Free Range Laying Hens Monitoring System Based on Improved MobileNet Lightweight Network

Abstract

In order to solve the problems such as the difficulty in artificial supervision, the high cost of managing to breed, and the difficulty in monitoring the number and state of the chickens, this paper proposed a layer monitoring system based on the improved MobileNet lightweight neural network. MobileNet (target classification network) is combined with SSD (target detection network), with MobileNetV2 network selected to replace VGG-16 in the SSD network as the basic feature extraction network, and all standard convolution in SSD regression detection is replaced by deep separable convolution to construct MobileNetV2-SSD target detection network. According to the VOC standard, the layer data set was built, the layer detection model was trained, and the layer monitoring system deployed on Raspberry PI 4B was constructed. The experimental results show that the monitoring system constructed in this paper achieves a detection accuracy of 79.17% on the self-built layer data set, and the detection speed of about 10fps can be achieved when running on Raspberry PI 4B, which basically meets the requirements of accurate and real-time detection, and can effectively monitor free-range layers and assist the management of chicken houses.