Development Of Aircraft In Remote Sensing Image Detection System Based On Improved YOLOv4-tiny

Remote sensing images,as a kind of important real information resources in the field of digital images,have been widely used in various fields.With the continuous improvement of the resolution of remote sensing images,the amount of remote sensing image data has exploded.It contains more types and numbers of objects than general images,and the feature information of the objects is more abundant.Object detection of remote sensing images is one of its key application areas.In this paper,taking the aircraft object as the research object,the deep learning technology and remote sensing object recognition technology are combined to be researched,taking YOLOv4-tiny algorithm to improve and apply,which effectively improves the detection speed and accuracy of the model.The specific process works as follows:(1)Based on YOLOv4-tiny,a lightweight algorithm in the YOLOv4 series,combined with ResNet and SPPNet,which tries to improve the backbone network structure of the algorithm,and goes through experimental verification that the improved algorithm can effectively improve the detection of the algorithm Speed and reduce the occupation of memory space,and this method is more suitable for deployment on embedded devices.(2)In view of the loss of accuracy in the previous improvement,the attention mechanism is proposed to be improved on the basis of further improvement,and three mainstream attention mechanisms are studied and ablation experiments are designed.The experimental results show the CBAM and ECA mechanism are far superior to the SE mechanism,but also the CBAM mechanism is slightly less than the ECA mechanism,which is different from the application conclusions of the attention mechanism in other studies.In this study,CBAM is more conducive to improving the accuracy of the model.(3)Through in-depth research on the model deployment work in the actual application process,this paper uses the Onnx Runtime framework based on C++ to deploy the trained model universally,and develops a target detection system based on MFC combined with Open CV on the Windows platform.(4)Based on the UCAS-AOD data set,the relevant data set was rearranged and produced,and the model was trained by optimization methods such as Mosaic.The robustness of the model is verified by designing related experiments,and the reliability of the algorithm is verified by comparing with the original algorithm and the YOLOv5 s algorithm.The m AP of the model reaches 98.9%,and the detection speed can reach125 FPS.In the last of the research,this paper analyze the problems of missed detection and false detection in the research and give relevant solutions.This paper contains 76 figures,11 tables,and 51 references.