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.

Monitoring of Cow Location in a Barn by an Open-Source, Low-Cost, Low-Energy Bluetooth Tag System.

Indoor localization of dairy cows is important for cow behavior recognition and effective farm management. In this paper, we propose a low-cost system for low-accuracy cow localization based on the reception of signals sent by an acceleration measurement system using the Bluetooth Low Energy protocol. The system consists of low-cost tags and receiving stations. The tag specifications and the localization accuracy of the system were studied experimentally. The received signal strength propagation model and dependence on the tag orientation was studied in an open-space and a barn environment. Two experiments for the evaluation of localization accuracy were conducted with 35 and 19 cows for two days. The localization reference was achieved from feeding stations, a milking robot and videos of cows decoded manually. The localization accuracy (mean ± standard deviation) was 3.27 ± 2.11 m for the entire barn (10 × 40 m2) and 1.9 ± 0.67 m for a smaller area (4 × 5 m2). The system can be used for recognizing long-distance walking, crowded areas in the barn, e.g., queues to milking robots, and cow's preferable locations. The estimated system cost was 500 + 20 × (cow number) € for one barn. The system has open-access software and detailed instructions for its installation and usage.

Assessing the potential of photogrammetry to monitor feed intake of dairy cows.

We address the hypothesis that individual cow feed intake can be measured in commercial farms through the use of a photogrammetry method. Feed intake and feed efficiency have a significant economic value for the farmer. A common method for measuring feed mass in research is a feed mass weighing system, which is excessively expensive for commercial farms. However, feed mass can be estimated by its volume, which can be measured by photogrammetry. Photogrammetry applies cameras along the feed-lane, photographing the feed before and after the cow visits the feed-lane, and calculating the feed volume. In this study, the precision of estimating feed mass by its volume was tested by comparing measured mass and calculated volume of feed heaps. The following principal factors had an impact on the precision of this method: camera quality, lighting conditions, image resolution, number of images, and feed density. Under laboratory conditions, the feed mass estimation error was 0·483 kg for heaps up to 7 kg, while in the cowshed the estimation error was 1·32 kg for up to 40 kg. A complementary experiment showed that the natural feed compressibility causes about 85% of uncertainty in the mass estimation error.