Monday 03 March 2025
The quest for more efficient and accurate fish monitoring in fisheries has led researchers to explore the capabilities of computer vision and deep learning algorithms. A recent study published in ICES Journal of Marine Science presents a novel approach, known as AutoFish, which leverages these technologies to automate the identification and measurement of fish species.
The authors of the study aimed to develop an automated system that can accurately recognize fish species, estimate their length, and detect any instances of discarding or misreporting. To achieve this, they designed and trained a deep neural network (DNN) using a dataset comprising 1,500 images of 454 different fish species.
The AutoFish system is comprised of two primary components: instance segmentation and length estimation. The former involves identifying individual fish within an image, while the latter calculates the length of each recognized fish. To facilitate this process, the researchers employed a convolutional neural network (CNN) with a custom-designed backbone architecture, specifically tailored for the task at hand.
The AutoFish system was evaluated using three different scenarios: separated images of single fish, touching fish, and combined images featuring multiple fish. Results indicate that the system achieved an impressive mean average precision (mAP) of 0.89 in the first scenario, 0.84 in the second, and 0.79 in the third.
When it comes to length estimation, the study reveals that the AutoFish system can accurately measure fish lengths with a mean absolute error (MAE) of approximately 0.62 centimeters. Notably, this performance is comparable to human measurement accuracy.
The authors also explored the potential benefits of maintaining individual fish IDs throughout the monitoring process. By tracking and re-identifying fish across multiple images, they found that the system’s accuracy can be further improved. This capability could potentially enable more precise estimates of fish biomass and population dynamics.
While AutoFish is still a nascent technology, its promise lies in its ability to streamline fisheries monitoring, reduce labor costs, and increase data accuracy. As the fishing industry continues to grapple with issues such as overfishing and misreporting, innovative solutions like this could play a crucial role in promoting more sustainable practices.
The AutoFish system is not without its limitations, however. The study highlights the need for further research into addressing challenges like occlusion, class imbalance, and variability in fish poses. Nevertheless, the results thus far are encouraging, and it will be interesting to see how this technology evolves in the coming years.
Cite this article: “Automating Fish Monitoring with Computer Vision and Deep Learning”, The Science Archive, 2025.
Fisheries Monitoring, Computer Vision, Deep Learning, Fish Identification, Length Estimation, Instance Segmentation, Convolutional Neural Network, Fisheries Management, Sustainable Practices, Automated System.
