Hyperspectral Analysis Of Total Viable Count On Chilled Mutton Surface And Design Of Detection Platform

The hygienic quality of chilled mutton can be reflected by the total viable count (TVC) on its surface. this study uses a hyperspectral imaging system within the wavelength range from 400 nm to 1100 nm to realize the acquisition of the hyperspectral data 85 samples of chilled mutton based on the hyperspectral technology, and uses a variety of algorithms to establish the TVC prediction model of chilled mutton surface and the classification model of the freshness of chilled mutton, analyzes and compares each model, chooses the best modeling method, designs a detection platform for the calculation of the TVC on the surface of chilled mutton with the combination of the graphical user interface (GUI) of MATLAB, and finally realizes the intelligentization of the detection of the quality of chilled mutton.The main results of this study are as follows:(1) Acquires the hyperspectral data of chilled mutton samples (within the wavelength range from 400 nm to 1000 nm). Uses a variety of methods to preprocess the original hyperspectral data, and based on the BP-ANN model, reaches that the best preprocess method is SNV combines WT and SG Also, applies the PCA algorithm to successfully reduce the dimension of the hyperspectral data.(2)Uses respectively the four classical methods of BP-ANN, RBF-ANN, PLSR and SVM to establish the TVC prediction model of chilled mutton surface. In order to improve the prediction accuracy of the model, introduces a new neural network algorithm, ELM. Realizes ELM and KELM algorithms, and proposes the PSO and GA optimized KELM model. Through analysis and comparison, the prediction accuracy of the 4 models based on ELM algorithm is better than the classical method, of which the best method is GA-KELM model. Its correlation coefficient(R) and root mean square error(RMSE) of the training set and prediction set is:0.9300,0.0016; 0.9835 and 0.0015.(3)Establishes a classification model of the freshness of chilled mutton from the aspect of TVC. Realizes the modeling of PSO-SVM and ELM respectively, as well as of their optimization methods CPSO-SVM and FA-ELM. Analysis and comparison result of the four methods shows that CPSO-SVM is the best classification model, whose training set accuracy reaches 95.313% and prediction set accuracy reaches 100%.(4) Designs a detection platform for the calculation of the TVC on the surface of chilled mutton by using the graphical user interface (GUI) of MATLAB, which includes four modules as follows:TVC detection, freshness identification, preprocessing and result analysis. By using the functions of the interface, realizes the fast, nondestructive and intelligent detection and analysis of the quality of chilled mutton.