Target Recognition Technology Based On Feature Fusion

The technology of target recognition which based on feature fusion is one of the key technologies used in the battlefield detection systems. This paper is the research of infrared images, including preprocessing of the infrared images, extraction and fusion of the targets'feature, classification and other key technologies. In the study of this subject, mainly in the following areas:(l)The preprocessing of infrared images, enhancement and threshold segmentation is studied. Image enhancement based on histogram equalization and gray value of the morphological gradient algorithm, and compared the effect of two methods;in the algorithm of threshold segmentation, adopting the iterative threshold segmentation and Otsu method, these two automatic threshold segmentation algorithm are described in detail, and compared results of the two segmentation algorithms;(2)The extraction and fusion of infrared image, in the aspects of feature extraction from infrared images,the features of geometric and image torque were analyzed; in the aspects of feature fusion, two types of infrared images which are close or far from the detector were analyzed, and according to the different characteristics of them I put forward methods respectively base on weighted summation and weighted combinations for fusion of feature, and analysis the characteristic quantities of befor and after fusion possessing stability and the characteristics of Distinction, and lay a good foundation for target identification in the future;(3)The designment of the classifier, I summarize some basic knowledge used in classifier's designing, after the infrared image preprocessing and feature extraction and fusion, the minimum distance classifier has been designed. Infrared images extracted by the two types of target characteristics and the characteristics's fusion of the two types of features,the target classification and recognition were tested, the results show that the fusion characteristics of the recognition rate is higher than non-integration features; also indicate that the minimum distance classifier method is more effective and it can be used as options for the technology of classifier.In this paper, the various algorithm of preprocessing, feature extraction and fusion, designment of the classfier were simulated and verified the feasibility of these algorithms.