AI-driven classification and precision cutting algorithms using machine vision in a customer-operated fish processing system

Abstract

Despite the high nutritional value of fish, it is often under-consumed due to its characteristic odor and laborious cleaning process. This sensory barrier significantly diminishes the appeal of fish, particularly in regions or cultures where individual exhibit heightened sensitivity to fish odor. Fish processing systems have been developed to facilitate cutting and cleaning steps in aquatic supply centers and factories. In this study, to upgrade a fish processing system to an intelligent machine, four high-consumption fish classes were classified using Artificial Intelligence (AI), and the corresponding cutting point determination algorithms were developed using a multipurpose backlighted pure blue background for each class. As the classification algorithms developed, the best results were selected based on the least total MSE value. The best ANN structure was determined as 6–23–4 with 99.62%, 96.72%, and 95.06% with corresponding MSE values of 9.51 × 10^–5, 2.03 × 10^–2, and 2.54 × 10^–2 in the train, validation, and test sets, respectively. This structure was recorded as the best one with the ‘Logsig’ function in both hidden and output layers with the LM learning algorithm. The total classification accuracy of the SVM classifier resulted in 99.69% and 98.75%, with the corresponding MSE values of 1.23 × 10^–2 and 1.25 × 10^–2 in train and test data sets, respectively. As soon as the fish were classified, their unique cutting point determination algorithms were applied for fish processing. Finally, the head and belly cutting points accuracy of Silver Carp, Carp, and Trout fish were resulted in 98.36% and 99.49%, 97.85% and 98.07%, and 96.61% and 97.90%, respectively.