Study On Positioning Algorithm Of Grazing Goats Based On Machine Vision

In recent years,with continued global warming and unscientific grazing behavior,the grassland in Otog Banner,Ordos City,Inner Mongolia Autonomous Region has been degraded in a large area.In order to realize the rational utilization of grassland resources and promote the sustainable development of pasture,the Agriculture and Animal Husbandry Bureau of Otog Banner has adopted a series of information technology means to strengthen the supervision of pasture grazing.Taking the first branch of Otog Banner of Inner Mongolia Yiwei White Cashmere Goat Co.,Ltd.as the pilot,14 PTZ cameras are set up in the field to realize monitoring of the pasture around the clock,and GPS is used for outdoor goats positioning.Due to the limitations of GPS in power consumption,manual maintenance and comfort of goat when wearing equipment,this subject takes the Albas cashmere goat in the first branch of Otog Banner as the main research object.Focusing on the current issue of GPS application in pasture goats positioning,this thesis studies and develops a grazing goats positioning algorithm based on machine vision by making full use of the video images taken by the existing PTZ cameras in the pasture.The algorithm realizes fast and accurate goats positioning under the condition of very low computing power and memory requirements.The main tasks of this thesis are as follows:(1)The cashmere goat image data set is constructed based on the processing images collected by PTZ cameras at the pasture.The coordinate conversion text data set is constructed based on the data of goat’s longitude and latitude from GPS,which are matched,sorted and integrated with the images taken by PTZ cameras.(2)A goat target detection algorithm based on shallow convolutional neural network is studied.A shallow convolutional neural network named Shallow SE is proposed,which uses channel attention mechanism SENet,Ge LU activation function and Layer Normalization.Custom＿YOLO object detection module is obtained after simplifying and optimizing the PANet part and YOLO Head part in YOLOv4.And the goat target detection algorithm model based on Shallow SE-Custom＿YOLO is constructed.The overall parameters of the model are only 4.5M,the average detection accuracy is 95.89 %,and the detection time is only 8.5 ms.(3)A 3D coordinate regression algorithm model based on fully connected network is proposed.The fully connected network is used to fit the conversion relationship between 2D coordinates and 3D coordinates of the target to predict the spatial coordinates of goats,and the average distance error is only 0.88 m.(4)Design and visualization of goats positioning algorithm based on machine vision.Combining the goat target detection algorithm and 3D coordinate regression algorithm,goats positioning algorithm based on machine vision is proposed.A visual interface for goats positioning is built to realize visualization of goat target detection and target positioning results.After testing the accuracy and the speed of positioning,the average error of positioning that this goats positioning algorithm has is only 0.94 m and the positioning time is 0.21 s,which realizes fast and accurate positioning.