Application of Image Processing and Machine Learning to the Classification of Fish Size

Abstract

The purpose of the research is to evaluate image processing and machine learning techniques for classifying fish size. In this study, the sizes of mackerel fish were classified into three categories based on their weight range: small, medium, and large. Initially, the fish were captured on top and side views using a light control box to regulate image quality. The fish images were then analysed by image processing and machine learning techniques using the MATLAB software to acquire fish characteristics. Five fish qualities were observed including fish area, fish perimeter, fish length, fish width, and the length-to-width ratio in top and side views. For the purpose of categorizing fish into separate groups, statistical analysis was used to determine the relationship between fish characteristics and size. The correlation analysis between fish attributes and size revealed that the fish area in top view exhibited the highest correlation value, following by the fish width in top view. Combining image processing and regression analysis can be used to classify fish by size. On the other hand, the decision tree algorithm was implemented to allocate fish sizes using machine learning. It was discovered that both image processing and machine learning can be used to classify fish by size. The classification accuracy of both techniques exceeded 90%, which was satisfactory. This can be further applied in automated fish-processing lines to supplant manual labours.