Urban Land Use Classification Using LiDAR Geometric, Spatial Autocorrelation and Lacunarity Features Combined with Postclassification Processing Method

Abstract

The economic and technological development zone (ETDZ) is a critical urban economic and functional area. Inefficient land exploitation and insufficient supervision have led to a great waste of land resources. The timely and precise extraction of residential and industrial building type, area, and density information is urgently needed and essential for sustainable land use development. This study attempted to extract residential and industrial buildings by integrating light detection and ranging (LiDAR) geometric, local indicators of spatial association (LISA), and lacunarity features with an object-based classification and postclassification processing approach. Geometric and LISA features were selected based on random forest cross-validation (rfcv) method. Grayscale lacunarity features were then incorporated for detailed classification using a support vector machine classifier and spatial neighbor processing method. An accuracy assessment indicated that the proposed method can effectively identify residential and industrial buildings. The overall classification accuracy and kappa statistics were 96.72% and 0.9538, respectively. A comparison with results derived from the normalized digital surface model (nDSM) alone and nDSM + multiple features (LiDAR geometric, LISA, and lacunarity) showed a significantly improved classification accuracy using kappa analysis. The results not only confirmed the applicability and effectiveness of the proposed method but also provided fundamental information for evaluating land use in the ETDZ.