P is designed to enhance the fusion between features. In addition, a small target detection layer is added to the backbone network to improve the low detection accuracy of the model for small targets. Finally, the weighted positioning loss function is designed by using NWD metric and CIoU loss, which further improves the model detection accuracy. Experiments are carried out on SSDD dataset, and the experimental results show that the average accuracy of this method reaches 98.38%, which is 2.09 percentage points higher than that of YOLOv7 network.

A infrared small target detection method based on FACET filtering weighted local contrast is proposed to address the problem of low contrast between targets and background and high edge brightness in infrared images under low altitude and complex backgrounds. Firstly, the target candidate pixels are obtained through FACET filtering operation. Then, local contrast is calculated by the ratio of grayscale differences between the target and background regions, and a weighted function is designed based on the heterogeneity of the target and background to enhance target saliency and suppress background. Finally, the real target is extracted through adaptive threshold segmentation. The performance of the algorithm is compared and analyzed with different algorithms on five sets of real infrared small target image datasets. The experimental results show that the algorithm proposed has good detection performance in different low altitude complex scenes.

A multi-scale asymptotic feature fusion algorithm based on YOLOv8 is proposed to address the issues of missed and false detections due to the diversity of target scales, dense small targets, and complex background environments in remote sensing images. Firstly, the Res2C2f module combined with multi-scale residual network is constructed to capture the features of different scales more effectively. Secondly, the pyramid pooling module with cross-level connection is designed to improve the problem of insufficient feature extraction ability of the original pyramid pooling module. Then the multi-scale asymptotic feature fusion network is reconstructed to realize the exchange of multi-scale information, and the features of different levels are fully utilized to enhance the effect of feature fusion. Finally, a small target detection layer with a size of 160×160 is added to improve the detection effect of the model on small targets in dense scenes. In the DOTA dataset, compared with the baseline model, the accuracy, recall, and mean value of average accuracy of the proposed algorithm are improved by 4.8, 4.0 and 3.7 percentage points, respectively.

Aiming at the problem of high miss rate due to the drastic change of target scale, complex background, small and dense targets from the perspective of UAV, an improved YOLOv5s real-time target detection model is proposed. Firstly, a novel hybrid attention mechanism is introduced and embedded into the backbone network to enhance the extraction of crucial target information. Secondly, a new dense residual pyramid pooling is created to improve network information fusion capabilities while reducing computational cost. Then, a C3-BoT module based on multi-head self-attention mechanism is designed to effectively capture the global contextual information of UAV images. Finally, a specialized layer for detecting extremely small targets is added to the YOLOv5s network, specifically tailored to mitigate the issue of miss rate of small objects. Experimental results on the VisDrone2019 dataset show that the improved model achieves an mAP0.5 of 40.6%, an improvement of 8.1 percentage points over the YOLOv5s baseline model, demonstrating superior detection performance in UAV aerial image tasks.

Because there are few visual features of small targets and it is difficult to locate them, many small object detection methods are based on Feature Pyramid Network (FPN) to perform multi-scale fusion to enrich the information of each feature layer. However, FPN only focuses on the local correlation of features and uses element-wise addition operations to fuse different feature layers, ignoring the differences in the receptive fields of different feature layers. Therefore, the Enhance Context Feature Pyramid Network (ECFPN) is proposed, the Context Information Enhancement (CIE) is designed to enhance context information, and the Attention Guided Feature Fusion (AGFF) fuses high-level feature maps and low-level feature maps. The experimental results show that the ECFPN has AP0.5 and APS of 75.05% and 19.48% respectively on VOC2012 dataset, and of 93.48% and 45% respectively on NWPU VHR-10 dataset, which has good small target detection performance.

