Surface quality is an important characteristic commonly assessed in wood products. Sanding relies on coated abrasives as tooling for both dimensioning and surface finishing, but their performance is dependant on chip loading and grit wear. Traditionally, the useful life of abrasive belts in sanding operations has been manually assessed. This type of inspection is highly dependent upon individual expertise and usually leads to either underutilization or overutilization of the abrasive, which in turn affects the production costs and quality of the product. A classification method that characterizes the abrasive loading curve with artificial neural networks and computer vision was developed. Controlled experiments were conducted to develop abrasive belts of known machining exposure. Image processing was complemented with pattern classification and recognition algorithms to support a decision-making framework. The results show 93 percent and 95 percent success rates in abrasive images classification. Also, classification of images from interpolated and extrapolated times of abrasive usage is achieved with high success rates. This approach is proposed as an input to a decision system that would help in evaluating the remaining life of the abrasive and would trigger optimal tool replacements.