Design Of Application System For Dense Counting Of Sheep Based On Deep Learning

The vast grasslands of the Inner Mongolia Autonomous Region provide a unique advantage for grassland animal husbandry.Therefore,Inner Mongolia Autonomous Region has become a vital livestock production base in China,and the booming sheep farming industry has become a landmark industry in Inner Mongolia.At the same time,a series of problems such as overgrazing and livestock claims have emerged,and sheep counting has become an essential part of solving these problems.The traditional sheep counting method is timeconsuming and inefficient.This paper proposes a method of dense sheep counting based on deep learning for the shortcomings of conventional methods,which provides a simple,fast and accurate platform for sheep counting and integrates the deep learning method into monitoring overgrazing in grasslands,which is an essential guideline for precision grazing and intelligent farm construction.By carrying an image acquisition device in the sheep pen for sheep image acquisition,the image is pre-processed and the data set is fed into a sheep intensive counting model based on a convolutional neural network for training so that the model can count sheep and thus achieve the purpose of intensive counting of sheep.Finally,to facilitate the relevant grassland management departments to carry out overgrazing monitoring and grassland actual livestock carrying capacity statistics,a sheep flock intensive counting application system is designed based on the counting model.The main work is as follows.(1)Cameras are built on sheep pens with heights ranging from 1.5 to 3 meters to collect images of sheep in sheep pens.The pixels of the collected images are not uniform.The collected images were selected and cropped,and MATLAB2015 a software was used to create a sheepfold dense flock dataset.(2)To make up for the shortcomings of data acquisition,large production workload,and lack of public flock dataset,the sheep image generation is performed using a conditional generative adversarial network to augment the flock dataset.(3)Comparing the counting effect of the improved Modified-SFANet algorithm and the Modified-Seg Net algorithm in the dense sheep flock dataset.(4)The system is based on an algorithmic model for better counting and uses the Django framework in combination with HTML and SQLite databases to enable users to count sheep through a web browser on Ubuntu systems.The proposed research on deep learning-based sheep dense counting method achieves accurate counting of sheep in highly shaded dense sheep within a specific error range.It provides data support for the relevant departments to carry out grassland ecological,environmental protection.The proposed method of data set augmentation using generative adversarial networks provides a reference for dense counting research.The designed sheepdense counting application system provides a realistic basis for sheep population counting and management in pastoral areas.