Study On Dead Birds Detection System Based On Machine Vision In Modern Chicken Farm

The automation has been realized widely in modern chicken farm,such as automated supply of fodder and water,automated collection of egg and dung,and automated control of temperature and wind speed.But a major labor-intensive task still remains,that is timely detection and removal of dead birds.There are two problems about this.For one hand,it may not remove the dead bird timely,which endangers alive birds.For the other hand,manual work efficiency is low,and extended working hours inside a laying-hen house may have some impact on the welfare of the workers.In order to solve this problem,an automatic dead birds detection system based on machine vision was proposed.The system consisted of ARM board,,GPRS MODEM,remote monitoring center and two algorithms of dead birds detection based on image processing.The one was based on red area extraction and "AND" operation.Through identifing the motionless red cockscombs,dead birds could be detected for individual cage.To pick-up the red cockscombs,a~* of the L~*a~*b~* color space was taken as characteristic variant,the improved method of maximum classes square error(Otsu) was used as segmentation criterion,and the morphological filter was taken to reduce noise.Knocked on the cage and collected another picture and processed it using the same method above.Finally,AND operation was taken to detect whether there was motionless cockscombs in the cage.If yes,sent the warning message and transmitted the original image to remote surveillance center through GPRS MODEM.The experimental results showed that the correct identification rate exceeded 85%.The other was based on contour extraction and SVM.Whether there was dead bird in the cage was identified by distinguishing whether the picture belonged to the dead bird image class.Firstly,took two pictures continuously on the same cage,then processed the images and extracted the contour of birds.Secondly,a center radiating vector representation was used to abstract features of the contour.Minused the corresponding vectors in the two pictures,and the difference was taken as input of SVM.Thirdly,trainning samples were used to train SVM.Finally,testing samples were input to classify alive bird pictures and dead bird pictures.If the dead bird picture was found,sent the warning message and transmitted the original image to remote surveillance center through GPRS MODEM.Experimental results showed that the correct identification rate exceeded 95%.In the future,the system can be installed in the vehicle that moves through the passage between cages,in order to detect and remove dead bird timely,which is important for the realization of all-around automation in modern hen-house.So far, the research on this subject is still under the way.So the paper is of some instruction to the research since.