Automated Classification of Snow Crab Shell Condition Using Analysis of Field Images

The snow crab (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Chionoecetes opilio</i>) fishery is a significant economic driver in Atlantic Canada. Effective management of this resource requires accurate identification of crabs that have recently molted, which is currently performed through subjective visual assessment of the crab’s shell condition. However, there are challenges in terms of accuracy and consistency in shell conditions identification, due to differences among skill levels of field technicians and at-sea observers. This study introduces an automated method for shell condition classification using field images of snow crabs as inputs. The images were gathered from both fishery and scientific survey operations in the southern Gulf of Saint Lawrence in 2023 and 2024. Both the Support Vector Machine (SVM) and Multilayer Perceptron (MLP) algorithms were able to reproduce reliable field identifications of five shell conditions (New-soft, New-hard, Intermediate, Old, and Very-old) to an accuracy of 87.6% and 91.7%, respectively. When the classification was simplified to two categories, “New-shells” and “Old-shells”, these accuracies improved to 96.0% and 97.9%, respectively.