A method for predicting the cutting points using random sample consensus partitioning technique and AI machine vision.

Abstract

Food processing and manufacturing companies that handle large amounts of fish want their fish to be packaged and delivered in uniform quantities. In particular, manual labor is performed by skilled workers for fish that are difficult to trim. Moreover, some fish processing sites employ partially automated processing machines. However, the structure of these machines is too simple to process fish with diverse patterns. Furthermore, these machines exhibit a large amount of error for the target weight. To improve the working environment of food processing and manufacturing companies by solving the aforementioned problems, this study proposes a technique for predicting the cutting points of a fish for each weight by adopting image processing and machine learning. The proposed technique employs a variety of image processing techniques and random sample consensus partitioning to extract the 3D model of the fish and its length, maximum major and minor axes, and volume information from the image of the fish. The model trained with the extracted 3D features and the measured weight information can predict the cutting points for the desired weight from the input fish image. The performance evaluation results of the proposed method indicated that there is an average error of less than 3% between the target and predicted weights. This error level is considered significantly better than 8%, which is the permissible error level in fish processing sites. It is expected that the proposed technique will significantly contribute to the development of an automated cutting system that considers the weight by integrating the technique with the cutting machine and conveyor belt.