Abstract
The vast majority of goat skin processed by traditional tanneries comes from small rural producers. Thus, with the predominance of rustic creation, slaughter, and skinning methods, the batches of hides processed by tanneries have a very heterogeneous quality. Thus, there is a need to categorize the samples according to the quantity and location of defects. The categorization process is subjective and strongly influenced by the experience of the professional classifier, causing a lack of homogeneity in the composition of the goat hide lots for sale. Aiming to reduce failures in the categorization of goatskin samples, the authors investigate the application of computer vision and artificial intelligence on a set of previously categorized wet blue goatskin photographic samples. That said, is analyzed the capacity of different classifiers, with different paradigms, in detecting defects in goatskin samples and in categorizing these samples among seven possible quality levels. A hit rate of 95.9% was achieved in detecting defects and 93.3% in categorizing quality levels. The results suggest that the proposed methodology can be used as a decision aid tool in the qualification process of goat leather samples, which can reduce sample labeling errors.