Research On Method Optimization And Accuracy Evaluation Of Vehicle Target Detection In Satellite Image

Vehicle target recognition and extraction using satellite imagery is one of the important traffic information acquisition techniques in the field of intelligent transportation, with the resolution continues to increase, the research of traffic flow parameters extraction based on commercial application of satellite imagery has a great prospect, and becoming a hot research topic in the international arena. This paper aims to optimize the existing vehicle target extraction methods, and verify the effectiveness of the optimization method through experimental results. The main research contents are as follows:(1) Vehicle target extraction process optimization. On the basis of summary and analysis the existed vehicle target extraction process and key links, this paper proposed an object-oriented optimization process for vehicle target detection, which includes three steps:segmentation-oriented enhancement, classification-oriented segmentation and object-oriented classification, and the process of vehicle target extraction is optimized from the general.(2) Vehicle target detection segmentation scale optimization. According to the theory of optimal scale and scale effect in the image processing, this paper tested the optimal segmentation scale in vehicle target detection using the method of "The Mean of Area Ratio" and "Ratio of Mean Difference to Neighbors (ABS) to Standard Deviation (RMAS)". Through the comparison of segmentation results, the RMAS method is selected as the optimal segmentation scale determine algorithm for the Vehicle target detection.(3) Vehicle target detection optimization of the classification. Against to the characteristics of target image object under small-scale segmentation, this paper analyzed the insufficient of nearest neighbor classification and membership function classification which applied in the vehicle target detection. Combined with the vehicle expertise and the advantages of object-oriented fuzzy classification, this paper proposes the method of knowledge-based fuzzy rules classification, which is helpful to improve the classification speed and accuracy of vehicle target.(4) Under the guidance of the optimization method which summarized in Vehicle target extraction processes and key links, this paper carried out the vehicle target extraction experiments using three different high-resolution satellite image data, through the error matrix analysis and Kappa coefficient calculation to evaluate the experimental results, and the main influencing factors are also analyzed.This study has an important reference meaning to improve the vehicle target detection method based on high resolution satellite images, and will make the vehicle extraction more rapid and accurate.