Monitoring cattle behaviour is core to the early detection of health and welfare issues and to optimise the fertility of large herds. Accelerometer-based sensor systems that provide activity profiles are now used extensively on commercial farms and have evolved to identify behaviours such as the time spent ruminating and eating at an individual animal level. Acquiring this information at scale is central to informing on-farm management decisions. The paper presents the development of a Convolutional Neural Network (CNN) that classifies cattle behavioural states (‘rumination’, ‘eating’ and ‘other’) using data generated from neck-mounted accelerometer collars. During three farm trials in the United Kingdom (Easter Howgate Farm, Edinburgh, UK), 18 steers were monitored to provide raw acceleration measurements, with ground truth data provided by muzzle-mounted pressure sensor halters. A range of neural network architectures are explored and rigorous hyper-parameter searches are performed to optimise the network. The computational complexity and memory footprint of CNN models are not readily compatible with deployment on low-power processors which are both memory and energy constrained. Thus, progressive reductions of the CNN were executed with minimal loss of performance in order to address the practical implementation challenges, defining the trade-off between model performance versus computation complexity and memory footprint to permit deployment on micro-controller architectures. The proposed methodology achieves a compression of 14.30 compared to the unpruned architecture but is nevertheless able to accurately classify cattle behaviours with an overall $F1$ score of 0.82 for both FP32 and FP16 precision while achieving a reasonable battery lifetime in excess of 5.7 years.

Monitoring and classification of dairy cattle behaviours is essential for optimising milk yields. Early detection of illness, days before the critical conditions occur, together with automatic detection of the onset of oestrus cycles is crucial for obviating prolonged cattle treatments and improving the pregnancy rates. Accelerometer-based sensor systems are becoming increasingly popular, as they are automatically providing information about key cattle behaviours such as the level of restlessness and the time spent ruminating and eating, proxy measurements that indicate the onset of heat events and overall welfare, at an individual animal level. This paper reports on an approach to the development of algorithms that classify key cattle states based on a systematic dimensionality reduction process through two feature selection techniques. These are based on Mutual Information and Backward Feature Elimination and applied on knowledge-specific and generic time-series extracted from raw accelerometer data. The extracted features are then used to train classification models based on a Hidden Markov Model, Linear Discriminant Analysis and Partial Least Squares Discriminant Analysis. The proposed feature engineering methodology permits model deployment within the computing and memory restrictions imposed by operational settings. The models were based on measurement data from 18 steers, each animal equipped with an accelerometer-based neck-mounted collar and muzzle-mounted halter, the latter providing the truthing data. A total of 42 time-series features were initially extracted and the trade-off between model performance, computational complexity and memory footprint was explored. Results show that the classification model that best balances performance and computation complexity is based on Linear Discriminant Analysis using features selected through Backward Feature Elimination. The final model requires 1.83 ± 1.00 ms to perform feature extraction with 0.05 ± 0.01 ms for inference with an overall balanced accuracy of 0.83.

This study aimed to evaluate the capacity of infrared thermography (IRT) to be used for the in situ clinical assessment of udder health status across lactation in goats. Two epidemiological studies were designed, with 106 purebred Skopelos goats being prospectively monitored over one lactation in the first study and 132 goats of the same breed from a second farm being enrolled in the second, cross-sectional study. Goats were categorized based on the clinical status of each udder half (healthy, fibrotic, both fibrotic and asymmetric, asymmetric, swollen supra-mammary lymph nodes, and abscessed). Skin surface temperature values of teats (TSST) and udder halves (USST), as well as temperature variations between different types of abscesses, based on the stage of their development (superficial developed, superficial fully mature, drained), and the symmetrical sites of the non-affected udder halves, were evaluated. Mixed linear regression analyses indicated that fibrosis, particularly when accompanied by asymmetry, resulted in a decrease of 0.2, 0.3, and 0.3 °C (p < 0.05) and of 0.6, 0.9, and 1.4 °C (p < 0.001) in the maximum, mean, and minimum USST, respectively. Recording season significantly affected USST and TSST (p < 0.001), with both of them increasing between February and August. Additionally, TSST appeared to be an unreliable indicator of udder health status, while abscesses exhibited distinct temperature patterns based on their developmental stage. It is concluded that IRT has the potential to be used on site for the detection of temperature alterations associated with chronic udder health issues, yet future studies on a larger and more diverse population of goats with various udder health conditions is needed to confirm its extensive applicability under real-world conditions.

Information on the harvest date of crops can help with logistics management in the agricultural industry, planning machinery operations and also with yield prediction modelling. In this study, the determination and prediction of harvest dates for different crops were performed by applying machine learning techniques on C-band synthetic aperture radar (SAR) data. Ground truth data were provided for the Vojvodina region (Serbia), an area with intensive agricultural production, considering winter wheat, maize and soybean fields with exact harvest dates, for the period 2017–2020, including 592 samples in total. Data from the Sentinel-1 satellite were used in the study. Time series of backscattering coefficients for vertical–horizontal (VH) and vertical–vertical (VV) polarisations, both from ascending and descending orbits, were collected from Google Earth Engine. Clustering of harvested and unharvested fields was performed with Principal Component Analysis, multidimensional scaling and t-distributed Stochastic Neighbour Embedding, for initial cluster visualization. It is shown that the separability of unharvested and harvested data in two-dimensional space does not depend on the selected method but more on the crop itself. Support Vector Machine and Multi-layer Perceptron were used as classification algorithms for harvest detection, with the former achieving higher accuracies of 79.65% for wheat, 83.41% for maize and 95.97% for soybean. Finally, regression models were developed for the prediction of the harvest date using Random Forest and the long short-term memory network, with the latter achieving better results: an R2 score of 0.72, mean absolute error of 6.80 days and root mean squared error of 9.25 days, for all crops considered together.