Ground surface structure classification using UAV remote sensing images and machine learning algorithms

The applicability of a machine learning algorithm can vary across regions due to disparities in image data sources, preprocessing techniques, and model training. To enhance the classification accuracy of ground surface structures, it is crucial to select an appropriate method tailored to the specific region. This study used highly-efficient UAV remote sensing photography and conducted training and tests using three supervised machine learning techniques, namely support vector machine (SVM), random forest (RF), and maximum likelihood (ML) as well as performed a cluster analysis using an unsupervised machine learning technique. The main objective of this study was to evaluate the effectiveness of four machine learning methods for classifying five distinct structures (forest, grassland, bare land, built-up area, and road) in UAV images. The machine learning methods will be trained using sample features extracted from the UAV images, and test classifications will be conducted for the five ground surface structures. The results demonstrated that the RF classifier outperformed the other methods, achieving performance metrics, including an accuracy of 91.78%, an area under the curve (AUC) of 0.93, a Kappa coefficient of 0.88, and a gain of 100%. The RF classifier showcased its capability to accurately differentiate between various ground surface structures by examining spectral composition, encompassing both natural and artificial elements, and making precise judgments based on factors such as color, color tone, and texture observed in the images.