Object Detection Under Complex Meteorological Conditions In Aerial Images

Aerial image object detection is one of the important research areas of computer vision.It is widely used in disaster prevention,port control,maritime rescue and military purposes.It also has prosperous prospect especially in the field of public security,anti-terrorism,security inspection and other fields.However,the quality of aerial image is susceptible to weather conditions.For instance,clouds and fog may blur aerial images or occlude objects.The shooting location is often at high altitude,and the object is relatively small in the image.The surrounding environment will cause strong interference to the detection of objects.For complicate meteorological conditions which cause the problem of low image quality of aerial images(regardless of extreme meteorological conditions which will lead to the impossibility of aerial images),this thesis presents dehazing algorithm to promote imagine sharpness.In the meantime,for the problem of difficult object detection in aerial image,the thesis proposes object detection method.The specific research content is as follows:The current mainstream image dehazing algorithm are studied and the dehazing effects of each algorithm are compared.Through the contrast experiment of dehazing image entropy function contrast experiment and image dehazing effect,it is concluded that the dark channel dehazing algorithm optimized by guided filter can more effectively reduce the influence of cloud on aerial image and improve the quality of aerial image.In this paper,the state of the object detection algorithms are studied,and the improved object detection algorithm YOLO-C based on YOLOv3 is proposed.In the comparative experiment,the proposed algorithm achieved mean Average Precision(mAP)of 90.1% on the public dataset of the China Aerospace Science and Technology Fourth Institute Command Automation Center,which exceeded other algorithms.Through the contrast experiment of object detection before and after dehazing,we have the conclusion that dehazing can improve object detection accuracy.The YOLO-C algorithm can improve the average precision of object detection by 16.1% after dark channel dehazing.The research results show that the proposed algorithm in this paper can provide an implementation method for aerial object detection under complex meteorological conditions.The image dehazing and object detection are combined to realize the aerial object detection system under complex meteorological conditions.