Fish fillet color plays a crucial role in customer acceptance. Traditional visual fillet color assessments are often labor-intensive, subjective, and lack standardization . Colorimeters have b een used for meat and fillet color assessments ; however, their accuracy leaves room for improvement as they provide point color measurement and fail to capture the non-uniformity of fillets. Rece nt computer vision technologies have shown promise in accurately assessing fillet color, but their application is often limited to algorithm optimization. T his study describes a portable computer vision tool for the rapid and precise color profiling of fish fillets. The tool consists of a single-board computer integrated with a red-green-blue (RGB) camera, facilitating data acquisition, onboard image processing, and data visualization. An automated algorithm was developed and optimized for resource-constrained edge devices to segment fish fillets and color palettes on the SalmoFan™ Lineal scale in the field of view of the camera. Various color difference metrics, such as Delta E (CIELAB 1976, 1994, and 2000), and hue angle were employed to assess the visual disparities in color between color palettes on the SalmoFan ™ Lineal scale and the fillet. The minimum Delta E and difference of the hue angle were used to rate fillet color on SalmoFan ™ scale ranging from 20 to 34. The performance of the tool was validated on 60 fillet portions (2 species × 5 fish/species × 2 fillet/fish × 3 Portions/fillet) obtained from 10 fish (species: Atlantic Salmon and Rainbow Trout) cultured in Recirculating Aquaculture System. The obtained fillet color values were compared with the visual ratings provided by three experts. An evaluation of the performance and accuracy of the developed digital tool for fish fillet color categorization will be presented.
Keywords: Precision aquaculture, Internet of Things (IoT) , Edge computing, Automation , Fish processing