Development of an early warning algorithm to detect sick broilers

Abstract The frequent occurrence of poultry diseases, such as bird flu, not only causes huge economic losses to farmers but also seriously threatens the health of human beings. Providing early warnings of new poultry disease outbreaks is essential in poultry breeding. With the rise of digital image processing technology and machine learning algorithms, real-time monitoring of poultry health status through cameras is an effective way to prevent large-scale outbreaks of disease. To analyze the postures of healthy and sick broilers, bird flu virus was inoculated intranasally into healthy broilers manually. The broilers were then placed in isolator cages for comparative experiments. The methods of observing the posture changes of broilers and extracting the key features are used to realize the automatic classification of healthy and sick broilers. In this research, broiler images are obtained, and two kinds of segmentation algorithms are proposed to separate the broilers from the background to obtain the outlines and skeleton information of the broilers. According to the preset feature extraction algorithm, the posture features of healthy and sick chickens are extracted, the eigenvectors are established, the postures of the broilers are analyzed by machine learning algorithms, and the diseased broilers are predicted. A series of experiments have been done. Data for each feature acquired by the algorithms are analyzed, and the effect of each feature on the recognition accuracy is obtained. Using some of the features proposed in this research, accuracy rates of 84.248%, 60.531% and 91.504% are obtained, but using all the features can yield an accuracy rate of 99.469%. Then, the recognition effects of several commonly used machine learning algorithms are compared. The Support Vector Machine (SVM) model obtains an accuracy rate of 99.469% on the test samples, which is superior to those of the other machine learning algorithms. The experimental results show that the algorithms proposed in this research can effectively separate broilers from the background, extract the posture information of broilers, and accurately and quickly identify the health status of broilers by means of SVM. The algorithms for digital image processing and machine learning are evaluated in the diagnosis of broiler health status and show high accuracy, good stability and good generalization performance, and can give early warning signals. This research can provide a reference for the intelligent identification of broiler health status in the future.