Aircraft Target Detection And Recognition In Remote Sensing Image Based On Deep Learning

Target detection and recognition for remote sensing images is an important research direction in the field of remote sensing technology and computer vision.Among them,the detection and recognition of aircraft targets for remote sensing images is also an important strategic means to maintain national security.With the rapid development of remote sensing technology and the improvement of the performance of hardware equipment such as imaging platforms,the resolution of remote sensing images can reach sub-meter level,and massive amounts of remote sensing image data can be generated every day.In the field of computer vision,the development of target detection technology is becoming more and more mature.In the face of remote sensing image data sets,higher requirements are placed on the detection and recognition of aircraft targets in the image.This article takes optical remote sensing images as the research object,and based on the target detection algorithm based on deep learning,selects representative algorithms from two types of methods based on candidate regions(two-stage)and regression-based(one-stage)for research and optimization.After verification,it is found that the regression-based method is faster and more suitable for practical production applications,and the detection accuracy of the candidate-based method is relatively higher.The main work and innovations of the paper are as follows:1.Since the remote sensing image aircraft target detection and recognition has great application value in the civil and military fields,the paper combines the current technical status and future development trends,briefly summarizes the research significance and application prospects of remote sensing image aircraft target detection and recognition Analyze and combine the main combat models currently equipped by the US military(F-22,B-1,B-52,C-17,C-130)with the current two types of targets for civilian aircraft,and discuss the current depth-based Insufficiency of learning target detection and recognition algorithms,the follow-up research objects and specific test methods of this paper are proposed.2.Because deep learning algorithms usually require massive amounts of data for training,in order to construct the data set required for the experiment,the existing open source data set was analyzed,and it was found that it is difficult to meet the training needs for aircraft target detection of remote sensing images.The thesis is based on the Wuhan University 's DOTA dataset,extracting the aircraft images,and the five main types of US military equipment(F-22,B-1,B-52,C-17,C-130)Perform analysis,use Google earth software to obtain remote sensing images of some US military bases,and then use web crawler technology to supplement and improve,and establish a set of data sets that can be used for the experiments in this article.Thedata set mainly includes civilian passenger aircraft and military aircraft.Military aircraft include five typical target samples(F-22,B-1,B-52,C-17,C-130).The civilian common aircraft is mainly called the PLANE data set,and the military aircraft data set is called the MA(Military Aircraft)data set.The method of establishing the data set shows that this article has certain military application value.3.Select the representative algorithm Faster R-CNN in the two-stage target detection algorithm based on deep learning for in-depth research.The Faster R-CNN algorithm is based on the detection of natural photos and uses its When remote sensing images are detected,they are prone to misdetection and missed detection.In view of the deficiencies of Faster R-CNN in the detection of aircraft targets in the remote sensing image data set,the following optimization is carried out: after fusion of multi-scale feature maps,combined with adaptive threshold setting for detection.Compared with the traditional Faster R-CNN algorithm,the optimized Faster R-CNN algorithm in this paper improves the accuracy of aircraft target detection of remote sensing images and can realize the recognition of military and civilian aircraft.4.In order to improve the efficiency of target detection,the paper conducts an in-depth study on the YOLOv3 target detection algorithm based on regression(one-stage).YOLOv3 is also based on training and detection based on natural photos,and there are missed and misdetected phenomena in the face of remote sensing images.In this paper,based on YOLOv3,the basic backbone network is simplified and optimized,and the DenseBlock structure is added to the network.After comparative test analysis,this paper proposes the optimization algorithm of YOLOv3,while ensuring the detection accuracy is equivalent,the network size is compressed,which greatly improves the detection speed.Compared with the YOLO series algorithm,the optimization algorithm enhances its application value in actual production,and also has certain detection and recognition effects on military aircraft.