UAV Detection Image Recognition Based On CNN

The important strategic position of military UAV(unmanned aerial vehicles)and the wide application of civilian UAV in many fields indicate that the era of UAV has come.Image recognition as an important visual function UAV,helps UAV accomplish various tasks in complex environments.At present,UAV image target detection mostly adopts the traditional method of combination of artificial feature extraction and machine learning classifier.With the application expansion in the horizontal field of UAV and the increase in the task difficulty in the vertical depth,the traditional target detection model has been difficult to meet the increasing accuracy requirements.Based on the background,the study aims at through deep analysis of UAV image detection and the long-term study of target detection algorithm,compare traditional model and target detection of deep learning,find out the feasibility and advantages of using deep learning in this study scenario.This paper designs a way of target detection method based on convolution neural network(CNN),completed the construction of the framework of basic experiment,achieved the process from the original dataset to the load enhancement dataset,network training verification from 0 to 1,and the control test is designed for different optimization schemes.In the last step feasible schemes are added into the iterative process of the networkThis paper proposed an aerial image target detection technology based on improved Faster R-CNN(Regions with CNN),which not only realized the algorithm model but improved a lot on the basis of experiments.Aim at the problem of small target false detection and scale difference in aerial image dataset,this paper designs the network structure of RPN(Region Proposal Network)and the optimization scheme of related algorithm.By improving CNN and OHEM(Online Hard Example Mining)algorithm,a lot of adjustments have been made to the.structure of Faster R-CNN so as to improve the overall network effect.