Automatic assessment of land parcel identification systems for agricultural management

For management and control of agricultural and environmental resources, remote sensing images are often interactively analyzed by domain experts for knowledge exploitation. To automate the image analysis for agricultural management, particularly for assessment of land parcel identification systems (LPIS), we propose an unsupervised two-step method. The first step is based on land cover identification by self-organizing maps based spectral clustering, recently proposed in [1], which combines advantageous properties of self-organizing maps (faithful quantization in a topology preserving manner) and of spectral clustering (accurate partitioning of clusters with varying statistics). The second step utilizes spatial information to extract artificial surfaces from high-resolution (5m) imageries using an anisotropic rotation-invariant textural measure, Pantex [2]. Our proposed method accurately determines the necessary required updates in the LPIS, as shown on three test zones with Rapideye imageries.