Validation of LiDAR point clouds for classification of high-value crops using geometric-and reflectance-based extraction algorithm

Abstract

This study provides an experimental analysis of high-value crop classification using geometric and reflectance-based feature extraction algorithms which can be used to validate extensive classification. The main goal of this study is to identify the class based on the 3D reconstruction of small scale LiDAR scanner and RGB images. The reference used in this study was obtained from the initially classified classes of Mapua-Phil LiDAR2 Project. In validating high-value crops, the classification results were compared to the results obtained using the reference data. The validation procedures involve comparing the locations of previously classified crops and analyzing the crop cycle in the tested fields. The proposed methodology used to identify specific class based on the geometric feature alone are found to be acceptable with an average accuracy of 92.5% while the reflectance based classification alone provides an average accuracy of 90%.