
Machine vision is transforming aquaculture, offering an innovative and non-invasive way to monitor the welfare of fish and other aquatic species. A recent study published in Aquaculture, Fish and Fisheries highlights its potential to enhance operational efficiency and improve animal welfare. However, significant challenges remain before the technology can be fully integrated into large-scale fish farming.
One of the most advanced applications is size and biomass estimation, which is crucial for managing aquaculture production. Systems such as ReelBiomass and BiomassPro have demonstrated impressive accuracy, with error rates below 5%, making them widely used in salmon farming. Yet, their high cost and maintenance demands pose barriers, particularly for smaller farms that operate with tighter budgets.
Feeding activity monitoring is another development approaching commercial adoption. Machine vision can track fish behaviour during feeding, helping farmers optimise feed distribution while minimising waste. Existing feeding cameras, already in use in many farms, can be upgraded with AI-driven algorithms to provide real-time insights. Trials have shown that lightweight AI systems such as YOLO-Tiny can detect subtle changes in feeding patterns with high precision, making them an affordable solution for farms looking to improve efficiency without major infrastructure changes.
The monitoring of fish behaviour is another promising area. Experimental systems have successfully identified early signs of stress and disease by analysing swimming patterns and schooling behaviour. However, the transition to commercial environments presents challenges. In controlled laboratory settings, visibility is high and fish populations are sparse, enabling algorithms to perform well. In contrast, in large sea cages, where fish are densely packed and water clarity is often poor, these systems struggle to deliver reliable results. The study notes that high-density environments cause overlapping images, a problem that current algorithms are not yet fully equipped to handle.
The research also highlights gaps in applying machine vision beyond fish. Despite crustaceans being a major part of global aquaculture, they remain largely overlooked in research. With growing regulatory pressure to improve crustacean welfare, addressing this gap is becoming increasingly important. Furthermore, the lack of standardised datasets from commercial fish farms has slowed progress. Publicly available databases, such as Fish4Knowledge, have been useful for research, but they often rely on images of wild fish, making them less effective for training AI models intended for farmed species.
Cost and technical barriers are also significant. While stereoscopic cameras provide high accuracy, their expense and processing power requirements make them inaccessible for many small and medium-sized farms. More affordable single-camera systems offer an alternative, but often at the expense of precision, making widespread adoption challenging.
Despite these obstacles, researchers remain optimistic about the future of machine vision in aquaculture. They recommend enhancing AI algorithms to cope with real-world conditions, integrating technology into existing systems such as feeding cameras, and developing multifunctional platforms that can simultaneously monitor size, health, and behaviour.
Machine vision is no longer a futuristic concept—it is an emerging tool that could revolutionise aquaculture. However, its large-scale adoption will depend on overcoming technical and financial barriers, ensuring it becomes an accessible and effective solution for the industry. By refining these systems, aquaculture could move towards a more sustainable, efficient, and ethically responsible future, aligning with increasing consumer expectations for higher welfare standards in seafood production.