Development of an optical catch monitoring tool for demersal trawl fisheries

Demersal trawl fisheries are ones of the most productive and widely used types of fishing activities globally and one of the most important sources of marine protein. This type of fisheries has always been influenced by fluctuations in stock sizes as well as continuously changing regulations of fishing activities. In recent years, demersal trawl fisheries and commercial fisheries in general, have been increasingly challenged by several factors. For instance, new ambitious management plans to stimulate sustainable development of the sector. The major issue is a considerable amount of unwanted species and sizes obtained by the demersal trawls together with the target catch, which challenge the biological and economic sustainability of these fisheries. To address this issue, ambitious management plans such as the EU Common Fisheries Policy’s landing obligation have been implemented in recent years, requiring fishers to declare all catches of listed species and count them against their quota. These plans are combined with technical regulations that aim to improve gears’ size and species selectivity through mesh size regulations, trawl modifications, and bycatch reduction devices. As the landing obligation directly couples the fisher's ability to fish selectively and the vessel's economy, even minor levels of bycatch will lead to additional expenses. There is, therefore, a need to develop methods and technologies that can transform demersal trawl fisheries from a blind process to a more informed process during which the fisher can continuously make decisions to optimize the ongoing catch process.

This thesis develops, tests and demonstrates a real-time optical catch monitoring tool developed specifically for demersal trawl fisheries that consists of an in-trawl image acquisition system to ensure high-resolution catch recordings during trawling and a software to automatically detect, classify and count the catch. The PhD thesis consists of a synopsis and three scientific articles published in open access journals. In an experimental pilot study, the thesis designs and tests a prototype of an in-trawl image acquisition system, and explores computer vision methods for automated catch description (Paper I). Since no in-trawl video data of catch was available at the outset, training a data-driven model to perform detection, classification and counting tasks was not possible initially. Instead, a model based on a classic computer vision feature extraction, and a data-driven model pre-trained on a large open-source dataset were first considered.

This pilot study provided the basis for two further developments of the optical monitoring tool. The first one covered the successful transfer of an experimental image acquisition system to real demersal trawling conditions, ensuring stable and consistent background and illumination during image acquisition (Paper II). The second one included software development for automated catch description (Paper III), based on a deep-learning approach, which became possible to implement due to the collected catch data during in-trawl image acquisition system test at sea.