| In recent years,the scale of mutton market in China has been increasing and the sheep industry is mainly free range.However,large-scale and intensive sheep farms are the mainstream in the future,therefore,applying modern information technology to sheep farms is an important trend of future development.At present,the development of sheep farm informatization is lagging behind,and some key intelligent breeding technologies,including intelligent identification of individual sheep,phenotypic data measurement of sheep growth and production traits,and the development of integrated intelligent breeding expert system,have made little progress.Therefore,the combination of intelligent breeding technology and sheep farm production data can effectively guide sheep production management and improve sheep raising efficiency.This research applies the sheep face recognition algorithm based on convolution neural network to solve the intelligent identification problem of sheep in the breeding farm,and uses the sheep contour extraction algorithm based on neural network and the sheep body size measurement algorithm based on computer vision to automatically measure the sheep body size.On this basis,an intelligent sheep farm expert management system is integrated to assist the sheep farm in daily management,breeding and scientific decision-making.The main research contents and results are as follows:(1)112 Hu sheep were captured from the sheep farm,from which 90456 sheep face images were obtained through image transformation,enhancement,similarity comparison and annotation,and convolutional neural network algorithm was applied to sheep face image recognition as well.It was found that the first layer of AlexNet neural network 11×11convolution kernel is suitable for extracting features of larger objects,therefore,it is replaced by four smaller convolution kernels to increase the receptive field level and extract multi-scale features of sheep face images.Since the role of Local Response Normalization(LRN)is not obvious,the LRN layer in the AlexNet network is removed to reduce the amount of computation and the first full connection layer in the network is replaced by the attention mechanism module and replace the Rectified Linear Unit(Re LU)with the Mish activation function.The results show that compared with the AlexNet neural network,the Visual Geometry Group(VGG)neural network,and the Goog LenNet neural network,the AlexNet network model based on attention mechanism has better comprehensive performance,with a recognition speed of 15.86 ms and an accuracy rate of 98.82%.(2)2924 body profile images of 112 Hu sheep were captured from the sheep farm and were labeled as a data set using Labelme software and sheep contour was extracted through semantic segmentation neural network.This project compares three semantic segmentation networks: UNet,Pyramid Scene ParsingNetwork(PSPNet)and Deep Lab V3+,and uses MobilNet V2 as the backbone network.The results show that when Deep Lab V3+ is used as a semantic segmentation network,the comprehensive performance of the model is the best.Its mean intersection over union can reach 98.51%,and the number of frames per second can reach 29.3.(3)In order to calculate the sheep’s body size parameters,the project first divides the sheep into three parts: head,trunk and leg according to the sheep contour,and then divides the sheep posture into three types: normal,bowed head and turned head by calculating the relative positions of the three parts.Then classify the sheep pose using the Support Vector Machine(SVM)algorithm,designing different sheep contour feature point extraction algorithms for different poses,calculating the position of seven feature points on the contour,converting the feature points on the sheep image contour to three-dimensional coordinates and calculating the sheep body size parameters through the feature point coordinates.The results show that the average error of the above algorithm is less than4.63%.This non stress measurement method of sheep size parameters can provide help for welfare sheep raising on farms.(4)Finally,this project uses the distributed microservice structure to design and develop a smart sheep farm management system,and successfully applies the sheep size measurement data to this system.The system is divided into web and mobile terminals,each with 9 modules,respectively: production management,biosafety management,material management,inventory management,breeding management,purchase and sales management,information entry,problem reporting and statistics reporting.The smart sheep farm expert management system developed in this project can provide intelligent reminders on the vaccine effect,treatment effect and performance measurement of breeding ewes,ewes to be tested for pregnancy,ewes to be produced,artificially inseminated ewes and weaned lambs,and provide timely and professional breeding information guidance for sheep farm management,and conduct dynamic analysis on daily weight gain and body shape change through body size and body weight data during sheep growth.The combination of body size parameters of sheep and production performance data aims to promote accurate sheep breeding and provide technical support and data support for scientific breeding of sheep industry.The AlexNet model based on attention mechanism has good performance in speed and accuracy,which is suitable for the production environment of sheep farm.The measurement algorithm of sheep’s body size with different posture has strong applicability and small error.The intelligent sheep farm management system can improve the production management efficiency of sheep farms.The project can improved the efficiency of sheep farm in intelligent breeding and scientific breeding. |
PDF Full Text Request