Artificial intelligence based fish classification on conveyor belt

Today, the development of artificial intelligence technologies offers solutions to various problems in the fishing industry, as in many other sectors. In the fishing industry, manually classifying different types of fish caught at the same time is both time-consuming and prone to error. These manual classification processes result in the incorrect identification of fish species and, consequently, a decrease in efficiency. This situation creates cost increases and quality control issues, highlighting the need to automate the classification process. In this study, images of fish moving on a conveyor belt are captured, and artificial intelligence-based object detection and classification processes are performed using the data obtained from these images. A total of 3,915 images of three different fish species are collected to create a special dataset. The latest versions of the YOLO algorithm, which demonstrates strong performance in image processing and object recognition, namely YOLOv8, YOLOv10, and YOLOv11, are used in this study. After the training, the average mAP50-95 performance values of the models are obtained as 77.7% for YOLOv8, 78.9% for YOLOv10, and 79.4% for YOLOv11, respectively. The results demonstrate that YOLO algorithms offer high accuracy and speed in the automatic classification of fish species. With the implementation of this system, the need for human intervention decreases, and the classification process becomes more consistent and economical.