Automatic nematode detection in cod fillets (Gadus morhua L.) by hyperspectral imaging
This is the accepted manuscript version. Published version available at http://dx.doi.org/10.1016/j.jfoodeng.2012.02.036 (PDF)
Detection of objects embedded in tissue, using visible light, is difficult due to light scattering. The optical properties of the surrounding tissue will influence the spectral characteristics of the light interacting with the object, and the spectral signature observed from the object will be directly affected. A method for calibrating the spectral signature of small objects, embedded in translucent material, by the estimated local background spectrum is presented. The method is evaluated under industrial conditions in a new hyperspectral imaging system for automatic detection of nematodes in cod fillets. The system operates at a conveyor belt speed of 400 mm/s which meets the industrial required speed of assessing one fillet per second. The local calibration method reduces the number of spectra needed to be classified by 89.6%. For one or more false alarms in 60% of the fillets sampled after the trimming station, the Gaussian maximum likelihood classifier detects 70.8% and 60.3% of the dark and pale nematodes, respectively. This is better than what is previously reported using a higher resolution instrument on a slow moving conveyor belt, and comparable or better to what is reported for manual inspection under industrial conditions.
CitationJournal of Food Engineering 111(2012) nr. 4 s. 675-681
MetadataShow full item record
The following license file are associated with this item: