Research On Determination Of Pig Carcass Composition Based On CNN

As the meat of choice for Chinese residents,pork has always been the most concerned issue in the pig breeding,slaughtering and food processing industries.Currently recognized pork quality evaluation indicators are the lean meat rate and intramuscular fat content in pig carcasses.As an effective non-destructive testing technology,CT technology has broad application prospects in the detection of pig carcass composition.With the rapid development of deep learning,the deep learning combined with CT non-destructive testing technology to achieve the determination of pig carcass composition has become a research hotspot.In this paper,a convolutional neural network is applied to the determination of pig carcass composition,and a convolutional neural network prediction model combining Inception module and MSVR algorithm is proposed.The specific work of this paper is as follows:(1)Acquisition and processing of experimental data.In order to construct a complete CT image dataset for the determination of pig carcass composition,firstly,CT images of live boars with body weights between 20-50 kg and corresponding label data(intramuscular fat content and lean meat rate of pigs)were collected..In order to facilitate the training and learning of the subsequent models,the original CT images were preprocessed,including format conversion,image denoising and image data enhancement.(2)Construction of predictive model of pig carcass composition based on CNN.Using the convolutional neural network to automatically extract the characteristics of image features,four different depths of CNN models were constructed based on traditional CNN.Experiments on the constructed datasets were used to verify the prediction of pig carcasses on CT images using convolutional neural networks.The feasibility of the composition.(3)CNN optimization model combining Inception module and MSVR algorithm.The network structure of the traditional CNN is improved,and some convolutional layers in the network are improved to the Inception module,thereby deepening the structure of the network model.And combined with the MSVR algorithm,the intramuscular fat content and lean rate were simultaneously predicted.Using the optimized CNN model as a feature extractor combined with MSVR modeling not only improves the accuracy of the prediction of the composition of the two pig carcasses,but also realizes the multidimensional output of the prediction results.