Real-Time Tuna Fish Quality Assessment Using Deep Learning Techniques

Fish quality evaluation is critical to guarantee food safety, facilitate international seafood trade, and reduce post-harvest losses. Conventional quality assessment relies on subjective expert visual inspection, which is labour-intensive and error-prone. Recent advances in computer vision and deep learning offer promising alternatives for automated, objective grading. This paper investigates two deep learning image classification approaches—a Roboflow ResNet18-based model and a YOLOv11 classifier—applied to a dataset of 257 tuna fish images labelled as Grade A, Grade B, or Grade C. Images were pre-processed with XnConvert and Roboflow, and augmented to improve model generalisation. Performance was evaluated using accuracy, precision, recall, F1-score, and confusion-matrix analysis. The Roboflow model attained a validation accuracy of 78.90% and test accuracy of 64.86%, while YOLOv11 achieved a validation accuracy of 78.95%, precision 74.96%, recall 70.87%, and F1-score 71.19%. Grade C tuna was classified most reliably, whereas Grade A proved most challenging owing to inter-class visual similarity and limited training samples. Results confirm the feasibility of computer-vision based automated tuna quality grading, with YOLOv11 identified as the preferred model for real-world deployment.