Research On Vehicle Classification And Identification From High Resolution Remote Sensing Image

With the continuous development of remote sensing technology and the improvement of satellite spatial resolution,the research of vehicle identification based on high resolution remote sensing image has made some achievements and established a relatively complete vehicle identification process.In this paper,based on vehicle identification process of high-resolution remote sensing image,the method used in the vehicle identification process is improved to obtain higher vehicle identification accuracy.The main contents of the paper are as follows:(1)In the aspect of image preprocessing,this paper uses the sobel edge filtering method to deal with the image and contrasts with the image processing methods such as the linear transformation used in the existing vehicle identification research.Sobel edge filtering can better highlight the vehicle edge feature and conducive to the follow-up vehicles identification.(2)The optimal scale problem of vehicle segmentation in high resolution remote sensing image is studied.Based on the analysis of the characteristics of the vehicle at different scale,this paper proposes the method of determining the optimal scale of vehicle segmentation based on the vehicle area-method of maximum of total vehicle area.The method is based on the area of the vehicle as the basis for determining the optimal scale of vehicle,the total vehicle area calculated at different segmentation scales,when the area reaches the maximum value corresponding to the segmengtation scale is the optimal scale of the vehicle scale.This method is better than the RMAS method based on spectral heterogeneity commonly used in the existing research..(3)The characteristics of the vehicle in the high-resolution remote sensing image are analyzed in detail and the corresponding feature class description is established.The classification methods of threshold classification and fuzzy rules are briefly introduced.Combining with the characteristics of the two classification methods,a dual classifier method of multi-feature threshold-fuzzy rules is constructed to improve the identification accuracy of vehicle with different brightness.(4)A variety of different types of high-resolution remote sensing images are used to experiment,the corresponding evaluation factors are also combined to evaluate the results of vehicle identification,results with 96%of correctness and 89%of completeness and 86%of quality are achieved.Compared with the identification results of bright vehicles at the same resolution,the identification method has higher accuracy than the previous research,and there is little error recognition and omission recognition.The results show that the method of determining the optimal scale of vehicles and the classification method have good universality and feasibility.