Study On The Detection And Identification Method Of Sheep Grazing Behavior In Grazing Grassland

In order to solve the problem of traditional detecting methods in monitoring sheep's grazing behavior in grassland,which mainly implemented by man's direct observation with time-consuming and laborious,inaccurate,also affecting the normal animal's feeding behavior,an automated testing system of sheep's grazing behavior was developed by using the modern measurement technique,also the correlation of grassland quality with sheep grazing behavior was established in the dissertation.It will provide a basis for grassland evaluation,grassland management and policy decision of grassland ecological restoration.In view of the characteristics of herd behavior of grazing sheep and the feeding behavior of individual sheep,a micro test device was designed to detect the movement behavior and feeding behavior of the grazing sheep.The main conclusions are as follows:(1)In the system,an embedded microprocessor STM32VET6 was employed to acquire the data from a GPS NEO-6M,a three-axis acceleration sensor MPU6050 as well as a temperature and humidity sensor,and to fuse the video signals from a micro-camera which was installed on the right side of the sheep's head,the match of sheep grazing data in time and space domains and the detection of related sheep grazing behavior were realized.The related test data was uploaded to the Tencent cloud server by a GPRS wireless data transfer system and was saved to the MySQL database under the Apache environment.The HTML was used to design web browsers to collect,display and store the typical behavior data of the grazing sheep.The accuracy,reliability and effectiveness of the device was verified by comparing with the other methods.(2)The wavelet threshold denoising method and an improved method were employed to denoise the three-axis acceleration data of grazing sheep behaviors.The experimental results showed that both of them can attain the good effect,and the improved wavelet threshold function denoising algorithm can achieve a better denoising effect.The K-mean clustering algorithm was applied to identify the resting behavior of grazing sheep accurately,then GPS velocity signal was matched to get the recognition result of grazing sheep's movement behavior and standing behavior.It has been verified that the accuracy of all detections is above 85% compared to manual direct observation method.(3)The video of sheep grazing herbage was preprocessed to extract the images of forage leaf intaked by the grazing sheep,that is ROI(Region of interest).By extracting the color,shape and texture features of herbage,by matching the effective features and reducing the feature dimension,the classification of herbage was completed by BP neural network training and learning,realized the recognition of sheep's grazed grass species.Compared with artificial simulation method,the overall recognition rate reached nearly 80%.(4)The vegetation coverage of typical steppe is normally used to characterize the quality of grassland.A prairie navigation car was developed to collect the pasture image data.By building the feature rule base for typical pasture,segmenting the herbage from the soil background using the fuzzy C-mean clustering algorithm based on 2G-B-R in the feeding pasture area,and then performing the vegetation segmentation for typical pasture by neural network method,the image pixels were statisticed in order to calculate the coverage of whole or single vegetation at the sheep's feeding location.It shows that the recognition accuracy is 89.6% by comparing with the manual direct observation method.(5)A statistic analysis on the grazing sheep movement and feeding behaviors were conducted in the dissertation,established the correlation study with the microscopic grassland quality.The experimental results show that the microscopic vegetation coverage and environmental factors have a certain correlation with the movement and the feeding behaviors of the sheep,further validated the feasibility,accuracy and effectiveness of the system detection strategy and the recognition algorithm for the movement behavior and forage recognition.