CAJ | 학술논문

针对非监督分类算法分类精度不高、监督法分类算法的训练样本需要人工选择且容易误选的问题,提出了一种基于模糊C均值聚类( FCM)和BP神经网络相结合的遥感影像自动分类算法. 首先利用FCM对影像进行初始聚类,然后根据聚类结果,由该算法自动选取其中的纯净像元作为训练样本,并送入BP网络进行学习,用最终训练得到的BP神经网络分类器对TM遥感影像进行分类,实验结果表明该算法具有较高的分类精度,能够满足大尺度地物类别判定的需要.
침대비감독분류산법분류정도불고、감독법분류산법적훈련양본수요인공선택차용역오선적문제,제출료일충기우모호C균치취류( FCM)화BP신경망락상결합적요감영상자동분류산법. 수선이용FCM대영상진행초시취류,연후근거취류결과,유해산법자동선취기중적순정상원작위훈련양본,병송입BP망락진행학습,용최종훈련득도적BP신경망락분류기대TM요감영상진행분류,실험결과표명해산법구유교고적분류정도,능구만족대척도지물유별판정적수요.
As for the problems that low classification accuracy of non-supervise classification algorithm and training sample of super-vise classification algorithm needs manual selection which is easy to be made wrongly, there is an automatic classfication algorithm of remote sensing image which is based on the combination of FCM and BP neural network. First, this paper uses FCM to make initial clusters of images. Then in accordance with the results of clusters, this paper picks out the endmembers which are automatically select-ed by the algorithm as the traaning samples, sends the samples to study in BP network and uses the BP neural network classifier which is got from the final training to classify the TM remote sensing images. The result shows that the algorithm owns high accuracy which could meet the requirements of determination of object types in a large scale.