Learning sample selection in multi-spectral remote sensing image classification using BP neural network

Through analyzing the influence of the learning samples' location in the sp ectral space on the accuracy of multi-spectral remote sensing image classif ication using BP neural network, a method for learning samples selection ba sed on x(2) distribution was presented and used in TM image classification. The classified results of the 6 ground objects with BP classifier using di fferent learning samples selecting methods and the Bayes classifier show th at the BP classifier with the presented learning samples selection method c an not only reduce the number of learning samples greatly which leads ro sh orter learning time, but also improve the classification accuracy compared with the existing methods.