Grid Level Analysis of the Performance of Artificial Neural Network Classifier on the Classification of Multispectral RS Data: A Case Study

Remote sensing data has found its use in a variety of social applications such as change detection, mapping, vegetation analysis, and land cover detection. The classification outcomes vary depending on the algorithms employed, the type of the classifier, and the data complexity. In this paper, we use an artificial neural network (ANN) classifier for mapping two Landsat-8 images of different complexity. The data are analyzed for LULC class separation in spatial space using Jeffries-Matusita and transformed divergence metrics. Further, we analyze the performance of ANN by studying the impact of three network generalization parameters; the number of hidden layers, number of training iterations, and training rate on the classification outputs. Accuracy assessment is carried out to verify the correctness of the output thematic maps using 1000 ground truth points. The study presents a detailed analysis of the classification performance by considering Anderson’s level 1 and level 2 classes. Lastly, we compare the classification performance of ANN with other well-known classifiers commonly employed in remote sensing.