Development and Analysis of a CNN- and Transfer-Learning-Based Classification Model for Automated Dairy Cow Feeding Behavior Recognition from Accelerometer Data.

Due to technological developments, wearable sensors for monitoring the behavior of farm animals have become cheaper, have a longer lifespan and are more accessible for small farms and researchers. In addition, advancements in deep machine learning methods provide new opportunities for behavior recognition. However, the combination of the new electronics and algorithms are rarely used in PLF, and their possibilities and limitations are not well-studied. In this study, a CNN-based model for the feeding behavior classification of dairy cows was trained, and the training process was analyzed considering a training dataset and the use of transfer learning. Commercial acceleration measuring tags, which were connected by BLE, were fitted to cow collars in a research barn. Based on a dataset including 33.7 cow × days (21 cows recorded during 1-3 days) of labeled data and an additional free-access dataset with similar acceleration data, a classifier with F1 = 93.9% was developed. The optimal classification window size was 90 s. In addition, the influence of the training dataset size on the classifier accuracy was analyzed for different neural networks using the transfer learning technique. While the size of the training dataset was being increased, the rate of the accuracy improvement decreased. Beginning from a specific point, the use of additional training data can be impractical. A relatively high accuracy was achieved with few training data when the classifier was trained using randomly initialized model weights, and a higher accuracy was achieved when transfer learning was used. These findings can be used for the estimation of the necessary dataset size for training neural network classifiers intended for other environments and conditions.

Evaluation of MEMS NIR Spectrometers for On-Farm Analysis of Raw Milk Composition.

Today, measurement of raw milk quality and composition relies on Fourier transform infrared spectroscopy to monitor and improve dairy production and cow health. However, these laboratory analyzers are bulky, expensive and can only be used by experts. Moreover, the sample logistics and data transfer delay the information on product quality, and the measures taken to optimize the care and feeding of the cattle render them less suitable for real-time monitoring. An on-farm spectrometer with compact size and affordable cost could bring a solution for this discrepancy. This paper evaluates the performance of microelectromechanical system (MEMS)-based near-infrared (NIR) spectrometers as on-farm milk analyzers. These spectrometers use Fabry-Pérot interferometers for wavelength tuning, giving them the advantage of very compact size and affordable price. This study discusses the ability of MEMS spectrometers to reach the accuracy limits set by the International Committee for Animal Recording (ICAR) for at-line analyzers of the milk content regarding fat, protein and lactose. According to the achieved results, the transmission measurements with the NIRONE 2.5 spectrometer perform best, with an acceptable root mean squared error of prediction (RMSEP = 0.21% w/w) for the measurement of milk fat and excellent performance (RMSEP ≤ 0.11% w/w) for protein and lactose. In addition, the transmission measurements using the NIRONE 2.0 module give similar results for fat and lactose (RMSEP of 0.21 and 0.10% w/w respectively), while the prediction of protein is slightly deteriorated (RMSEP = 0.15% w/w). These results show that the MEMS spectrometers can reach sufficient prediction accuracy compared to ICAR standard values for at-line and in-line fat, protein and lactose prediction.

The effects of body posture and temperament on heart rate variability in dairy cows.

Reactivity of cattle affects many aspects of animal production (e.g. reduced milk and meat production). Animals have individual differences in temperament and emotional reactivity, and these differences can affect how animals react to stressful and fear-eliciting events. Heart rate variability (HRV) is a good indicator of stress and balance of the autonomous nervous system, and low parasympathetic activity is connected with higher emotional reactivity. The study had two specific aims: (1) to compare HRV in dairy cows for standing and lying postures (no earlier results available), and (2) to assess whether dairy cows' emotional reactivity is connected to their HRV values. Eighteen dairy cows were subjected twice to a handling test (HT): morning (HT1) and afternoon (HT2), to evaluate emotional reactivity (avoidance score, AS). HRV was measured during HT (standing). HRV baseline values, both standing and lying down, were measured one week before HTs. HRV was analyzed with time and frequency domain analyses and with the Recurrence Quantification Analysis (RQA). Heart rate (HR), low-frequency/high-frequency band ratio (LH/HF), % determinism (%DET) and longest diagonal line segment in the recurrence plot (Lmax) were higher (p<0.05) while the cows were standing than when lying down, whereas the root mean square of successive R-R intervals (RMSSD) (p<0.05) and power of the high-frequency band (HF) (p<0.1) were higher while the animals were lying down. HR, the standard deviation of all interbeat intervals (SDNN), RMSSD, HF, power of the low-frequency band (LF), % recurrence (%REC), %DET, Shannon entropy (p<0.05), and HF (p<0.1) were higher during the handling test compared to standing baseline values. AS (i.e. tendency to avoid handling) correlated positively with SDNN (r=0.48, p<0.05), RMSSD (r=0.54, p<0.05), HF, RMSSD (r=0.46, p<0.1) and LF (r=0.57, p<0.05), and negatively with %DET (r=-0.53, p<0.05), entropy (r=-0.60, p<0.05) and Lmax (r=-0.55, p<0.05) in the baseline HRV measurements. AS correlated positively with SDNN (r=0.43, p<0.1) and HF (r=0.53, p<0.05) during HT. Some HRV parameters (HR, LF, %REC, %DET) indicated that the handling test may have caused stress to the experimental cows, although some HRV results (SDNN, RMSSD, HF, entropy) were controversial. The correlations between HRV variables and AS suggest that the emotional reactivity of the cow can be assessed from the baseline values of the HRV. It is debatable, however, whether the handling test used in the present study was a good method of causing mild stress in dairy cattle, since it may have even induced a positive emotional state. The posture of the cow affected HRV values as expected (based on results from other species), so that while standing a shift towards more sympathetic dominance was evident. Our results support the idea that linear (time and frequency domain) and non-linear (RQA) methods measuring HRV complement each other, but further research is needed for better understanding of the connection between temperament and HRV.