The bacon quality grade intellectual pattern recognition based on neural network of hyperspectral imaging

This paper researches on bacon quality pattern recognition. We take the hyper spectral microscopic image of plaque area, the acid value, the peroxide value (POV) and total volatile basic nitrogen (TVBN) obtained from the spectra at the Self Organization Map (SOM) neural network as inputs to conduct multi-data fusion. From that we set up a model about level detection of bacon quality. Eventually we refine the level of bacon quality to four categories based on the national standard with rough division of 2 classes, Those are: Safe to eat, Edible, Do not recommend eating and Inedible, which increase the detection accuracy of pattern recognition, and can propose an early warning for eating bacon safely. The study found that the detection model's root mean square prediction error (RMSEP) is respectively 3%, 9%, 4% of acid value, peroxide value and TVBN built after pretreatment and band selection. The number of wavelengths used here is about one-third of the whole number of wavelengths, which has improved the detection accuracy of pattern recognition. Theories and methods of the research can be further extended to detect other related agricultural meat products.