EveryFish

Transición digital del seguimiento de las capturas en las pesquerías europeas

Duración: 2023-2026

Financiación: Horizont Europe

El proyecto Everyfish pretende desarrollar, testar y validar diferentes soluciones tecnológicas de monitorización para aplicando IA y machine learning poder contribuir a una mayor transparencia en las pesquerías europeas. El proyecto liderado por SINTEF Ocean desarrollará innovaciones tecnológicas que permitirán al sector asegurar un correcto reporte de capturas. La información generada aportará datos y conocimiento para el desarrollo de pesquerías más sostenibles.

La participación de DATA FISH se desarrollará en el WP6: Data Collection and Validation in European Fisheries, liderado por el centro tecnológico AZTI, en el cual participará en la toma de datos y validación para pesquerías europeas.

Labeling progress in AZTI–DataFish WP4

We are working on EVERYFISH project as a partner labeling images and developing tools to identify species of tuna catches automatically following the next steps:  

Figure 1: Frame segmentation: segmentation of all the frames in the input video.

CatchHawk: Species identification and size estimation in tropical tuna purse seiners:
  • Video input.
  • Manually segmented and labelled images to train the models. 
  • Segment distribution in classes (segment tracking and segment classification).
  • Application of data segmentation techniques.
  • Histogram equalization.
    Segment tracking and classification: Tracking of each segment through the video and prediction of its class:
    • An ID is assigned to each different fish that is identified. 
    • The segments for a given ID present in the different frames are classified. 
    • A final class is predicted for each ID.
      Size estimation:
      • It is difficult to estimate the size of a fish without knowing the distance between itself and the camera. 
      • We will try to use 3D data from stereoscopic cameras to estimate their size considering the depth information.

        Conclusions and next steps:

        • Some species like BET and YFT are very difficult to classify. 
        • Different strategies are being tested to improve the classification.  
        • Data for validation will be available soon.

          Figure 2: Id:1, BET:0.98, SKJ:0.05, YFT:0.07; Id:2, BET:0.62, SKJ:0.01, YFT:0.37 

          Figure 3: Infrared camera use testing.

          Socios