Feeding Behavior And Identification Of Pigs Based On Improved Optical Flow Method And Deep Learning

With the development of the technology of feature extraction,traditional feature extraction and manual monitoring methods have been unable to meet people’s needs,and computer vision technology has developed rapidly.At present,the main problem in my country’s breeding industry is that the health problems of livestock cannot be detected in time,which leads to the death of a large number of livestock due to common infectious diseases,resulting in heavy economic losses.The computer vision technology can not only improve automatic identification without contact,but also reduce manual labor.At the same time,this technology can identify the feeding behavior of pigs very efficiently.Therefore,it is great for the feature extraction.This paper proposes a method for pig feeding behavior and identification based on improved optical flow method and deep learning.I will introduce the main research content of this paper.(1)To improve the efficiency of using the optical flow method to identify the feeding behavior of pigs,this paper first performs framing processing on the bird’s eye view angle video collected in the group,and then preprocesses it,and then extracts the preprocessed The video frame,the key frame is obtained,and the data set is provided for the subsequent pig feeding behavior recognition and identity recognition.Finally,the optical flow method is used to identify the pig feeding behavior.First,the divided video frames are preprocessed by a combination of logarithmic transformation,bilateral filtering,and histogram equalization.Then,the analytic hierarchy process is used to roughly extract the key frames,and then the improved support vector machine to refine and extract key frames.The improvement of SVM mainly lies in the use of G-EIPSO to optimize the penalty factor C and the kernel function parameter g of the SVM.The optical flow method is used to identify the feeding behavior of pigs on the extracted key frames.The accuracy of pigpen A is93.33%,the recall rate is 97.46%,and the accuracy is 93.83%,which is much higher than that of ordinary LK optical flow.Method and pyramid layered LK optical flow method.(2)In order to extract better texture features,MB-ACDLDP is used to extract texture features,and the multi-block operation can extract more Details,the absolute value difference operation can reduce the computational load of the extraction process.The feature dimension extracted by LBP of an 8*8 picture is 8*8*256,which is 16384 dimensions.The feature dimension extracted by CS-LBP is less than that of LBP,so details will be lost.The texture feature dimension extracted by MB-ACDLDP is 8*8 *3*3*256/2 is73728 dimensions,the amount of information extracted is much larger than that of LBP and CS-LBP,and the calculation amount is greatly reduced because of the absolute difference operation,which provides a good foundation for subsequent pig identification.(3)To improve the efficiency of identification of feeding pigs,a method based on MB-ACDLDP features combined with deep learning methods was adopted.MB-ACDLDP extracts texture features,and the deep learning network uses PIG-VGG16.The PIG-VGG16 network adds a BN layer and a mean pooling layer on the basis of the VGG16 network.Only identify the network.Incorporating the extracted pig texture features into the network can increase the richness of the entire network and improve the sensitivity to pig texture features,thereby improving the recognition accuracy of the network.The average accuracy rate of the ALex Net network is 94.52%,while the average accuracy rate of the ALex Net network is 91.28% lower than that of the PIG-VGG16 network.The accuracy rate of the Google Net network is 94.89%,slightly higher than that of the PIG-VGG16 network,but the overall processing speed is 3 times that of the PIG-VGG16 network.,for pig identification,the PIG-VGG16 network is the most cost-effective.