Lacunarity-Based Characterization of Hyperspectral Spatial Patterns

This study explores image heterogeneity within the Salinas and Pavia University Datasets using lacunarity analysis. We identify the “maximum information band” for each dataset based on standard deviation and analyze texture properties across different spatial scales within selected Regions of Interest (ROIs). Lacunarity patterns reveal transitions and saturation points, emphasizing the importance of scale selection for accurate texture characterization. Varying box sizes in Salinas ROIs affect lacunarity differently, with subtle variations in some and clearer patterns in others. In PaviaU, increasing box size initially increases lacunarity, then decreases, underscoring the need for optimal scale selection. The observed trends in lacunarity across ROIs highlight the complexity of texture structures, reflecting shifts in complexity and homogeneity at different scales. This study contributes to a deeper understanding of spatial heterogeneity and texture complexity in hyperspectral imagery, aiding in the identification of agricultural and urban features across various scales.