Rapid detection of salmon louse larvae in seawater based on machine learning

• Published in: Aquaculture
• Main Authors: Zhang, Chao, Bracke, Marc, da Silva Torres, Ricardo, Gansel, Lars Christian
• Format: Article in Journal/Newspaper
• Language: English
• Published: 2024
• Subjects: Aquaculture; Atlantic Salmon farming; Computer vision; Machine learning; Salmon lice; Atlantic salmon
• Online Access: https://research.wur.nl/en/publications/rapid-detection-of-salmon-louse-larvae-in-seawater-based-on-machi; https://doi.org/10.1016/j.aquaculture.2024.741252

The proliferation of salmon lice is one of the most challenging problems in salmon farming, seriously affecting the welfare of both farmed and wild salmonids. The detection of salmon lice in seawater, especially in the (pre-) infectious stage, supports the accurate management of salmon lice proliferation. Methods for detecting salmon lice in seawater to date involve traditional light and fluorescence microscopy and Polymerase Chain Reaction (PCR) methods. However, these methods require sample pre-processing and are laborious and time-consuming, rendering them difficult to apply for the task of continuous, rapid salmon lice detection in large water volumes. This study aims to develop a rapid and easy-to-use method for the detection of salmon lice in seawater. A novel approach was introduced for the detection of salmon lice at different larval stages in seawater by leveraging machine learning. A setup for the rapid acquisition of images of salmon lice in seawater was developed, and a dataset comprising 8316 images encompassing salmon lice “Nauplius Live,” “Nauplius Shell” and “Copepodite” was created. Then 16 object detection models including one- and two-stage Convolutional Neural Network (CNN-) based models and Transformer-based models were trained, tested, and compared on this comprehensive dataset. RT-DETR showed the best performance among all models tested with a precision of 0.972 and a recall of 0.978. The YOLO series models, except for YOLOv3, showcased remarkable inference speeds (ranging from 152 to 416 frames per second) while maintaining high precision (ranging from 0.968 to 0.977) and recall (ranging from 0.964 to 0.970). To our knowledge, this is the first time machine learning has been applied to the task of rapidly detecting salmon louse larvae at different stages in seawater samples. The results demonstrate that the machine learning approaches can achieve rapid easy-to-use and cost-effective detection of salmon lice in seawater. In the future, this study has the potential for application in ...