Chicken Object Detection Using Faster Region-Based Convolutional Neural Networks

Monitoring chickens is important for managing health factors and assessing welfare. Increased infectious disease transmission brought on by increased poultry production can result in significant financial losses as well as widespread chicken mortality. Because of its non-invasive and invasive characteristics as well as its ability to display a wide range of information due to advancements in information technologies, computer vision can be trusted for the real-time automation of chicken health monitoring systems. Using a Faster Region-Based Convolutional Neural Network (Faster R-CNN) and sample images totaling 605 taken from poultry houses, this work introduces deeper learning to identify various breeds of chickens in the context of poultry. The best convolution architecture for real-time chicken detection was determined by testing and implementing three different convolutional architectures: ResNet-50, ResNet-101, and ResNeXt-101. Our findings show that we were more successful with Faster R-CNN in the context of ResNet-50. ResNet-50 was the most appropriate architecture to be implemented for real-world use in poultry management systems since, in spite of having more convolutional layers, its real-time processing time was quicker than ResNet-101's. The efficiency of ResNet-50 and Faster R-CNN has been used to monitor poultry more quickly and precisely.