Animal behavior is an important indicator of animal welfare evaluation, is the entry point of accurate poultry breeding, and for poultry farming, due to the small size of chickens and large breeding scale, automatic monitoring, especially individual monitoring has great difficulties, the realization of PLF of chicken farms is a difficult problem, in view of this problem, this paper proposes a method based on improving Faster R-CNN for individual behavior recognition of laying hens. The VGG16 in the Faster R-CNN object detection network was replaced with the ResNet50 network, and the RPN was improved, the number of proposals was reduced, the ROI increased, and the accuracy was improved, which could automatically identified the four basic behaviors of individual laying hens of standing, lying down, feeding and drinking. Simulate the construction of the living environment of laying hens in real state, collected videos of laying hens in natural state as input, divided the videos into pictures, and identify individual behaviors of 3200 pictures, and the recognition rates of standing, lying down, feeding and drinking water were 96.4%, 92%, 86.3% and 80.2%, respectively. At the same time, it can reduce the contact between farmers and chickens, and increase the safety of biological epidemic control.