Recognition Of Respiratory Diseases In Pigs With Multiple Sources Based On Deep Learning And Evidence Theory

With the development of China’s economy and social progress,people’s living standards have been significantly improved,especially in the dietary structure,people began to favor healthy and environmentally friendly food.Pork has become the main meat slices in daily life due to its delicious quality and rich nutrition.The scale of domestic pig breeding industry has also increased.Due to the intensive and humid environment in which pig pens are kept,it is difficult to keep clean and easy to spread diseases,which seriously affects the growth and development of pigs,resulting in a decline in pork production and economic losses suffered by farmers.Therefore,it is urgent to strengthen the prevention and identification of pig diseases.Traditional pig respiratory tract disease recognition methods are not real-time and efficient in today’s large-scale breeding,so the application of computer technology in disease recognition is of great significance for pig disease recognition and treatment.This article combines deep learning algorithms and DS evidence theory,is used to identify the swine respiratory disease,with deep learning network to the sick pigs anatomical organ image training,the trained model can for an organ to the pig,which can identify the disease with DS evidence theory fusion,the result of the different organs disease identification fusion results as the final result.The main contents are as follows:According to the three respiratory diseases studied in this experiment,relevant disease reports are sorted out and pathological characteristics are analyzed.The lungs are the respiratory organs of pigs,the liver is the detoxification organ of pigs,and the spleen is the lymphatic immune organ.Therefore,the study of these three organs is of great relevance to the identification of respiratory diseases.This article USES the deep learning Caffe ResNet to 18 under the framework of the network,the ResNet network adopted the deep web gradient residual block to solve the problem,the image in the convolution layer through convolution operation feature extraction,the pooling layer for data dimension reduction,compression and the number of parameters,to network res3a ResNet-18,res4a,res5a,res5b characteristics after the four layers of convolution calculation of deconvolution deconvolution layer to get the characteristics of the input layer to the classification of weighted fusion,obtain more sufficient characteristics,Finally,multi-scale feature fusion is output to the classification layer for image recognition.The modified network training is used to obtain the model,and the similarity between the features of the organ image and those of the network learning can be calculated when the anatomical images of pig organs are input,so as to identify diseases.Information fusion is carried out for the recognition results of various organ diseases obtained by deep learning,and the deep learning network is used to identify individual organ diseases and obtain the basic probability of disease recognition for each organ.DS evidence theory fusion results of three organs disease recognition,determine the disease recognition framework,based on the deep learning network provides the basic probability of each disease,calculate the uncertainty factor K,according to the K value recognition probability of each disease combined with individual organs after calculate the fusion recognition within the framework of the probability of disease,as the final disease recognition probability.Information fusion is used to fuse the results of disease identification of three organs,which can reduce the uncertainty of disease identification of a single organ and increase the reliability.This experiment in animal hospital,livestock epidemic disease inspection conducted by the anatomical diagnosis of pig lung,liver,spleen image diseased organs as sample,and get the laboratory diagnosis report as a verification,use of disease recognition system identification results of this study and the comparision of the diagnostic report,the results showed that the swine respiratory disease based on depth of learning-information fusion recognition accuracy of 85%or more,with high accuracy,the recognition is real-time,objectivity,porcine respiratory disease recognition provides a new effective path.