Automated Length Measurement of Pangasius Fingerlings Using a YOLO-Assisted Vision System on Raspberry Pi

Abstract

Regular measurement of fish length is essential for monitoring growth performance and evaluating the health status of fish in aquaculture. Accurate length measurements provide important information for optimizing feed management and facilitating size-based grading and sorting processes. However, conventional manual measurement methods are time-consuming and may cause stress or physical injury due to repeated handling. This paper presents an automated length measurement method for Pangasius fingerlings by integrating the YOLO26n-Seg-NCNN instance segmentation model with OpenCV-based image processing techniques. The proposed approach improves the robustness of object recognition and mask extraction under varying illumination conditions, thereby enhancing the reliability of fish length estimation. Experimental results show that, compared with the conventional OpenCV-based method, the YOLO-OpenCV approach reduced the maximum absolute error from 12.12 mm to 2.71 mm, corresponding to an improvement of 9.41 mm, while increasing the minimum measurement accuracy from 90.06% to 97.41%. Furthermore, a lightweight edge-AI framework is developed by optimizing the deep learning-based image processing pipeline for deployment on resource-constrained devices, enabling automated monitoring and management of Pangasius aquaculture production.