Research On Yellow-feather Broiler Heat Stress Behavior Recognition And Evaluation Index Based On Visual Technology And Deep Learning

In southern China,the summer has high temperature,high humidity and long duration.This makes broilers extremely prone to heat stress,which seriously affects the large-scale and welfare-based breeding of broilers.In order to accurately analyze the status of the broilers in chambers in the high temperature environment on summer days,this experiment uses yellow feather broiler as the research object,analyzes and recognizes their behaviors in the heat stress environment,and combines the temperature and humidity data to evaluate the existing heat stress,which are optimized.Based on the above description,the contents of this research are as follows:(1)According to the differences in broiler behavior under heat stress environment,the behavior classification standard was determined,and the broiler behavior image data set under heat stress environment was produced,laying the foundation for subsequent research.(2)Through deep learning model experiments,combined with the current CNN model’s low classification accuracy and poor aggregation behavior classification problems,the Res Ne Xt network is introduced,and an improved Cascade R-CNN model is presented for broiler behavior recognition.Through the comparison and analysis of multiple algorithms,it is found that the improved algorithm model solves the problem of inaccurate behavior recognition in the case of broilers gathering,and the final recognition accuracy rate reaches88.4%.(3)Three machine learning models were established to test the classification of heat stress.Combining the behavior image data set with the temperature and humidity data,a new numerical data set is extracted,including the proportion of broiler behavior and the corresponding temperature and humidity data.The machine learning algorithm is used to realize the classification of heat stress,which proves that machine learning can be applied to the classification of heat stress in broiler chambers.Among them,Random Forest has the best classification effect.By comparing the accuracy of adding different behavior variables,the variable group is determined with the most obvious lifting effect.(4)Optimization of heat stress assessment model.A variety of statistical analysis methods are used to fit the behavioral variables and temperature and humidity,and finally the conclusion drawn by the partial least squares regression method is determined as the optimization model,and the behavior recognition results and optimization formulas are verified to prove the optimization formula The feasibility of this is conducive to the reasonable management and control of the broiler chamber.The experimental results of this paper show that the improved Cascade R-CNN algorithm can accurately identify the behavior of broilers under heat stress and solve the problem of low recognition accuracy of clustered broilers.The recognition accuracy is higher than that of the Retina Net algorithm and Faster R-CNN algorithm and original Cascade RCNN algorithm.At the same time,the original heat stress evaluation index was tested and improved,and finally the broiler behavior variables were added on the basis of the original evaluation formula and verified.The research in this article is conducive to the healthy breeding of yellow feather broilers.It can detect whether the broilers have heat stress problems in time and provide a reference for early warning of their heat stress environment.