Research On Target Detection Algorithm Of UAV Image Based On SSD

In recent years,with the continuous development of UAV technology,UAV aerial photography technology has been widely used in agriculture,geological exploration,urban planning and other fields.In these application scenarios,the target detection of UAV aerial photography image is a very important task,which is of key significance to realize the autonomous flight and target tracking of UAV.SSD-based target detection algorithm is an advanced image target detection algorithm with real-time performance and high accuracy,so it is also widely used in UAV aerial image target detection task.However,in the UAV aerial image task,the image quality is not high,the target size is uncertain,the number of targets and other factors will affect the effect of the detection algorithm,so it is necessary to optimize and improve the algorithm for these special cases.In this thesis,Mosaic data enhancement method is introduced,dual backbone network structure is proposed,feature pyramid model is added,and exclusion loss function is introduced to improve the target detection accuracy.(1)Mosaic data enhancement method is introduced to solve the problem of category unbalance and background noise in data set.This method splices multiple images together as input to the network to enhance the diversity and robustness of the data.Compared with traditional data enhancement methods,this method improves the robustness and accuracy of the detection algorithm.(2)Aiming at the problems of uncertain target size and low detection accuracy of small targets in images,a dual backbone network structure is proposed and a feature pyramid model is added.In this network architecture,two different backbone networks are used to extract high-level and low-level features respectively,which are fused together by a feature pyramid model to fit targets of different sizes.Experimental results show that the detection accuracy of the algorithm is improved.(3)For the problem of mutual occlusion between targets,the repulsion loss function is introduced.The loss function can effectively reduce the mutual interference between targets by introducing the repulsion term,and improve the stability and reliability of the detection algorithm.The experimental results show that the proposed algorithm achieves good performance in UAV aerial photography image target detection task,and the accuracy is increased by more than 5% compared with the traditional algorithm.The algorithm performance is improved and some new ideas and methods are provided for the research of UAV aerial photography image target detection.