Bluetooth-enabled multimodal sensor boards, data collection software stack, and machine learning model to identify early signs of lameness in dairy cattle

Objectives

To demonstrate the use of custom electronic sensor boards with software stack to automatically collect real-time movement data from multiple cows and to develop a machine learning model to process the collected data for the identification of early signs of lameness.

Materials and methods

We developed an electronic sensor board incorporating state-of-art sensors, including 3-axis accelerometer and gyroscope, electret microphone and infrared temperature sensor. Each sensor board was controlled by an on-chip, Bluetooth-enabled 2.4 GHz Nordic Semiconductor microprocessor and powered by a 3.3 volts, 800 mAh coin cell battery. A software stack was built in Python and included modules to wake up specific sensor boards, activate the desired sensors, collect the data for a pre- specified time duration, and store the recorded data in user- defined file formats. The performance of each sensor, the sensor boards, and the software stack was characterized in an Electrical Measurement Laboratory over several days to evaluate different performance metrics, such as sensitivity, resolution, temperature range, amplitude range, frequency and range of operation, power consumption, and battery life. The sensor boards were attached to both ear pinnae of individual cows to record multimodal sensor parameters, such as head acceleration, head rotation, head tilt, vocalization, ear skin temperature, and ambient temperature. A local server communicated with the sensor boards on ten cows to collect data from individual sensors every one second within an operating range of 50 to 100 feet. Video recordings from portable camcorders were collected for select time durations to externally validate the data recorded from sensor boards. [...]