The Research On Fresh-frozen Pork Detection Based On Convolutional Neural Network And Near-infrared Spectroscopy

Pork is one of the most popular meats in the world.With the continuous improvement of people's living standards,people put more stringent requirements on quality and safety in the process of pork consumption.At present,the safety hazards and quality problems caused by the phenomenon of frozen meat posing as fresh meat on the market have aroused great concern.Compared with traditional identification methods such as sensory analysis and physical and chemical detection,NIR spectroscopy has many advantages such as lossless,fast,and green.Improving the accuracy and stability of detection is the key to near-infrared spectroscopy.At present,the optimization of prediction performance is mainly focused on the methods of spectral data preprocessing and effective feature band filtering.However,there is little involvement in the proposed and improved aspects of the correlation spectrum prediction model.Convolution Neural Networks(CNN)is a learning model with high performance and deep structure.Different from traditional neural networks,it combines the two processes of feature extraction and model training and automatically acquires relevant important information from the multi-dimensional and cumbersome spectral data for training,avoiding complex feature engineering and cumbersome data pre-processing.Therefore,this dissertation innovatively introduces the convolutional neural network algorithm into near-infrared spectroscopy and applies it to the direction of fresh-frozen pork.Firstly,the near-infrared spectrometer was used to collect the spectral curves of fresh and frozen pork samples,and the data set of the experiment was obtained by pre-processing the abnormal samples.Then,the convolutional neural network was designed and optimized through a series of experiments.Experiments show that the discriminant model based on convolutional neural network can obtain better prediction accuracy than the traditional model,and the accuracy rate reaches 97%.The feasibility and effectiveness of the fresh-frozen pork discriminant model based on convolutional neural network are validated.Secondly,the effects of different preprocessing methods and different training sets on the model are investigated,and the advantages of the convolutional neural network model are analyzed from multiple angles.Finally,based on the Keras neural network framework of Python language,this paper designs and implements a fresh-frozen pork discriminating system based on convolutional neural network,which can analyze the collected pork spectrum online and in real time.