FISH-SPEC: Fast identification system for handheld spectroscopy and species classification

Authors: SUEKER, MITCHELL - University Of North Dakota; MACKINNON, NICHOLAS - Safetyspect Inc; BEARMAN, GREGORY - University Of North Dakota; TABB, AMANDA - Chapman University; KIM, DIANE - Chapman University; HELLBERG, ROSALEE - Chapman University; AKHBARDEH, ALIREZA - Safetyspect Inc; MARATEB, HAMIDREZA - Safetyspect Inc; Qin, Jianwei; Kim, Moon; VASEFIN, FARTASH - Safetyspect Inc; KASHANI ZADEH, HOSSEIN - University Of North Dakota

Submitted to: Applied Food Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/20/2025
Publication Date: 11/22/2025
Citation: Sueker, M., MacKinnon, N., Bearman, G., Tabb, A.M., Kim, D., Hellberg, R.S., Akhbardeh, A., Marateb, H., Qin, J., Kim, M.S., Vasefin, F., Kashani Zadeh, H. 2025. FISH-SPEC: Fast identification system for handheld spectroscopy and species classification. Applied Food Research. 5(2). Article e101536. https://doi.org/10.1016/j.afres.2025.101536.
DOI: https://doi.org/10.1016/j.afres.2025.101536

Interpretive Summary: Mislabeling of fish fillet species threatens consumers’ health, undermines consumers’ confidence, and damages the reputation of the seafood industry. To help solve this problem, we developed a handheld sensing device that can identify the species of the fish fillets in seconds. The device shines different types of light on the surface of the fish fillets then reads and analyzes how the light reacts with the tissue. We developed AI computer programs to teach the device how to recognize the fish species based on the collected light signals. The device was tested using a total of 68 fillets from 11 fish species. Data were collected from each fillet sample in both frozen and thawed states. A classification accuracy of 90% was achieved to differentiate the 11 species for the thawed, the frozen, and the combined fish fillets. The handheld device can be used for on-site detection of the fish fillet mislabeling, which can help authenticate the fish fillets for the consumers, the seafood industry, and the regulatory agencies.

Technical Abstract: A major challenge in the seafood industry is the correct identification of the species of fish fillets. We have developed a handheld multi-mode point spectroscopy system as an accurate and non-destructive solution to identify fish fillet species in seconds. It utilizes a combination of spectroscopic modes, pairing fluorescence (365 and 395 nm excitation) in the visible near-infra-red range (~350-900 nm), and reflectance in the visible near infra-red range and short wave infra-red range (~900-1700 nm). We obtained tissue spectra with our device for each spectroscopic mode at 25 different positions on fillets from 11 fish species, with a minimum of three fillets per species, and the entire dataset comprising 68 total fillet samples. Each fish was measured in frozen and thawed states to develop separate machine-learning models to classify the correct species under different conditions and an overall model that included thawed and frozen samples. Feature level fusion was used to incorporate the data from each spectroscopic mode, demonstrating better results than each mode’s performance individually. The model that included fish in both states could classify the 11 different species correctly 85 ± 2.8% of the time. We developed dispute models to enhance performance by training separate models with commonly misclassified species, with the dispute models’ improving the accuracy of the model with fish in both states to 90% ± 6.1%. The thawed and frozen models individually achieved 90 ± 6.0% and 90 ± 5.4% respectively, with dispute models in the thawed dataset increasing accuracy to 93 ± 4.3%.