Identification of bone fragments embedded in lean pork slices by comparing the pixel areas of bone regions segmented from single-band images

Abstract

This paper was concerned with the detection of hazardous bone fragments embedded in lean pork slices (LPS-BFs) by comparing the pixel area of bone regions segmented from single-band image. All images were collected using a Videometer A/S multispectral imaging system with 19 bands (405–970 nm). An area library of bone regions was established by processing 60 images captured from LPS-BFs and the minimum value of the library was defined as a boundary point for judgement. The accuracy of 90 % for 120 test samples (60 LPSs and 60 LPS-BFs) was achieved and all results could be visualized for monitoring the geometrical morphology of bones. Results of satisfactory identification of bone fragments embedded in LPSs indicated that image features at single-band could serve as a feasible approach for online industrial applications.