In order to solve the problems of small targets and target occlusion in UAV aerial images, an improved target detection algorithm YOLO-RC based on YOLOv8s is proposed. The Receptive-Field spatial Attention (RFA) is introduced into the backbone network structure to avoid the sharing of convolutional kernel parameters, so as to improve the image feature extraction performance of the model. The C2f module is improved and deep separation convolution is introduced to reduce the computational cost of the model. A small object detection layer of hybrid attention convolution is added to improve the detection accuracy of small objects. In order to fully consider the geometric features of the predicted image, the MPDIoU loss function is used to optimize the network. Experiments on the UAV image dataset VisDrone2019 show that the mAP@0.5 of the proposed improved algorithm is 44.7%, which is 5.4 percentage points higher than that of YOLOv8s, and the number of parameters is reduced by 1.81×106 with the addition of a small target detection layer. On the DOTAv1.0 dataset, the mAP@0.5 increased by 5.6 percentage points. The improved algorithm has stronger robustness and is suitable for UAV perspective target detection tasks.

To solve the problems of missing target features and strong clutter interference in infrared weak target detection and tracking in complex backgrounds, the target motion feature is introduced into the multi-frame processing method to improve the performance. However, for highly maneuverable targets such as UAVs, the existing multi-frame processing methods based on mechanism models make it difficult to cover their complex and variable motion forms, which makes the result of detection and tracking unsatisfactory. In this regard, a data-driven infrared weak maneuvering target detection and tracking method under complex backgrounds based on a multi-frame processing framework is proposed. Firstly, the MPCM algorithm is used to enhance weak targets. Then, the enhanced results of multiple frames are projected onto a 2D subspace to construct a 2D trajectory detection model based on YOLO. Finally, the 2D detection results are backtracked in 3D space-time to construct a 3D trajectory detection model based on LSTM. When constructing the detection model, data augmentation is performed on the images of real image to ensure that the training samples cover as many target motion forms as possible. In the 2D subspace, a high, performance YOLO detection network is used to quickly eliminate a large amount of clutter. In the 3D space-time, a small amount of difficult-to-eliminate clutter is finely screened out by the LSTM temporal network. The results of comparison experiment demonstrate that the proposed method is capable of achieving real-time performance in terms of time consumption, and the method exhibits an average tracking accuracy of 86.6% across multiple scenarios, a remarkably low false alarm rate of only 9.2%, and an impressive AUC value of 0.982 0 for the ROC curve.

Ground infrared target detection is an important research content in the fields of target reconnaissance, intelligent perception and camouflage protection. Aiming at the target detection model based on anchor frame, it needs the guidance of anchor frame when extracting features, which will produce a large number of calculation parameters related to anchor frame, and lead to inaccurate detection, poor generalization performance and easy to miss detection. Based on the target detection model without anchor frame based on the idea of image segmentation, a backbone network based on deformable convolution for feature extraction is constructed, and the convolution kernel is used to adapt the target shape to enhance the network's extraction effect on target features. Combining the attention mechanism of space and channel, focus on the target from the spatial dimension and channel dimension, realize the three-dimensional attention to the target feature, and improve the target information acquisition ability of the target detection model. The proposed ground infrared target detection model reached a detection accuracy of 91.3% on the Infrared-VOC dataset, and the overall performance of the ground infrared target detection model based on the Anchor-Free frame is optimized.

In the field of hyperspectral target tracking, the continuous variation in target scale leads to diverse appearances of tracked objects between consecutive frames, resulting in a decline in tracking accuracy. To address the challenge of target scale variations, a Transfomer-based Scale-adaptive Hyperspectral target Tracker (TSHT) is proposed. The method aims to significantly improve the accuracy of hyperspectral target tracking. Firstly, principal component analysis is applied to hyperspectral images to obtain three-band images. Subsequently, the search images is preprocessed by SwinBlock-Crop25, and then the template image and the preprocessed search image are input into the D-swin feature extraction module for the extraction of multi-block and multi-layer deep features. Following that, the obtained features are concatenated and subjected to a self-attention mechanism to better capture critical information about the target at different scales. Finally, through a multi-layer perceptron, the search image features processed by self-attention are mapped to the final target bounding box, completing the target tracking process. Experimental results demonstrate that TSHT exhibits high success rates and accuracy. In comparison with several state-of-the-art trackers, it outperforms the advanced target tracking algorithm MixFormer by 0.9 percentage points in handling challenges related to target scale variations, while maintaining real-time performance.

