An Angler‐Friendly AI Pipeline for Self‐Reporting and Automatic Catch Analysis in Recreational Fisheries

Abstract

Monitoring recreational fisheries is difficult: anglers are widely dispersed, gear and practices vary, and many species are involved, which leads to fragmented and scarce data. To address these issues, we developed an Artificial Intelligence (AI) pipeline that turns angler‐reported photos into standardised records of catch composition and individual body lengths. The workflow consists of four steps: (i) automatic fish detection, (ii) pixel‐accurate segmentation, (iii) species classification trained with few labelled images and (iv) length estimation calibrated with a measurement board carrying fiducial markers (machine‐readable reference tags). We validated the system on smartphone images from a mixed‐species fishery in the western Mediterranean under realistic conditions (variable lighting, occlusions, mixed catches). The detection–segmentation stage achieved F1≈0.93. The classifier reached 85% accuracy across 38 species using only 12 training images per species, showing strong data efficiency. Length estimates were robust, with centimetre‐level error suitable for size‐class analyses. Leveraging large, pre‐trained foundation models, only a lightweight adapter requires training, keeping data and compute demands low and enabling rapid extension across regions, gear types and species lists. When integrated with existing catch‐reporting smartphone applications and agency workflows, the pipeline delivers instant angler feedback and streams standardised, analysis‐ready records to managers. In practice, this unlocks operational monitoring: near‐real‐time indicators of catch composition and length structure, automated size‐limit compliance checks and cost‐efficient inputs to stock assessment and adaptive management.

Bibliographic Information