Application Of Hyperspectral Band Selection In Detection Of Oil Slick Thickness

Since the development of imaging spectrometer is increasingly sophisticated, the study of high-spectral remote sensing images is increasingly urgent, hyperspectral and multispectral data exist a lot of redundant information,in the process of estimating the film thickness, the dimension reduction is an important part of entire process,it is a necessary condition for good follow-up treatment. In this paper, the hierarchical agglomerative clustering algorithm based on band selection method is applied to AVIRIS band image sensor 50, and on this basis the application of the shortest distance, most long-distance method, the middle distance, center of gravity method, the average distance and minimum variance comparative analysis method, experimental results show that most long-distance method to choose the 25-band when the running time 44 minutes 54 seconds for the classification accuracy can reach 87.0636%. This method has the highest classification accuracy, run the shortest distance in the calculation of the six, so this paper select the method of the most long-distance calculation. Finally the hierarchical agglomerative clustering algorithm is applied to the existing laboratory equipment, remote sensing images (infrared imaging system, multi-spectral camera, UV cameras) to obtain the 5-band image.Among the band selection model, Select widely used hierarchical agglomerative clustering algorithm, and the method used to estimate the film thickness, the core of this paper is to establish a hierarchical agglomerative clustering model. As the experimental constraints,this paper AVIRIS sensor has 50-band imaging and laboratory equipment for remote sensing images of 5-band image and the model test experiments, the experimental results obtained. Experimentally investigated the recognition accuracy of this model and speed, and on this basis, using spectral angle mapping to class 5-band image of the film thickness to classify, identify the film thickness, the estimated volume of oil spills.The results show that the hierarchical clustering algorithm for identification of model performance is the best. It is a promising approach. More suitable for oil spill hyperspectral image.