Aiming at the problems of tracking failure and model degradation caused by low utilization rate of polarization factor, complex background and drastic change of target scale and appearance in traditional infrared ship target tracking scene, a bidirectional correlation filtering target tracking algorithm based on joint optimization framework is proposed. On the basis of constructing a new polarization constraint self-adaptation filter-location joint optimization framework, the response map of the main filter is calibrated by polarization mask filter, a parallel double filter structure based on polarization characteristics is constructed, and the convergence of the proposed optimization framework is proved. In addition, a dataset of infrared dim ship target is constructed, and experiments on the dataset prove that the proposed algorithm can realize real-time tracking, and it is more than 15% higher than other related filtering algorithms recently proposed in terms of accuracy, robustness and other common performance indexes.

Aiming at the problems that the intelligent deployment method of air defense formations cannot take into account both regional cover and target cover at the same time, the artificially formulated complex rules are difficult to solve, and the algorithm execution efficiency is low, an air defense formation deployment method based on Independent Multi-Agent Proximal Policy Optimization (IN-MAPPO) is proposed. An independent actor-critic network is designed to adapt to the different roles of fire units. It promotes the collaborative cooperation of fire units to complete hybrid deployment tasks through centralized value functions and reward functions, and improves the resistance capability and the overall deployment performance of the formation. Experimental results show that IN-MAPPO can complete the mixed deployment tasks according to the role of the agent, improve the resistance capability of remote fire units, and reduce the training time by 13.7% compared with other MAPPO algorithms. Compared with existing intelligent algorithms, the coverage area of fire units is increased by 4.2%, the effective cover width is increased by 12.3%, and the execution efficiency of the algorithm increased by 95.9%.

Aiming at the problem of missed target detection and false alarms in sea surface target detection in complex nearshore environments and partial target occlusion conditions, a sea surface target detection method fusing lidar and machine vision is proposed. Firstly, a feature extraction module based on attention mechanism and deformable convolution is designed to improve the ability of YOLOv7-tiny to extract features of sea surface obstacle targets, thereby reducing the missed detection rate and false alarm rate caused by complex nearshore background interference. Then, the lidar clustering results and the improved YOLOv7-tiny network model prediction results are fused to reduce the missed detection rate caused by partial occlusion of the target. Finally, experimental verification is conducted on the sea surface target detection image data set. The results showed that compared with the original YOLOv7-tiny network model, the mAP of the improved YOLOv7-tiny network model is increased by 3.8 percentage points. Experimental verification is conducted on real ship experimental data in a scene where the target is partially occluded. Compared with the NMS algorithm, the missed detection rate of the proposed fusion method is reduced by 6.9 percentage points, which verifies that this method can reduce the missed detection rate and false alarm rate of sea surface target detection in complex nearshore environments and partially blocked target scenarios.

At present, the image dehazing method based on dark channel mostly has the problems of image color distortion and blurred contour of the sky boundary region after restoration. In order to solve these problems, a dark channel dehazing method fused with color enhancement information is proposed. Based on the dark channel prior algorithm, the interval estimation of atmospheric light value is carried out by gray-scale open operation to obtain accurate atmospheric light value and initial transmittance. The grayscale image corresponding to the spatially restored image of YCbCr is used as the guide image, and the coarse transmittance image is refined by gradient domain guided filtering, and then the adaptive tolerance mechanism is used to adaptively correct the transmittance of the sky region and the bright region. Finally, the brightness compensation model is used to correct the image to improve the brightness and color saturation of the restored image. Experimental results on synthetic foggy images and real foggy images show that, compared with the existing single-image dehazing algorithms, the proposed method improves both the peak signal-to-noise ratio and structural similarity, and effectively solves the problems of image color distortion and detail loss after dehazing.