Study On Non-destructive Assessment Of Microbial Viable Cells On Chilled Mutton Surface Based On Hyperspectral Imaging Technology

The number of microbial viable cells in chilled mutton surface is an important index for evaluating the quality of food hygiene. The traditional detection method for it always produced of low efficiency, long cycle and the meat was seriously damaged. Hyperspectral nondestructive testing technology integrate images and hyperspectral spectral information, using spectral differences between different objects to reveal the quality parameters of the advantage, has received widespread attention. The implementation of agricultural and livestock products has important practical significance to the rapid nondestructive detection.This study in chilled mutton as the research object, used hyperspectral imaging technology to carry out the nondestructive detection on microbial viable cells in different wavelength range.The quantitative prediction model and quality grading model of microbial viable cells on chilled mutton surface were established, then extracted the surface pollution combined with hyperspectral imaging technology. Through the comparative analysis of different wavelengths, different pretreatment methods and different modeling methods to determine the best index and method for detection of microbial viable cells on chilled mutton surface.These detection algorithms offer foundation and theory for developing a online nondestructive detection system. The main research achievements are as follows:(1)The prediction models of TVC and PPC on chilled mutton surface show that the optimal preprocessing methods for different wavelength range were MSC+SD and MSC.(2)The prediction models of the microbial viable cells on chilled mutton surface were established by the PLS, BP-ANN, RBF-ANN, and the best prediction result was based on RBF-ANN model. The correlation coefficient R were0.9872,0.9988and root mean square error RMSEP were0.8210,0.2507for prediction of TVC in range of400～1100nm wavelength and900～1700nm, correlation coefficient R were0.9780,0.9802and root mean square error RMSEP were0.7370,0.6639for prediction of TVC in range of400～1100nm wavelength and900～1700nm.(3)The quality grading models were established by the BP-ANN, RBF-ANN, SVM. It has a better result by the supporting vector machine model, and the recognition rate were78.94%and84.21%in the range of400～1100nm and900～1700nm wavelength.(4)The prediction results and quality grading results in range of900～1700nm wavelength were better than that in the range of400～1100nm, which were the best spectral range for nondestructive detection.(5)Dimensionality reduction using principal component analysis, characteristic bands of the principal component analysis and band ratio algorithm can extract the contaminated area on the surface effectively, and the last one had the higher correct rate, which reached92.5%.