Snapshot hyperspectral imaging for production improvements in Atlantic salmon farming: A proof-of-concept study

• Hyperspectral imaging (HSI) evaluated for sexual maturity in Atlantic salmon. • Snapshot HSI integrated with halogen illuminators for image acquisition. • Mean spectral reflectance differed between mature and immature fish classes. • Key spectral bands for maturity detection identified in the 350–500 nm range. • Extra Trees and Random Forest classifiers achieved the best accuracy. When a significant percentage of farmed fish mature prematurely, it can negatively impact product quality, production, and profitability. Therefore, early and accurate detection of sexual maturity in recirculating aquaculture systems (RAS) raised Atlantic salmon ( Salmo salar ) is essential for effective production management. Traditional detection methods relying on external morphological traits like belly softness and color are often subjective, time-consuming, and unreliable. This study investigates the use of hyperspectral imaging (HSI) combined with machine learning to non-invasively assess maturity status in Atlantic salmon. A snapshot hyperspectral camera was configured to acquire images of 52 Atlantic salmon (females) specimens after harvest. The imaging system acquired images across 164 spectral bands spanning 350–1000 nm. The regions of interest (ROIs) were extracted, and reflectance spectra were analyzed. Spectral ratio analysis and principal component analysis (PCA) revealed class-dependent spectral variation, particularly in the visible and near-infrared regions. Multiple classification algorithms were tested on full-spectrum and feature-reduced datasets, with features selected using Random Forest importance and Jeffries–Matusita (J-M) distance. Key wavelengths contributing to class separation were consistently identified in the range of 350–500 nm. Ensemble models, particularly Extra Trees and Random Forest, achieved the highest classification accuracies, with the former reaching 81.8 % accuracy on the full dataset. These findings demonstrate the potential of HSI as a non-invasive tool for maturity detection and lay the groundwork for developing real-time and low-cost, multispectral sensing solutions for maturity detection in the RAS environment.