Hyperspectral remote sensing describes the spectral characteristics of ground objects in detail, which improves the reliability of ground object recognition. However, with the increase of spectral dimension, it also brings a lot of redundant data, which increases the burden of hyperspectral data processing and information recognition, and also affects the accuracy of ground object recognition. Therefore, it is very important to reduce the dimension of hyperspectral data and select characteristic bands when identifying ground objects. Support Vector Machine (SVM) is a kind of machine learning algorithm, which was first proposed by Vapnik of bell laboratory to solve the classification and regression problems (Vapnik, 1995). SVM has good generalization ability and overcomes the dimension disaster of machine learning to some extent. In recent years, SVM and improved SVM based on other algorithms have been widely used in hyperspectral image classification, and achieved good classification accuracy (Melgani et al., 24; Li Zuchuan et al., 211). However, for the redundancy of hyperspectral data, Particle Swarm Optimization (PSO) algorithm has a good advantage in finding the best combination of characteristic bands and further improving the classification accuracy of SVM.
PSO algorithm is a machine learning algorithm to find the optimal solution through the cooperation between individuals and groups, and it has the ability of self-adaptation, self-organization and getting the optimal solution quickly. PSO algorithm was first proposed by Kennedy and Eberhart, and later, in order to make PSO have a wider range of applications, they proposed binary PSO algorithm (Kennedy et al., 1995, 1997; Khanesar et al.，27； Zhang Hao et al., 28). Since the PSO algorithm was put forward, it has been widely concerned in various research fields. In the application of hyperspectral remote sensing, Monteiro and Kosugi(27) put forward the method of selecting the best band combination and the best band number of hyperspectral images based on PSO, and compared with the traditional band selection method through experiments, it proved the superiority of feature band selection based on PSO. Ding Sheng et al. (21) proposed a PSO－BSSVM classification model, which was used to select the characteristic bands of hyperspectral images and optimize the parameters of SVM. Compared with other methods, the model can improve the classification accuracy. Li Linyi and Li Deren (211) also used PSO algorithm in the selection of fuzzy features. In a word, PSO has been successfully applied in the feature band selection of hyperspectral image classification, but it is easy to get premature and fall into local optimum, so it is very meaningful to improve PSO in order to achieve higher SVM classification accuracy. Tent mapping is a typical example of chaotic mapping in chaos theory. Tent mapping is random and ergodic, so adding Tent mapping to PSO can improve the situation that PSO algorithm is easy to fall into local optimum. In this chapter, the improved Tent mapping is applied to the binary PSO algorithm to find the optimal combination of feature bands for SVM classification of hyperspectral images.