Interpretation Artificial Neural Network in Remote Sensing Image Classification

The responses of neural networks for uniform and normal distribution are studied, especially the BP and RBF neural networks and the question of combination between neural networks and fuzzy logical is answered by experiments. Linear relationship among sample feature components which impact the time consumption and convergence accuracy of networks has been discussed also. In the condition of feature vector included original bands and good separating degree components, BP and RBF neural networks combined with Fuzzy Reasoning have been used for TM image classification. Overall classification accuracy and Kappa coefficients reached 0.915 and 94.33% in RBF network which is higher than 0.845 and 89.67% in BP network.