Research On Herding Image Segmentation And Counting Algorithm Based On Machine Vision

In the process of animal husbandry production,it is very important to analyze and estimate the balance of grassland and livestock,the health of grassland and the capacity of carrying livestock continuously and dynamically for making scientific policy of grassland utilization,promoting the optimal management of animal husbandry and the healthy and sustainable development of grassland.At present,there is little research on the automatic identification and counting of sheep in pastoral areas.In China,the number of livestock is still counted manually,especially the counting of sheep is labor intensive and inefficient,and the actual carrying capacity is difficult to be counted in time.To solve this problem,combined with the uniqueness of sheep detection in outdoor environment,this paper proposes an automatic counting method of sheep based on machine vision,which is simple and fast to count the number of sheep grazing,and has certain application value.In this study,CMOS digital camera is used as the image acquisition hardware device in the DJI Phantom 4 UAV.The images of sheep herd and sheepfold at different heights of 15 m,20 m,25 m,30 m,35 m and 40 m are collected respectively.Through the image calibration at different heights and the effect comparison of image processing at the later stage,combined with the complexity of calculation,25 m is the best shooting height.By taking pictures and cutting them by hand,100,50 and 50 overlapped images of single,double and multiple sheep were obtained as training and testing images respectively.By using machine vision technology,through analyzing the background composition and the cause of noise of sheep image in complex background,the preprocessing of sheep image is carried out,which mainly includes the enhancement of image by gray histogram equalization,the denoising by Wiener filter and the denoising by image morphology processing combined with area threshold,which can effectively remove the background debris and noise.After preprocessing,the maximum variance between classes method is used to binarize the target image.Considering the range of body size of sheep of different childbearing ages,an automatic counting algorithm based on the pixel area threshold of sheep is proposed.The accuracy of the established sample image is More than 90.53%,the error is mainly due to the small size of the sheep in the image of adhesion,which leads to the small pixel area and the small counting.Through the error analysis,a sheep counting algorithm based on the segmentation of adhesion target and the area threshold is proposed.By comparing three different watershed image segmentation algorithms based on distance,gradient and marker,it is found that the watershed algorithm based on marker has the best effect on image segmentation of conglutinated sheep.After segmenting the conglutinated sheep,the parameters of counting algorithm based on body size area threshold are improved,and the accuracy of counting is 96.45%.The automatic counting method of sheep based on machine vision is proposed in this paper.The method of image acquisition is simple and easy.After collecting the images of sheep in foraging state in grazing grassland,the automatic counting of sheep is realized with high accuracy.