Automatic quality control of internal defects in cod - results from hyperspectral, ultrasound and X-ray imaging

Summary

Three different technologies were evaluated for the detection of embedded defects in cod fillets: ultrasound, x-ray imaging and the interactance acquisition mode of hyperspectral imaging. Previous studies had demonstrated interactance hyperspectral imaging to be a promising method to automatically identify nematodes in cod fillets. Improvements had been made to the illumination system that significantly increased the signal to noise ratio since that previous work. The detection rate with the new hyperspectral setup was approximately 40%, less than the former study. This is attributed to the storage time of the fillets as compared to the previous work and also differences in sample handling between the training and test set. Future work needs to be performed on-site in order to avoid results more pessimistic than would be encountered in industrial implementation. The ultrasound images were noisy and produced too little contrast between the muscle and nematodes, making reliable detection not feasible. Dual energy ray CT proved capable of detecting deeply embedded nematodes, but it is too slow and expensive for industrial application. Measurements on planar x-ray were inconclusive and require further study. While some high-density spots were observed in the loin in locations where nematodes were located during dissection, not all nematodes were identified and contrast was low.