In view of current low accuracy issue in small target detection from Unmanned Aerial Vehicle （UAV） perspective， a multi-scale small target detection algorithm from UAV perspective based on Channel-Prior-Multi-Scale cross-axis attention-YOLO （CPMS-YOLO） was proposed. Firstly， a multi-scale attention module named CPMS （Channel-Prior Multi-Scale cross-axis attention） was incorporated into the backbone network， and the module was designed to better extract and emphasize useful features in complex backgrounds. With this module， the algorithm was able to learn the location details of the region of interest more easily and improve the feature extraction ability of small targets at different scales in complex backgrounds. Secondly， the Backbone network and feature fusion network were restructured by adding a feature layer with the enriched small target semantic information， and the fusion module MultiSEAM （Multi-scale Separated and Enhancement Attention Module） was adopted to complement contextual feature information for each other， thereby detecting and recognizing small targets better. Thirdly， a C2f-RFE （C2f-Receptive Field Enhancement） module was designed to improve the deep C2f （Faster Implementation of CSP Bottleneck with 2 convolutions） module in the Neck network， so as to expand the receptive field of the feature map， thereby realizing more accurate， faster， and multi-angle localization of target features， and thus enhancing small target detection ability. Finally， a loss function named WIoUv3 （Wise-IoU （Intersection over Union） v3） was introduced to optimize the loss weights of small targets dynamically， so as to solve the difference problem between positive and negative samples in the bounding box regression task， thereby further improving the detection ability for small targets. Experimental results on the public dataset VisDrone2019 show that compared to the baseline algorithm YOLOv8s， the proposed algorithm improves the precision， recall， mAP₅₀ （mean Average Precision at IoU threshold of 50%）， and mAP_50-95 （mean Average Precision at IoU thresholds from 50% to 95%） by 5.9， 5.8， 6.3， and 3.6 percentage points， respectively. It can be seen that the multi-scale small target detection algorithm for UAV perspective based on CPMS-YOLO can capture and recognize small targets more accurately.

To improve the security and reliability of image transmission, an image encryption algorithm based on combination Low Density Parity Check (LDPC) coding with chaotic system was proposed. Firstly, the algorithm used parity encoding to extend pixels' value of image into 10 bits and calculated its deviation acted as chaotic initial value. Secondly, Arnold transformation was used to scramble the positions of image pixels and Henon mapping was used to diffuse the values of pixels. Finally, the high 2 bits were separated from 10 bits of pixel value and transmitted faultlessly by LDPC code, the other 8 bits acted as the encryption result. The experimental results show that the proposed algorithm has strong sensitivity to the keys and plaintext, possesses favorable avalanche effect, and it can resist plaintext attack and differential attack effectively. Moreover, the encryption result has strong ability of resisting cutting and noise attacks.

Selective Neural Network Ensemble (NNE) methods based on Decimal Particle Swarm Optimization (DePSO) and Binary Particle Swarm Optimization (BiPSO) were proposed in this paper. The basic idea of the methods was to optimally select Neural Networks (NNs) to construct NNE with the aid of PSO. This may maintain the diversity of NNs and decrease the effect of collinearity and noise of sample. Meanwhile, chaos mutation was adopted in order to increase the diversity of particles of PSO. The experimental results show that the chaos PSO algorithm is an effective ensemble method, and it may improve the generalization ability of NNE in comparison with the available ones.