Design Of Pet Dog Monitoring System Based On Deep Learning

In foster care for pet dogs,due to improper nursing work in foster care institutions,pet dogs had physical and mental health problems and even escape loss.In pet dogs care,accurate and quantitative behavioral activity testing had a wide range of health benefits for dogs,therefore,it was necessary to monitor the activities of pet dogs.In the research on monitoring methods of dog activity,manual monitoring wasted human resources,machine vision monitoring equipment cost too high and there were monitoring dead corners.Thanks to the development of wearable devices and deep learning,the pet wearable device connected with the Internet was a good choice for pet dog activity monitoring,improving the nursing efficacy of pet dog and reducing the monitoring cost.A study on the design of pet dogs activity monitoring system was carried out as follows:1.It analyzed the requirement of pet dogs activity monitoring,put forward the scheme design of the system and the production of pet dog behavior data set.Designed a pet dog behavior and location monitor,which integrated GPS positioning,4G LTE modules,accelerometers and gyroscope sensors.The mobile monitor was fixed to the neck collar of the pet dog,and the video camera was used to record the behavior of the pet dog and to synchronize with the data collected by the sensor in the mobile monitor on the time axis,and labeled six kinds of dog behavior,such as running,waiting,beating,drinking,eating,scratching,completed the data set production.2.The traditional machine learning method was used to construct the classification model of pet dog behavior activity.Random forest,k-nearest neighbor algorithm and artificial neural network were selected for experimental comparison,the method of category weight was used to solve the problem of data set sample imbalance.The experimental results showed that the artificial neural network model had the best classification effect among the three algorithms,the classification model achieved an average accuracy of 91% on the validation set,but in drinking water,eating such similar behavior classification task classification effect was not ideal.3.The complex model of convolution neural network and long-short term memory neural network was used to identify the behavior of pet dogs.Using the method of transfer learning to solve the problem that the amount of data of pet dog behavior activity was small,introduced multi-pair model design and multi-scale feature fusion to simulate the longer time series dynamic changes,and improved the performance of multi-channel sensor time series classification network,finally,the classification network was designed based on the Filter-Net model.The model achieved 95% average accuracy on the test set,4% increased compared with artificial neural network model,the performance of the model and the effectiveness of the improvement strategy were verified.4.The specific design realized and tested the function of the system.The design of pet dog activity monitor,pet dog GPS location and electronic fence,pet dog activity information web visualization page were completed.System test results showed that the dog monitor can collect data at 30 HZ frequency;GPS positioning error was within10 m;the accuracy of electronic fence entry and exit alarm reached 95%;The activity identification model can accurately detect the important canine behavior related to health,the recall rates of drinking water and eating behavior were 95% and 98%,respectively.