Improving land use land cover mapping of a neural network with three optimizers of multi-verse optimizer, genetic algorithm, and derivative-free function

Abstract

For land management and planning, information on the Land Use Land Cover (LULC) is vital. In this research, three optimizers of the Multi-Verse Optimizer (MVO), Genetic Algorithm (GA), and Derivative-free Function (DF) are developed in MATLAB programming language to improve the accuracy of remote sensing image classification using a Small-sized Neural Network (SNN). The results are compared to a Medium-sized Neural Network (MNN) developed in MATLAB programming language. Based on the test data, the MNN has the best performance with the Overall Accuracy (OA) of 92.64% for the object-based Landsat-8 imagery with a spatial resolution of 15 m. Based on the test data, the Derivative-free Function Multi-layer Perceptron (DFMLP) for the pixel-based Landsat-8 imagery with a spatial resolution of 15 m has the best performance with the OA of 89.31%. The Genetic Algorithm Multi-layer Perceptron (GAMLP) for the pixel-based Landsat-8 imagery with a spatial resolution of 30 m has the least performance with a value of 74.47% for the OA. The most significant improvement was for the pixel-based Landsat-8 imagery with a spatial resolution of 15 m where the DF and GA optimizers have improved the results of the SNN classifier with 7.37% and ~8% for the OA index, respectively.