Research On Small Target Detection And Location Method Based On Aerial Image

With the progress of the times and the development of science and technology,human society has entered the era of artificial intelligence.Computer vision has replaced the human eye in many areas of life and work,so as to realize the automatic analysis and processing of interesting information in videos or images.As an indispensable research direction in the field of computer vision,target detection and localization have achieved rapid development in the past ten years.This subject integrates high-tech topics in the fields of machine vision,image processing,and pattern recognition.It is also an important cornerstone of other complex vision tasks.It has been extensively used in fields such as intelligent recognition,military investigation,traffic law enforcement,and visual navigation.Therefore,the detection and positioning of small targets in aerial images are of great research value.This paper conducts research from three aspects: small target detection based on the improved YOLOv4 algorithm,small target positioning under the oblique perspective based on BP neural network,and system design and application.Research and improve the detection algorithm of small targets.Aiming at the problem that small target pixels in aerial images account for a small proportion of pixels and it is difficult to extract information so that less information can be provided to the detection model and the details of small target detection are easy to lose,a small target detection algorithm based on improved YOLOv4 is proposed.The algorithm improves the training strategy by estimating the target size and data segmentation based on the density map.It also uses jump connection,feature fusion,and other methods to modify the YOLOv4 network structure,which merges the shallow detail information with the deep semantic information to enhance the small characterization of the target.Experiments and evaluations on a self-made data set show that the improved algorithm almost maintains its running speed,and the average accuracy of small targets is improved by 7.9% compared with the original algorithm.A target localization algorithm based on BP neural network under variable inclination angle is proposed.Aiming at the problem of target positioning under the oblique angle of view,the symmetry center of the target in the image pixel coordinate system is inconsistent with that in the world coordinate system,and the target is deformed in the image due to camera distortion,a new method based on BP neural network is proposed to solve small target localization under variable inclination angle.First of all,produce and process a large amount of target positioning data.Afterward,design the BP neural network structure to fit the positioning data,which convert the pixel information of the target to the distance information in the earth ellipsoidal coordinate system.In the next part,the position of the target in the earth ellipsoidal coordinate system is obtained.In the end,a comparative experiment was performed on the test set,and the results showed that the proposed algorithm can overcome the effect of changing camera inclination on the target positioning based on the linear camera-based imaging model,the positioning results with an average error of about 0.96 m are obtained,and the effectiveness and feasibility of the proposed algorithm are verified.Design and develop system software.Aiming at two different conditions and requirements in application scenarios,based on the above-designed algorithm,small target detection and positioning system based on aerial images and small target detection and positioning system based on aerial video are designed and implemented.Through a large number of experimental verifications,it is shown that these systems can detect,track and locate small targets in aerial images,and can display the target’s location on the AMap.