Aircraft Recognition In Remote Sensing Image

Aircraft is an important military target. The study of aircraft recognition has been exploring since the development of remote sensing technology. This paper presents a space-based remote sensing image of the aircraft target recognition. First, the text segmentation for targeting aircraft, a scale based on affine invariant feature transform (ASIFT) of the SVM segmentation. On this basis, the plane target extraction based on wavelet fractal dimension, affine invariant moments and density of the weighted fusion Characterization of aircraft, the establishment of the sample characteristics of the training library. Finally, the binary tree-based kernel clustering SVM classification decision on the aircraft type.ASIFT is scale invariant feature transform (SIFT) based on the generated image is a local feature. It can accurately simulate changes caused by the camera axis orientation of all deformed, the establishment of affine invariant scale space, and then to SIFT, not only keep the target on the basis of important information, reducing the amount of data to be processed and computational, also has strong anti-affine performance and scale, translation and projection invariance. Concave shape of the aircraft is a polygon, and the high reflectance area of the airport, so it can be good target ASIFT characteristics of the local representation. In this paper, SVM using the features of segmentation, targeting a region of interest aircraft (ROI). Experimental results show that the method can overcome the shadows and environmental factors have produced the phenomenon of segmentation.Aircraft in remote sensing images relatively small proportion, so to strictly distinguish between them, while in access to global features, you need to take into account their local details. However, the local and global characterization of the capacity of the aircraft is different. Therefore, this wavelet to extract the fractal dimension plane, affine moment invariants and shape of the three characteristics of density and weighted them, integration into a feature for identification of input. To determine the optimal weights, the first feature based on the classification accuracy of a single level of the initial value of weights selected, then certain steps to test the weights constantly updated, with the highest classification accuracy corresponding weights, is the best Weights.SVM has strong learning ability of small samples, so the text using support vector machine to identify the category of aircraft. Aircraft Type Recognition is a multi-classification problem, in order to improve the training and testing time, this binary tree-based kernel clustering support vector classification. A text database containing 300 aircraft, which come from 6 categories, was constructed. In addition, a training database containing 240 aircraft was adapted to train and the remaining for testing. Experiments show that features fusion, the recognition rate of the aircraft has been improved significantly, and the use of nuclear clustering BT SVM-based classification of the test to shorten the training time.