土地督察易混地物最大似然分类改进

Comprehensive Improvement of Maximum Likelihood Classification for Easily-mixed Land Features in Land Supervision

鉴于中高分辨率影像最大似然分类对一些类型的建筑和裸土容易发生相互错分(易混地物),直接影响土地督察应用高分一号多光谱数据对大型工程项目的发现和阶段认定,提出了一种基于最大似然分类综合优化的精度提升方法。通过样本过滤和正态化提升样本的典型性,同时获得地物的先验概率,然后进行最大似然预分类,再通过混淆矩阵依据先验概率分布和像元的地物识别概率对像元归属进行调整,进一步提高分类精度。通过对比实验验证,综合改进后建筑和裸土的用户精度均有11%以上的提升,总体精度提升9%以上。

Building and filling have a high probability of mutual misclassification in the largest likelihood classification(MLC)of GF-1image,and they are easily mixed ground objects,which directly affect the judgment of construction status.This paper proposes a comprehensive optimization based on MLC accuracy improvement method.Through the sample filter and normal promotion sample representativeness,firstly,it obtains the feature prior probability.Secondly,it applies maximum likelihood classification.And then, through the confusion matrix based on the prior probability distribution and the like yuan feature recognition probability adjust like yuan belonging,it further improves the classification accuracy.By comparison experiments,the user's accuracy of building and filling was improved by more than 11%respectively,and the overall accuracy was increased by more than 9%.