Research And Implement Of Sows' Parturition Intelligent Detection System Based On Machine Vision And FPGA

China is the largest pork producer and consumer country in the world.Pigs' health is directly related to food security,human health,environmental pollution and other issues.In the process of breeding pigs,sows' healthily breeding is not only related to their own reproductive rate and productivity,but also piglets' birth and health.For the current pig large-scale,intensively breeding way,the sow breeding generally adopts the limit column,and the feeding amount of daily limit on a sow.Relative to other pigs,the welfare of the sows is poorer,and lactation piglet mortality is higher.In traditional breeding way,to judge the time of sow parturition depends on continuous observation of feeder,that makes human burden and low efficiency and the death of piglets.Thus,real-time monitoring the sow parturition state accurately to find the newborn piglets and provide early warning,has practical significance in enhancing the health degree of sows and welfare breeding level.With the rapid development of computer and machine vision technology,image processing technology has been widely applied in the field of livestock welfare breeding.Growth and health information of livestock is obtained from their images.According to the actual demand in sows breeding and production,a sow cough monitoring system is designed based on machine vision technology in this paper,which can provide a judgment gist for the early warning of sows' parturition detection and recognition.This paper mainly studied the following aspects.(1)Sow parturition images acquisition is studied based on FPGA and CMOS camera.And a parturition image acquisition environment is built on the basis of existing conditions of the pig farm,to acquire sow parturition images.(2)A sows' parturition detection system is designed and realized,including obtaining sows' parturition information,developing a WEB monitoring platform and an Android management terminal.It is convenient for breeders to record,query and manage related information of sows.(3)This paper studies the elementary process of sow parturition image processing,to realize parturition images' gray scale operation.Than do image edge detection by Canny operator.Secondly,image transformation by Otus algorithm into binary image.Thirdly,moving average algorithm and morphological opening operation was applied for binary image denoising.The maximum connected domain was extracted using group sequence detection algorithm to partition the targeted sow.Finally,according to the segmentation area,targets were recognized.The test is carried on in pig farm of Nanjing Pukou agriculture museum.Eight Landrace sows from production date for about three days are selected as subject.Among them,4 is with obvious about to parturition symptom,and another 5 is normal.The study shows,sow parturition images could be obtained quickly,steadily and automatically.Effective methods are designed to do sow target segmentation and newborn piglets recognition.Ignoring individual differences,the overall recognition accuracy of sow parturition reaches about 95.5%.The recognition algorithm works effectively.The monitoring system basically meets actual requirements.