ABSTRACT
There is a growing trend in applying machine learning and metaheuristic optimization algorithms in land cover classification with high-quality results. These optimizers search for the global optimum of objective functions, usually error-minimizing tasks. This chapter summarizes the metaheuristic method and its potential combination with neural networks (to form hybrid models) in land cover classification with a case study in Vietnam. Atom Search Optimizer and Salp Swarm Algorithm were used for verifications. The SPOT 7, with 1 Panchromatic (1,6 m) and four multiple spectral bands, was used to prepare the training dataset. The hybrid models’ workflows starts with image segmentation (more than 54,000 image objects with associated attributes were generated), data standardization, and model training using Root mean square error as the objective function. The results show a considerably high accuracy (overall accuracies around 0.98, Root mean square error around 0,2). The selection of these algorithms, among many others, does not quarantine their outstanding performance against the others but instead verifies their capability in land cover classification study. The last section discusses the remaining problems and future remarks.
