计算机辅助图像分类信息挖掘与应用探讨*

以沿海地区TM图像应用为例，通过量化方法由计算机对训练区的分离度进行确定与比较，进行最佳信息维数的确定与最佳信息维组合的选择，并将信息维数对分类精度的影响进行探讨。结果证明，这种方法可以得到能满足工作需要的分类精度，组合维数以6维最佳。

Study on Method of Improving Image Classification Precision Using Multi-Dimensions Information

We mining and collect message from the original image up to 75 dimensions and select 18 dimensions from which the most benefit for classification. After trying all the possible combinations of the multi-dimensions information by introducing in three statistical functions, we found that the combination of 6 dimensions can give the best result and the result of all the three methods show that more dimensions of the information give no help for improving precision. Then we select one of the best results with best kappa coefficient and best overall accuracy and apply it to the TM image of CIXI region in Zhejiang prov-ince, EASTCHINA. The result of the final classification is far better than that by the original bands of the image. The road can be identified from the adjacent environment and the lake can be differentiated from the ocean. All this will benefit the relative work followed.