A Novel Classifier Ensample for Combining Pixel-Based and Object Based Classification Methods for Improving Feature Extraction from LIDAR Intensity Data and LIDAR Derived Layers

摘要

Information extraction from LIDAR data is a hot research topic. Airborne LiDAR (Light Intensity Detection and Ranging) provides three different kinds of data: elevation, 3D point clouds, and intensity. This study evaluated the use of the LIDAR intensity data and LIDAR derived layers for land-cover classification. Two classification approaches were tested and their results were compared. The two approaches are pixel-based and object-based classification approaches. First, the pixel-based classification approach presented by the maximum likelihood classification (MLC) technique was used to classify the LiDAR intensity data. Then, more bands such as DSM, texture of the intensity data, and terrain slope were added, as different bands, to the intensity data to improve the classification accuracy resulted into six approaches. Secondly object-based classification (OBIA) was performed. An overall accuracy of 65.3% was achieved using the sixth approach of pixel-based classification technique. The overall accuracy of the results is improved to 69.5% using the object-based classification technique. Finally, classifier combination or classifier ensemble was developed for improving the classification results. The combined approach achieved the highest accuracy reaching 75.32% and kappa index of agreement of 0.79 and improving accuracy of individual classes.