Land-cover and land-use classification generates categories of terrestrial features, such as water or trees, which can be used to track how land is used. This work applies classical, ensemble and neural network machine learning algorithms to a multispectral remote sensing dataset containing 405,000 28x28 pixel image patches in 4 electromagnetic frequency bands. For each algorithm, model metrics and prediction execution time were evaluated, resulting in two families of models; fast and precise. The prediction time for an 81,000-patch group of predictions wasmodels, and >5s for the precise models, and there was not a significant change in prediction time when a GPU was used. Logistic regression was the best fast model, with >91% accuracy & f1 on the holdout dataset. While this model struggled with the trees and grassland classes, it provided inferences that the green wavelength band and the perimeter of the blue wavelength were important. All precise models had better performance than prior work, and the best precise model resulted from an 8-hyperparameter simultaneous neural network optimization. This model possessed >97% accuracy & f1 on the holdout dataset, and >97% accuracy & f1 on each of the 6 land categories. For both best-of-family models, there was little overfitting. Finally, the model was validated by predicting the land categories of a sample 159 MB multispectral image and visually verifying correct predictions. Using these classification models to automatically monitor land-cover and land-use classification is a promising approach for tracking changes over time and potentially reducing analyst workload.
World Congress in Computer Science, Computer Engineering, and Applied Computing (CSCE 2022)
Turner, Lorelei [*]; Wagner, Torrey J.; Auclair, Paul; and Langhals, Brent T., "Machine Learning Land Cover and Land Use Classification of 4-band Satellite Imagery" (2022). Faculty Publications. 1407.