Research On Key Technologies Of Intelligent Compression Of Detection Images

In order to acquire the high quality reconnaissance images through lowcapacity channels,this dissertation focused on the research on the key technologies of reconnaissance image intelligent compression,mainly solved the problem of automatic selection and rapid recognition of target features.The main research contents and achievements of the dissertation are as follows:(1)Aiming at the problem of less usable information and diverse scenes in reconnaissance images,a method of target characteristic analysis based on non-parametric kernel density estimation was proposed.Firstly,the features of the target and the background were extracted.Then,the nonparametric kernel density estimation method was used to establish the background region feature distribution model.Finally,the matching probability between the target feature and the background feature distribution model was calculated to guide the feature selection.Experiment results of different scene images show that the proposed method can accurately calculate the matching probability of the target and the background using a single image,which met the actual feature recognition accuracy,proved that the method can be used for reconnaissance images feature selection.(2)To improve the speed of target recognition,a fast target recognition algorithm based on regional characteristic gradient was proposed.First,the gradient integral image was used to calculate the average gradient of each region in the image to be inspected,which size is as same as the target template.Then,the region with similar gradient mean value to the template image gradient was defined as a potential target region.Finally traditional target recognition algorithm was used to detect the target in the potential area and the main direction of gradient was used to solve the target rotation problem at same time.Experimental results of the reconnaissance images in different scenes show that the proposed algorithm can hold the high recognition accuracy and improve the recognition speed at same time,also can solve the problem caused by rotation.(3)Completed the development and testing of target characteristic analysis software and intelligent compression software.