FilletCam AI: A handheld tool for precise fillet color profiling of Atlantic salmon and rainbow trout
High-throughput and objective color measurements of fish fillets are crucial for quality control and establishing pricing benchmarks. While fillet color scoring traditionally relies on subjective visual inspection using color reference cards or labor-intensive color measurement using a colorimeter,...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2024-12-01
|
| Series: | Journal of Agriculture and Food Research |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2666154324004988 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | High-throughput and objective color measurements of fish fillets are crucial for quality control and establishing pricing benchmarks. While fillet color scoring traditionally relies on subjective visual inspection using color reference cards or labor-intensive color measurement using a colorimeter, this study aims to develop a smart handheld device (FilletCam AI, hereafter referred to as ‘FilletCam’) for precise and rapid profiling of fish fillet color. FilletCam comprises an imaging sensor integrated with an embedded computing unit for image acquisition. A graphical user interface (GUI) was developed to streamline data collection, analysis, visualization, compilation, and data sharing with authorized users. An Artificial Intelligence (AI) model was developed and deployed on the edge for region-of-interest (ROI) detection. A single-stage YOLOv8 model was trained on a custom image dataset comprised of Atlantic salmon (Salmo salar) and rainbow trout (Oncorhynchus mykiss) fillet images to detect the fillet and color palette with their respective SalmonFan™ (SF) score on the reference card. Delta E metric was adopted to compare the color of the fillet and reference color shades on the SF card. The SF score corresponding to the lowest Delta E value was assigned to the fillet. The object detection model performed well and achieved a mean average precision (mAP0.5) of 99.5 % and an F1 score of 0.99. The FilletCam-predicted fillet color scores were compared with the expert ratings and colorimeter scores to evaluate the performance of the developed tool. The minimum Delta E values for FilletCam were consistently lower than those for the colorimeter, indicating FilletCam's ability to detect minor color differences accurately. FilletCam exactly predicted the color scores at 73 % of instances compared to 30 % of instances for the colorimeter. Additionally, for only 3 % of instances, the predicted SF score deviated by more than two points. Overall, the developed research prototype shows promise as a valuable digital tool for the fish processing and retail industries. |
|---|---|
| ISSN: | 2666-1543 |