Lobster Position Estimation Using YOLOv7 for Potential Guidance of FANUC Robotic Arm in American Lobster Processing

Abstract

The American lobster (Homarus americanus) is the most valuable seafood on Canada’s Atlantic coast, generating over CAD 800 million in export revenue alone for New Brunswick. However, labor shortages plague the lobster industry, and lobsters must be processed quickly to maintain food safety and quality assurance standards. This paper proposes a lobster estimation orientation approach using a convolutional neural network model, with the aim of guiding the FANUC LR Mate 200 iD robotic arm for lobster manipulation. To validate this technique, four state-of-the-art object detection algorithms were evaluated on an American lobster images dataset: YOLOv7, YOLOv7-tiny, YOLOV4, and YOLOv3. In comparison to other versions, YOLOv7 demonstrated a superior performance with an F1-score of 95.2%, a mean average precision (mAP) of 95.3%, a recall rate of 95.1%, and 111 frames per second (fps). Object detection models were deployed on the NVIDIA Jetson Xavier NX, with YOLOv7-tiny achieving the highest fps rate of 25.6 on this platform. Due to its outstanding performance, YOLOv7 was selected for developing lobster orientation estimation. This approach has the potential to improve efficiency in lobster processing and address the challenges faced by the industry, including labor shortages and compliance with food safety and quality standards.