Abstract:In broiler production, the temperature under the wing is an important indicator of animal health and welfare condition. Body temperature detection method of broiler based on infrared thermography was proposed to achieve measurement of broiler body temperature accurately and rapidly. The detected region of interest (ROI) model of broiler head and leg, based on a convolutional neural network, was developed to extract the maximum temperature of its head and leg. Besides, combined with ambient temperature, humidity and light intensity, two different broiler wing temperature inversion models were proposed by multiple linear regression and back propagation （BP）neural networks, respectively. And the experimental results showed that, based on the deep convolutional neural network, the ROI detected model achieved a precision and recall rate of 96.77% and 100% on the test dataset, respectively. What’s more, the temperature inversion models achieved an average relative error of 0.33% with multiple linear regression, while BP neural network was 0.29%. Deep learning method was used to obtain the ROI temperature, which was superior to the image processing method, high in efficiency and high in generalization ability. BP neural network model error was less than the error of multiple linear regression network model. Therefore, BP neural network can be applied as a temperature inversion model of broiler wings. BP neural network had the ability of selflearning and selfadaptation, and its generalization ability was strong. Applying it to the inversion of temperature under the wing can improve the accuracy and adaptability of the model. This model provided reliable technical support for realtime monitoring of broiler body temperature.
