Recognition Of Broiler Chicken Behaviour And Health Status Based On Machine Learning

Broiler health not only influences reproductive efficiency,but also has the potential to spread diseases and endanger human health.Intelligent broiler disease detection systems can help the breeding business cut labor expenses and halt the growth of an outbreak.There are numerous broiler diseases in commercial broiler breeding,and diagnosing specific diseases in broilers is cost consuming and inaccurate.At the moment,there is some research supporting the identification of general or specific behaviors such as estrus and lameness in livestock and poultry using machine learning,however there is minimal research supporting the identification of livestock and poultry health status.However,as a result of broiler disease,numerous pathological behaviors like as lack of energy,prolonged lying down,and loss of appetite will arise.The goal of this research is to understand the welfare status of broilers through the identification of their behavior,and to realize the identification of the health status of broilers and even early warning of diseases through behavioral data.The purpose of this study was to plan and construct a broiler behavior research site in a flatraised broiler house.The broilers were separated into three groups:poisoned,mildly poisoned,and normal.By injecting aflatoxin AFB1(Aflatoxin B1)into broilers,abnormal behavior is induced.A three-axis acceleration sensor and camera were used to collect acceleration and video data from broiler behavior.After denoising the triaxial acceleration data with a Kalman filtering algorithm,18 eigenvalues in the time and frequency domains of the denoised data were extracted.A total of five algorithms K-Nearest Neighbour(KNN),Support Vector Machine(SVM),Decision Tree(DT),Random Forest(RF),and Gradient Boosting Decision Tree(GBDT)were used to develop broiler behavior recognition and diseased chicken detection models.The following are the primary research conclusions:(1)To classify broiler chickens’ physical state,physiological indicators were collected from those in the poisoning group,the moderate poisoning group,and the normal group.AFB1 was demonstrated to successfully produce pathological behaviors such as lethargy and loss of appetite in broilers by the study of video recordings of their behavior.The specific manifestations are increased lying time and decreased feeding,drinking,standing and walking behavior.;(2)The Kalman filter algorithm is capable of effectively reducing noise in three-axis acceleration data.The time and frequency domain eigenvalues calculated from triaxial acceleration data exhibit a high link with the degree of broiler poisoning and the static behavior of broilers;(3)The models for recognizing broiler behavior and unhealthy chickens developed using the five machine learning algorithms perform better in classification and recognition,and the classification performance of the models constructed using DT,RF,and GBDT is superior to that of KNN and SVM.The GBDT model’s accuracy rate for recognizing broiler behavior can reach 94.02%,while its accuracy rate for recognizing diseased broilers can reach 97.65%.