CLASSIFICATION OF DATA FROM AIRBORNE LIDAR BATHYMETRY WITH RANDOM FOREST ALGORITHM BASED ON DIFFERENT FEATURE VECTORS
Modern full-waveform laser bathymetric scanners offer the possibility of a practical application of airborne laser bathymetry (ALB) data algorithms as a valuable source of information in the study of the aquatic environment. The reliability of the obtained results and the efficiency of the classification depend on the applied features. The input data for the classifier should consist of variables that have the ability to discriminate within the data set, for the detection and classification of objects on the seabed. The automatic detection of underwater objects is based on machine learning solutions. In this paper, the ALB data were used to present a classification process based on the random forest algorithm. The classification was carried out using two independent approaches with two feature vectors. The quality of classifications based on the full-waveform features vector and the geometric features vector was compared. The efficiency of each classification was verified using a confusion matrix. The obtained efficiency of the point classification in both cases was about 100% for the water surface, 99.9% for the seabed and about 60% for underwater objects. Better results for the classification of objects were obtained for the features vector based on features obtained directly from full-waveform data than for the vector obtained from geometric relationships in the point cloud.