- health control based on pH and temperature from a ruminal sensor
- fat-to-protein ratio control, based on milk parameters analysis, in order to detect improper nutrition
N-SAAW is a one-of-a-kind Deep Neural Network (DNN)-based system for monitoring farm animal health and wellbeing. This method was created specifically to examine the milk protein to fat ratio, encompassing extremes associated with malnutrition. This type of detection aids in lowering the danger of ketosis or acidosis induced by starvation. N-SAAW has already been deployed and evaluated in real-world agricultural settings, where it has been shown to identify malnutrition about 3.5 times sooner than current analytical approaches. Furthermore, because of the pre-processing, there was no possibility of the DNN missing starvation.
N-SAAW may be effortlessly linked with farm sensors and provides a way for autonomous monitoring to collect data that is then analyzed to provide insights to farmers. It also includes a visualization and reporting tool, allowing the provenance to be disclosed farther down the supply chain.
The solution developed and demonstrated was a tool to support the monitoring and analysis of milk cows’ bio parameters (temperature, pH). The data on temperature and pH are gathered from a ruminal probe.
The tool’s core functionality is based on a Recurrent Neural Network trained to predict the possible health deterioration of a specimen. In case of a predicted threat, an alert is triggered to inform the farm manager about the malnutrition of an animal. The triggering value has been set to a pH = 5.8, sustaining over a period of a few hours. Figure 1 presents the tool’s user interface, showing measurements and predictions tables, alerts, and pH values over time.
DAC conducted an initial validation of the health deterioration (malnutrition) prediction model. The results of the validation are considered satisfactory. As shown in the confusion matrix below (Figure N), the accuracy of “bad” and “good” predictions is reliable, producing an insignificant number of false positives. The mismatch between “warning” and “good” can be reduced with an auxiliary analytical algorithm.
Figure 2: Confusion matrix of the Recurrent Neural Network (RNN) trained to predict possible health deterioration resulting from malnutrition.