Improving segmentation accuracy for detecting deforestation using texture feature derived from landsat 8 oli multispectral imagery

Abstract

For the identification of deforestation, numerous techniques and features have been developed and used to maximize the segmentation accuracy. Choosing appropriate texture features and extraction methods are one of the most significant issues in remote sensing image applications. This work proposes a Modified Sum of Difference using diagonal texture feature extraction with multiresolution segmentation algorithm to maximize the segmentation accuracy. First, the Modified Sum of Difference using diagonal matrix (MSODDM) algorithm is applied to the input image; this divides the image into a number of coarse objects. Secondly, the multiresolution algorithm with the threshold value of 200 is applied to the coarse image produced by MSODDM for getting finer segmentation result. Finally, the accuracy of the segmented image is compared with the sample input image used in the experimental study. Thus, the proposed method generates more promising result of reaching 97.02% of segmentation accuracy, which results in better classification of remote sensing images in detecting deforestation.