Analysis of Posterior Probability Uncertainty for Classification of Hyperspectral Images by SupportVector Machines

This paper analyzes the uncertainty of classification posterior probability of support vector machine (SVM) using urban hyperspectral images. The hyperspectral images in Zhangye are selected as the study zone, and the sample parameter data were acquired based on the high resolution images and the ground survey information, the images were classified with parameter-optimized SVM to obtain the posterior probability graph for each class, and the posterior probability graphs were truncated using the threshold values of 0.2, 0.4, 0.6, 0.8 and 0.9 for analysis of the accuracy change of ground object classification at different probabilities. The results show that with the increase of truncation probability, the user accuracy in the classification results increases continuously, while the producer accuracy shows a declining tendency, and the overall classification accuracy also shows a declining tendency. The analysis of the posterior probability distribution of various types of ground objects shows that it is difficult to distinguish the posterior probability of some mixed ground objects. The untrained water body targets can be easily distinguished by the truncation probability, but the posterior probabilities of untrained red materials and white materials are mixed together. This shows that there exist some conditions in which the posterior probability of optimized SVM can not directly and effectively indicate the distinction of ground objects. The posterior probability should be used optionally, and at the same time, it is necessary to construct a more robust calculation method for the posterior probability.