AI Empowers Offshore Processing: FIRI Develops Non-Contact Intelligent Weighing and Grading System for Squid Tentacles at Sea
Squid is a core commercially valuable species for China’s distant-water fishery. As a high-value processed product, the weight of squid tentacles directly determines their grade and transaction price. However, the offshore processing environment on fishing vessels is challenging. For a long time, weighing and grading squid tentacles has relied heavily on manual work. Constant vessel movement at sea makes manual weighing labor-intensive, low in efficiency and unstable in accuracy. Conventional mechanical weighing equipment is also susceptible to vibration from the ship, which has become a key bottleneck restricting the productivity of offshore squid processing.
To adapt to the harsh working conditions onboard, the research team for aquatic product processing equipment from the Fishery Machinery and Instrument Research Institute (FIRI), Chinese Academy of Fishery Sciences, has developed a new technology for non-contact, multi-target simultaneous weighing of squid tentacles. The related paper A deep learning-based model for accurate weight estimation of squid tentacles has been published in Measurement, a JCR Q1 journal with an impact factor of 6.1. It offers a brand-new technical route for intelligent processing of aquatic products on distant-water fishing vessels.
The research took Argentine shortfin squid as the test subject. Built on the YOLOv8s-seg algorithm, an image segmentation model named SQ-YOLO was developed to identify overlapping squid tentacles. A complete detection pipeline was established: image segmentation → morphological parameter extraction → weight prediction. This solution does not require 3D scanning hardware. Weight estimation can be realized merely from 2D images, achieving industry-leading performance. It effectively avoids disturbances to mechanical weighing caused by vessel vibration and enables simultaneous measurement of multiple stacked squid tentacles, breaking the efficiency limit of traditional one-by-one weighing.
The algorithm framework is highly transferable. With recalibration of morphological parameters, it can be adapted for online inspection and grading of other aquatic products such as octopus, sea cucumber and fish fillets, showing great potential for industrial application.
The project is still under technical optimization. The research team plans to expand the training dataset to cover more real marine working scenarios. Combined with pre-treatment methods such as vibratory conveying to reduce stacking of raw materials, the weighing accuracy will be further improved to accelerate the transformation of laboratory technology into practical production lines. This research was jointly funded by the National Key R&D Program and research grants at academy and institute levels.
Industry Insights
Digital transformation of equipment for distant-water fisheries has long been a tough part of industrial upgrading. High humidity, constant shaking and limited space at sea prevent many mature shore-based devices from being deployed directly on vessels.
This non-contact weighing solution based on machine vision replaces physical weighing sensors with AI vision, circumventing the inherent drawbacks of mechanical onboard equipment. If further sea trials succeed, it will not only solve pain points in squid processing but also inspire new approaches for automatic grading of various other offshore catches.
Summary: The SQ-YOLO vision model addresses the challenge of weighing stacked squid tentacles at sea. It requires minimal hardware, supports multi-species adaptation, and is expected to move from academic research to production lines on fishing vessels.
Comments