Pixel-Based Classification of High-Resolution Satellite Images for Crop-Shelter Coverage Recognition

Intensive agricultural production contributes to environmental changes by acting on different factors. Protected cultivation, in particular, could be considered as a driving force that exerts pressure on the environment as the economic activity connected to it leads to changes in the state of the territory. To analyze the environmental effects of crop shelters using, for instance, causal chain frameworks as well as to adopt methodologies and tools oriented to reduce the negative effects of protected cultivation, knowledge of location and planimetric extent of crop shelters as well as their spread over time are required. This knowledge can be acquired by using and analyzing suitable thematic cartography that, if available, is often difficult to update because of the temporary character of crop shelters. Experts engaged by local authorities to carry out the visual recognition of large territories could take advantage of pattern recognition methods to reduce time-consuming operations. In this regard, methodologies for automatic acquisition of crop-shelter coverage from remote-sensing images have been proposed in some studies. Knowledge of location and planimetric area of crop shelters as well as their spread in a region is much more relevant when there are no planning tools concerning the setting of crop-shelter cultivation, and no specific regulations regarding materials and building techniques. In this paper findings of experiences on the definition and implementation of image processing methodologies based on per-pixel methods for crop-shelter classification applied to high resolution multi-spectral satellite images were analyzed and discussed. The spectral information contained into these images was analyzed and mapped through indices. The results showed that improvements of crop-shelter classification could be achieved by using this information in comparison to RGB spectral information.