Automatic inspection of baked goods based on cost-effective RGB-D camera

Nowadays, growing attention is paid by consumers to agricultural and food products that meet high-quality requirements. Considering the food production activities, therefore, many efforts are aimed at developing new methods for verifying in an automatic way the correspondence between the qualitative parameters obtained and those expected. Indeed, these inspection activities are still often performed by operators, thus suffering some critical issues such as slowness, laboriousness, and the need, in most situations, of physical contact between the operator and the analysed product. These aspects can undermine the economy, the extensiveness and, sometimes, the success of such operations. Non-Destructive Systems (NDS), which are automatic or semi-automatic processes used for inspection, allow overcoming these limitations, providing a large amount of information about the analysed products. In this work, an innovative NDS for the inspection of baked goods is presented, which is based on a cost-effective depth camera. The system was specifically conceived to identify products with morphological defects due to unsuitable leavening or baking. Three product categories were considered in the experimental campaign: first choice, second choice, and non-compliant products. A set of 8 indices were computed by processing the 3D point cloud model provided by the system, in order to determine the optimal k-mean based classifier. The accuracy, obtained by processing 24 items, was 83%, with a defects underestimation error lower than 4% and an overestimation error of 12%. The system was proved to be reliable since it does not classify any unsuitable product as compliant with company requirements.