Monitoring System Of Sow Delivery Nursing House Based On Image Recognition

In recent years,with the rapid development of modern information technology,intelligent equipment and wireless sensor network technology has penetrated into various fields.However,in the domestic animal husbandry field,especially the sow breeding industry,the management mode is still relatively traditional.China is a big pig country,but not a big pig country.Its sow management model is far from that of developed countries such as the United States.Sows are considered to be the "engine" of a pig farm,and the management of sows is more difficult than that of fattening pigs and the breeders need to provide more technology and energy.The change of sow management directly affects the economic benefit of pig farm.This paper mainly introduces a monitoring system of sow nursery based on image recognition.The system consists of environment data acquisition,image acquisition unit,control equipment unit and terminal equipment.Among them,environmental acquisition units are distributed around the delivery bed of sow farms,which are used to collect temperature,humidity,light and other data.They are uploaded to the data measurement and control center through ZigBee network,and then sent to the terminal device for display.The image acquisition unit is mainly used for image information acquisition in the breeding field to provide support for image recognition process.The control equipment unit mainly consists of ventilation system,temperature control system of nursing house and automatic feeding system,which can be automatically adjusted according to the environment and remotely controlled by terminal equipment.The terminal equipment is used to display the environment and image data,control the equipment,and identify the sow's birth behavior.Firstly,this paper summarizes the development status of sow management system at home and abroad,and puts forward the overall design scheme of the system according to the domestic status.In the part of image recognition,the deep learning algorithm is selected,and the Faster RCNN deep learning algorithm is applied to the sow delivery behavior recognition.In the early stage of image recognition,3000 sow delivery images were prepared,and the establishment of data set was carried out by using the software labelImg to modify the hyperparameters of the Faster R-CNN algorithm,so as to carry out the training of deep learning model.Analysis of the test results and expansion of the data set can improve the recognition accuracy,which eventually reaches 96.8%.The design of the farm monitoring system and the installation of image recognition environment were carried out for the terminal equipment.Finally,the overall system was built and the system was verified.According to the test,the wireless communication between the devices maintains a good stability,and the recognition effect of sow's delivery behavior is particularly prominent,which proves the stability and reliability of the system.