Prior ReseARCH

Source

Umstätter, C., et al. “Classification of Sows' Activity Types from Acceleration Patterns Using Univariate and Multivariate Models.” Computers and Electronics in Agriculture, Elsevier, 21 May 2010, https://www.sciencedirect.com/science/article/abs/pii/S0168169910000116.

Similar Research Found in Literature Review

Both the research presented in the "Classification of Sows' Activity Types from Acceleration Patterns Using Univariate and Multivariate Models" paper and our E205 final project use 3-axis accelerometer data to estimate state
Both the prior research in the paper and our project use the accelerometer vector equation as described in Figure 1 below

Motivation

It is difficult to monitor individual animals when large groups are housed together. The researcher's objective is to automatically classify the activities of group-housed pigs.

Accelerometer data is collected by affixing a sensor to each pig’s collar. This allows for monitoring of animal welfare and monitoring during the full duration of the reproductive cycle.

Figure 1. Equation for length of acceleration

Figure 2. Dynamic Linear Model that smooths out data

mETHOD - kALMAN fILTER

Data was collected from the accelerometer in all three directions. The researchers classified 5 types of activity: feeding, walking, rooting, lying laterally, and lying sternally.

They collected a learning data set, which consisted of about 10 min for each activity type. The test data set is 10 x 2 min for each activity type.

A Multi-Process Kalman Filter that contains 20 Dynamic Linear Models (5 activities x 4 axes) is used. Each time series is modeled using a Dynamic Linear Model (smooths out data) with cyclic components.

Figure 3. Block diagram for Kalman Filter inputs and outputs

Results

“It can be seen that each activity type displays a distinct acceleration pattern” (264).

Feeding and lateral/sternal lying activities are best recognized
Walking and rooting activities are mostly recognized by a specific axis corresponding to the direction of the sow’s movement while performing the activity (horizontal and vertical)

Figure 4. 10 minutes of acceleration data for each of the 5 states

Figure 5. Probability of pig being in each state during a trial 2 minute trial

Page updated

Report abuse