Class Exclusion and Reassembling: A Novel Approach for Improving Classification Accuracy in Remote Sensing Imagery

Abstract This paper introduces a new approach called Class Exclusion and Reassembling, which outperforms traditional band arithmetic approaches in terms of classification accuracy. The study was conducted in Nanded district and Parbhani district, India, with the goal of improving the accuracy of classification for all classes. Traditional band arithmetic approaches often improve the classification of some classes, but they can reduce the accuracy of other classes and the overall classification accuracy remains insufficient. In the proposed method, classes are classified one by one using band arithmetic approaches. After all classes are classified, the classified pixels are reassembled to obtain the complete classified image. The proposed method was tested on two sites: one with training and testing data, and another with only testing data. The results were compared with the composite image of all 13 bands using the random forest classification tool. The proposed method achieved 100% accuracy for Site 1, which was an improvement from the previous accuracy of 86%, and 95.5% accuracy for Site 2, an improvement from 89.5%. Overall, the Class Exclusion and Reassembling approach is more effective than traditional band arithmetic approaches for classification accuracy.