The Study On Detection Of Lamb TVBN And Color Parameters Based On Hyperspectral Imaging

Lamb meat is as an important source of meat consumption in China, which is loved by people because of its high nutritional value. However, lamb meat is prone to deterioration during storage so that detecting the quality of mutton is very important. Traditional methods for the detection of lamb meat quality have disadvantages including the low detection efficiency and the complicated operation, which can not meet the fast and nondestructive online-detection.In this study, we took the lamb meat produced in Xinjiang as the research object, and did the study on the rapid and nondestructive detection of the color parameters(brightness L *, redness a *, yellowness b *) and total volatile basic nitrogen(TVBN) by using hyperspectral image acquisition system,whose spectral range is 400-1000 nm. The test totally obtained the spectral information and color physical and chemical parameters of 140 samples and the spectral information of content of TVBN for 72 samples determining the content of TVBN by experiment in the same time, and found the impact of the results of the prediction model by combining image processing technology and chemometrics knowledge to explore lamb meat color parameters and TVBN content, and using different characteristic wavelength selection methods, sample partition and spectral pretreatment methods in order to detect the color parameters and TVBN content of lamb meat rapidly and nondestructively. The specific content and results are as follows:(1)Use the same spectral pretreatment method, and compare the effects of L*, a*, b* and TVBN contents on PLSR model respectively by using the different methods(RS method, KS method, three separate selection method and SPXY method). The optimal sample set partitioning method for L* and b* was determined as the SPXY method, and the method for the content of a* and TVBN was divided into three separate selection method.(2)Compare the results of L*, a*, b* and TVBN contents on PLSR model respectively by using the different spectral pretreatment methods or the combination methods(mean-centering, S-G, MSC, SNV, wavelet transform, 1D, 2D and so on), and determine that the optimal pretreatment method of L*, a*, b* was 2D with SG(17).What difference was that b* was also combined SNV and a* was SNV coupled with MC. The optimal pretreatment method of the contents of TVBN was 1D, SG(21), SNV and MC.(3)The study compared and analysed the effect on model results of L*, a*, b* value and TVBN contents by using different modeling methods. Through using of three modeling methods concluding PLSR, SMLR and PCR to model, it determined the optimal modeling method for L*, a*, b* value and TVBN contents was PLSR(4)The study selected the characteristic bands of L*, a*, b* and TVBN contents by using si PLS method and GA algorithm and combined the selected characteristic bands with the corresponding physical and chemical indexes to build the PLSR model. The results showed PLSR models establish with GA characteristic bands were able to predict color patameters, As for L *, a * and b * the correlation coefficient of calibration set(RC) was 0.98, 0.93 and 0.81, and root mean square errors of calibration set(RMSEC) was 0.86, 0.55 and 1.02, respectively. The correlation coefficient of prediction set(RP) was 0.98, 0.87 and 0.81, and root mean square error of prediction set(RMSEP) was 0.98, 0.73 and 1.04, respectively. PLSR models establish with GA characteristic bands were able to predict TVBN contents. The correlation coefficient of calibration set(RC) and root mean square errors of calibration set(RMSEC) was 0.98 and 0.91, respectively, The correlation coefficient of prediction set(RP) and root mean square error of prediction set(RMSEP) was 0.89 and 2.92, respectively.