An analytical framework to predict slaughter traits from images in fish

Yuuko Xue*, John W.M. Bastiaansen, Haris Ahmad Khan, Hans Komen

*Corresponding author for this work

Research output: Contribution to journalArticleAcademicpeer-review

1 Citation (Scopus)

Abstract

Accurate measurements of breeding traits on individuals are critical in aquaculture for obtaining breeding values and tracking the progress of the breeding program. Modern breeding programs prioritize not only production traits but also complex traits related to production, product quality, body composition, disease resistance, and fish health, such as slaughter traits. Slaughter traits can be selected indirectly and incorporated into breeding programs. Indirect selection is cost-effective, but there is often little genetic correlation between measured and target traits. Accurate phenotypic prediction of the target traits using modern phenotyping technology can be game-changing in indirect selection. This paper proposes an analytical framework for predicting slaughter traits using images. The framework demonstrated that using images in addition to body weight improved fat percentage prediction accuracy from 0.4 to 0.7 when compared to a model that only used body weight and its numerical derivations. The framework also allowed for the interpretation of the prediction by providing imaginal features. In the case study, the dorsal side, the upper edge of the pectoral fin, and operculum edge were discovered to be the three regions on seabream that have properties that are negatively correlated with fillet fat percentage. The framework showed that both body weight and visceral weight are highly correlated with total fish body area. The framework also revealed that the lower edge of the pectoral fin, operculum edge, and anal fin are the regions with properties that explain variation in the visceral percentage. Future research will be required to segment and quantify each predictive imaginal feature to calculate its heritability. The framework can potentially predict other harvest, post-slaughter, and metabolic traits for aquacultural study.

Original languageEnglish
Article number739175
JournalAquaculture
Volume566
DOIs
Publication statusPublished - 15 Mar 2023

Keywords

  • Machine learning
  • Machine vision
  • Multi-input framework
  • Non-invasive prediction
  • Novel phenotyping

Fingerprint

Dive into the research topics of 'An analytical framework to predict slaughter traits from images in fish'. Together they form a unique fingerprint.

Cite this