Line of action:

Status:

Execution date:

General Objective:

To consolidate the progress made with the FOTOPEIX project (PLEAMAR 2017) in terms of automatically measuring the size of fish caught by Mallorca’s artisanal fleet from images, but also to take the first steps to automatically classify the species in each box of fish landed.

Specific objectives:

• Consolidate the automatic size estimation method: (1) make it more robust and (2) more accurate.
• Significantly increase the number of images processed.
• Adapt the automatic size estimation method to other species.
• Lay the foundation for an automatic sorting of species inside a fish box.
• To disseminate the applicability and usefulness for the fishing sector of the tools developed.

FOTOPEIX II has satisfactorily met the objectives set in the initial planning, although there are aspects of the methodology that require improvements for its operational application in practically real time.

During the project, four neural networks have been put into operation for the detection of species and size estimation and sampling of fish in the fish market has been carried out for the development of these nets, as well as to validate the results. It has been proven that the MASK RCNN net is the most suitable for working in the identification and extraction of fish sizes from fish market images, since due to the arrangement of the fish in the boxes it is necessary to carry out a process of segmentation of the image for a correct separation between each piece. The correct adaptation of these networks, however, requires a great effort and dedication for the design and preparation of the data that feed them, a factor that has been underestimated and that is intended to be worked on in the near future.

The project has made it possible to generate a trained network to classify and identify two species of red mullet of high interest in the Fish Market: Mullus surmuletus and Mullus barbatus and a protocol has been created to implement the strategy to new species in the future.

A great deal of effort has also been devoted to scientific dissemination, presenting project results at three international meetings and publishing an article in the ICES Journal of Marine Science. This article describes the precision and accuracy results of the application of nets for the measurement of fish sizes.

One of the most interesting points of the development of FOTOPEIX II has been the sharing of the work carried out with another project of the Pleamar 2018 call, SICAPTOR, of the Institute of Marine Research of Vigo, due to the similarity of both projects. Telematic meetings have been held in search of synergies and a visit to share the knowledge acquired by both entities. From this contact, a proposal has emerged for the Pleamar 2019 call, in which work continues with the insertion of artificial intelligence in the field of fisheries research and management to increase the effectiveness of efforts through collaborative actions.

FOTOPEIX II has confirmed the usefulness of neural networks to work with fish databases, mainly images, highlighting the promise of deep learning techniques.

The proposed methods still need to be improved in order to be considered fully operational tools, improving the predictive capacity of the nets, for which it will be essential to significantly increase the number of patterns used in training, the number of images processed and the number of fish sampled. In short, the project has highlighted the need for greater effort in the preparation of input data to these networks and for more personnel and dedication to achieve an optimization of the results.