Design And Implementation Of Cow Behavior Recognition Base On Inertial Sensor And DAC Of Its Control Circuit

Promoting the dairy industry and its production capacity of dairy products is conducive to improve people’s quality of life.It is of great significance to monitor the behavior of dairy cattle to improve the safety of dairy products and the breeding level of dairy farms because the cows tend to behave abnormally when they are sick or in estrus.The information approach of cow behavior monitoring based on computer vision has the disadvantages of a large amount of data and high demand for computing resources,on sound or vibration sensor is easy to be affected by the external environment,so this paper selects inertial sensor for cow behavior recognition.With the progress of the micro mechanical system and integrated circuit technology,MEMS / ASIC integrated inertial navigation microsystem is invented,which provides new information technology for low-cost monitoring of cow health.Aiming at the bottleneck of cow behavior recognition,this thesis studies the deep learning cow behavior recognition method based on inertial navigation data,meanwhile optimizing the precision and the power consumption of the DAC of the measurement and control logic loop in the MEMS /ASIC integrated inertial navigation microsystem.The main research contents and relevant conclusions are shown below:（1）Construction of deep learning behavior recognition model based on inertial navigation.The inertial navigation data of different behaviors of cows are collected using a commercial inertial measurement system.After Kalman filtering to remove the noise,the cow behavior recognition model based on CNN and LSTM is constructed.And then the LSTM model is combined with the CNN model to complete the CNN-LSTM cow behavior recognition based on inertial navigation data.Finally,by comparing the results of the three models,the conclusion is that the average recognition accuracy of the CNN-LSTM model for the four behaviors recognition results of cows standing,lying down,eating,and walking reaches 96.0%,which is 6.1%,and 4.9% higher than that of the CNN model and LSTM model,respectively.The superiority of the CNN-LSTM model for inertial guidance-based cow behavior recognition is well demonstrated.（2）DAC design in MEMS / ASIC inertial guidance integrated microsystem for cow behavior recognition.The gyro calibration DAC of the measurement and control logic loop in the MEMS / ASIC integrated inertial navigation microsystem for cow behavior recognition developed by the project team is designed and optimized.The DAC adopts thermometer combined with R-2R wide swing segmental structure,and improves the accuracy and power consumption of DAC through the design of weighted resistance network,MOS switch and output feedback resistance of operational amplifier.Under the simulation condition that the operating temperature is between-45°and 125°and the sampling frequency is 800k Hz or3.2MHz,the simulation results show a signal-to-noise ratio of 80.7856d B and a spurious-free dynamic range of 85.5457 d B.The power consumption of the DAC achieves 390μW while ensuring the acquisition accuracy of the inertial sensor.Finally,the back-end layout design,parasitic parameters extraction,and post-simulation were completed.The gyro measurement and control system on chip which integrates the DAC has been taped out to foundry in November 2021.