Fast identification method of healthy and diseased meat based on refined feature information extraction
[Objective]Aiming at the rapid identification of healthy and diseased meat,herein we investigate the information extraction and classification methods of the surface Raman spectra of healthy and diseased meat.[Methods]Taking the surface-enhanced Raman spectrogram of mutton as a sample,we use two methods,principal component analysis-support vector machine and convolutional neural network for classification,respectively.Through the refined feature extraction of the spectrogram,the filtering of the spectrogram degradation and interference information is accomplished,thus providing more accurate and rich feature information for the classification model.In the experimental validation,240 Raman spectra containing healthy and diseased mutton were used as the training set samples to build the classification model,and other 120 samples were used to validate the identification effect between healthy and diseased meat.[Results]Experiments show that the principal component analysis-support vector machine model constructed after refined feature extraction can clearly find the classification boundary between healthy and diseased meat,and the recognition accuracy of the validation samples rises from 82.5％to 93.3％.At the same time,if the convolutional neural network that learns and classifies refined extracted features is used,the recognition accuracy rises from 90.2％,achieved by the conventional method,to 95.5％.[Conclusions]The refined feature information extraction and classification method of meat spectra based on surface-enhanced Raman proposed herein can effectively achieve the rapid classification and identification of healthy and diseased meat in mutton samples.Additionally,it can be applied to the detection and classification of other meats,and this application leads to great potential in guaranteeing food safety.
万方数据
维普
