Abstract

In this presentation, we will explore the various tasks undertaken by the Fish Ecology Lab at the Palma Fish Market since 2018. Our focus lies in automating the processes employed for extracting fish length measurements. The dynamic distribution of fish lengths plays a crucial role, not only in making informed short-term operational decisions within a fishery co-management framework but also in providing input for traditional fishery models that illuminate mid- and long-term trends in exploited stocks. Traditionally, the estimation of fish length in most fisheries has been a manual endeavor, resulting in precise measurements at the individual fish level. However, due to the high cost and inefficiency associated with supervised sampling, the sample size tends to be limited. Consequently, the precision of population-level estimates often falls short, and biases may arise, especially when adequately stratified sampling programs are economically unfeasible. Conversely, the application of machine learning and artificial intelligence in fisheries science presents a promising avenue for large-scale, unbiased sampling of fish catches. The Fish Ecology Lab has been working with the daily images recived from the auction of the fish market to extract information on various commercial species such as hake, dolphinfish, and red shrimp. This presentation will specifically focus into the deep learning techniques employed for each of these cases.