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.

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.

Monitoring plant growth is crucial for effective crop management, and using color and depth (RGBD) cameras to model lettuce has emerged as one of the most convenient and non-invasive methods. In recent years, deep learning techniques, particularly neural networks, have become popular for estimating lettuce fresh weight. However, these models are typically specific to particular datasets, lack domain adaptation, and are often limited by the availability of open-access datasets. In this study, we propose a method based on plant geometric features for estimating the rosette structure and volume of lettuce. This new approach was compared to existing methods that reconstruct surfaces from point clouds, such as Ball Pivoting and Alpha Shapes. The proposed method creates a tight hull around the plant's point cloud, preserving high detail of the rosette structure while filling in surface holes in areas not visible to 3D cameras. Using a linear regression model, we estimated fresh weight for this dataset, achieving a root mean square error (RMSE) of 18.2 g when using only the estimated plant volume, and 17.3 g when both volume and geometric features were included. Additionally, we introduced new geometric features that characterize leaf density, which could be useful for breeding applications. A dataset of 402 point clouds of lettuce plants, captured before harvest, was compiled using one top-down and three side-view 3D cameras.

Monitoring individual cow feed intake is necessary for calculating the cow individual feed efficiency. The cost and maintenance time necessary for research systems make them impractical for most of the commercial producers. We developed a measurement system with producer convenience and low investment as key design criteria. The goal of this study was to design the system and validate its ability to rank cows by their feed conversion efficiency in commercial farms. The new system consisted of three principal parts: (a) a hanging weighing system, (b) a visual cow identification system and (c) an automatic cleaning system. The weighing system consisted of hanging a single load cell to provide feed mass measurements. The image-based cow identification system (replacing Radio-Frequency Identification) entailed cameras installed above the feeding area and an image processing algorithm that recognized cows by their collar numbers. The new system worked within normal farm routines: the feed supplying truck distributed the animal feed, and a tractor cleaned feed residual. To validate the accuracy and convenience of the system and to rank the cows by their efficiency, an experiment with six scales and 12 cows was conducted in a research barn, succeeded by eight-scale system in a commercial farm with 16 cows. The feed intake of each cow participating in the experiments was monitored for one month. The validation experiment showed that the system had the following specification: scales were accurate within 120 g; the visual cow identification rate was greater than 96%; feeding duration was accurate to 52 s; and routine farm practices (feed distribution, pushing, and residual removal) continued as usual. The cost for a feeding station (utilized consequently for a number of cows) was about 1 500 USD. An example of application of the system to rank cows by their efficiency under commercial conditions was shown. The system can potentially be used for ranking cows by their efficiency in commercial facilities.

Field robots are an important tool when improving the efficiency and decreasing the climatic impact of food production. Although several commercial field robots are available, the advantages, limitations, and optimal utilization methods of this technology are still not well understood due to its novelty. This study aims to evaluate the performance of a commercial field robot for seeding and weeding tasks. The evaluation was carried out in a 2-hectare sugar beet field. The robot’s performance was assessed by counting plants and weeds using image processing. The YOLOv8 model was trained to detect sugar beets and weeds. The plant and weed densities were compared on a robotically weeded area of the field, a chemically weeded control area, and an untreated control area. The average weed density on the robotically treated area was about two times lower than that on the untreated area and about three times higher than on the chemically treated area. The testing robot in the specific testing environment and mode showed intermediate results, weeding a majority of the weeds between the rows; however, it left the most harmful weeds close to the plants. Software for robot performance assessment can be used for monitoring robot performance and plant conditions several times during plant growth according to the weeding frequency.