Relationship between Fisher Classification and Distance Measurement

Fisher classification and distance measurement methods are classification method commonly used for remote sensing monitoring of forest resources. With the advent of the era of big data, there are more and more related studies, but there is a lack of research on the relationship between the two. This paper takes the botanical garden of Central South University of Forestry and Technology, Hunan Province as the study area. Five imaging hyperspectral images taken by the SOC710 imaging spectrometer in October and November 2016 were used as data sources. Four samples were selected from each image (500 sample points of Eucalyptus and 500 sample points of Pinus massonians in each sample). There were a total of 15 experimental groups and 5 validation groups. It calculated K-L divergence, Hellinger distance, Bhattacharyya distance, J-M distance and Fisher's classification accuracy between two tree species in each sample, and calculated the Pearson correlation coefficient of each distance result and Fisher's classification accuracy to determine the relationship between 4 distances and Fisher's. On this basis, classification accuracy model of Fisher was established using the sample with the highest correlation in the experimental group. The research conclusions are as follows: In the same sample, Fisher classification and four distance measurement methods are used to calculate the separability. The correlation between J-M distance and Fisher's classification accuracy is the highest, and the correlation coefficient reaches 0.96. There is little correlation between K-L divergence and Fisher's classification accuracy. In this paper, the estimation model of Fisher's classification accuracy is obtained. The fitting effect of the model is good, and the average value of mean absolute deviation in verification groups is 0.083. Research has important significance for the optimization of the distance measurement method and the Fisher classification.