SBIR Phase I: Deep Learning Classifier Using Hyper Spectral Imaging for Automated Processing of Fish Fillets

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project will be to advance the development of solutions for the fish processing industry. Current fish fillet preparation depends highly on visual inspection and manual trimming by trained technicians to identify and remove blemishes and food-borne parasites, which remain a significant cost of seafood preparation. Preparation of seafood by hand also poses significant health risks for consumers. The presence of human technicians means significant portions of seafood processing must be performed at room temperature, increasing the risk of bacterial infection and enzymatic degradation. This decreases the freshness and quality of the final product, while increasing chances of foodborne illness. Additionally, the manual detection rate for nematode infections can be as low as 50% under industrial conditions due to the small size of the infections and the potential depth in the fish. The proposed automated hyperspectral imaging (HSI)-based platform addresses these issues by (1) fully automating the complete fish processing supply/value chain, (2) obviating industry dependence on manual inspection and (3) being much less error-prone and more precise than manual inspection and trimming, reducing lost yield and promoting safer products for consumers. This Small Business Innovation Research (SBIR) Phase I project will advance the development of a hyperspectral imaging (HSI) system for the automatic processing and remediation of fish fillets. HSI has already been validated in the detection of anatomical features of fish fillets and detection of harmful parasites like nematodes. However, its use has not been validated on an industrial scale. The proposed research will focus on: (1) compilation and annotation of a fish fillet data set, (2) training and validation of deep learning algorithms to detect fish fillet center line and nematode infestations, and (3) assessing the feasibility of using a state-of-the-art robotic actuator and customizing it for automatic nematode removal with the company's novel algorithms. The result will be a computer vision platform capable of recognizing nematodes/other foodborne parasites, detecting blemishes, blood spots, and other undesirable cosmetic marks, as well as conducting precision removal of identified parasites and blemishes without excess loss of fish meat. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.