Advancing Fisheries Science: Enhancing Data Collection and Analysis through Electronic Monitoring

The global demand for seafood continues to rise, placing immense pressure on marine ecosystems and intensifying the challenges faced by fisheries management. Accurate catch data is pivotal for assessing stock health, guiding sustainable harvest practices, and ensuring ecosystem resilience. However, traditional methods of catch documentation are often inadequate due to unreported and misreported catches. This Ph.D. thesis investigates the potential of electronic monitoring (EM) combined with computer vision and deep learning technologies to address these challenges and advance fisheries science.

The research is framed around four key objectives: improving species identification and length measurement accuracy, developing ground-truth datasets for training machine learning models, assessing the performance of a readily deployable deep learning architecture in complex catch scenarios with occlusion, and exploring practical applications of EM data for fisheries management. Through a series of interconnected studies, this thesis collects, develops, analyses, and evaluates the use of EM data for species classification and length estimation of catches. Whereas several of the proposed methods are applicable across different types of fisheries, the present research focuses exclusively on the application of demersal fishing vessels with conveyor belts. The findings demonstrate that while deep learning offers transformative potential for automating catch documentation, challenges such as occlusion, variation in vessel layouts, and the scarcity of labeled training data remain significant barriers.

A critical contribution of this thesis is the development of a benchmark dataset for video instance segmentation suitable for a multitude of fisheries applications. This dataset is the first of its kind in fisheries. The data collection method used is adaptable to other fisheries, and the authors highly encourage increased efforts in the fisheries community to keep collecting and sharing such datasets to further advance fisheries science.

The findings in this thesis underscore the necessity of integrating EM systems into management frameworks, not only to enhance data accuracy but also to support adaptive management strategies and promote transparency in global fisheries. Enhanced data collection and analysis capabilities can enable more precise stock assessments, facilitate scalability of gear selectivity studies, and support the transition toward more flexible and catch-based management frameworks.