Search

The detection of pig behavior helps detect abnormal conditions such as diseases and dangerous movements in a timely and effective manner, which plays an important role in ensuring the health and well-being of pigs. Monitoring pig behavior by staff is time consuming, subjective, and impractical. Therefore, there is an urgent need to implement methods for identifying pig behavior automatically. In recent years, deep learning has been gradually applied to the study of pig behavior recognition. Existing studies judge the behavior of the pig only based on the posture of the pig in a still image frame, without considering the motion information of the behavior. However, optical flow can well reflect the motion information. Thus, this study took image frames and optical flow from videos as two-stream input objects to fully extract the temporal and spatial behavioral characteristics. Two-stream convolutional network models based on deep learning were proposed, including inflated 3D convnet (I3D) and temporal segment networks (TSN) whose feature extraction network is Residual Network (ResNet) or the Inception architecture (e.g., Inception with Batch Normalization (BN-Inception), InceptionV3, InceptionV4, or InceptionResNetV2) to achieve pig behavior recognition. A standard pig video behavior dataset that included 1000 videos of feeding, lying, walking, scratching and mounting from five kinds of different behavioral actions of pigs under natural conditions was created. The dataset was used to train and test the proposed models, and a series of comparative experiments were conducted. The experimental results showed that the TSN model whose feature extraction network was ResNet101 was able to recognize pig feeding, lying, walking, scratching, and mounting behaviors with a higher average of 98.99%, and the average recognition time of each video was 0.3163 s. The TSN model (ResNet101) is superior to the other models in solving the task of pig behavior recognition., Competing Interests: The authors declare no conflict of interest.

For both pigs in commercial farms and biological experimental pigs at breeding bases, mounting behaviour is likely to cause damage such as epidermal wounds, lameness and fractures, and will no doubt reduce animal welfare. The purpose of this paper is to develop an efficient learning algorithm that is able to detect the mounting behaviour of pigs based on the data characteristics of visible light images. Four Göttingen minipigs were selected as experimental subjects and were monitored for a week by a camera that overlooked the pen. The acquired videos were analysed and the frames containing mounting behaviour were intercepted as positive samples of the dataset, and the images with inter-pig adhesion and separated pigs were taken as negative samples. Pig segmentation network based on Mask Region-Convolutional Neural Networks (Mask R-CNN) was applied to extract individual pigs in the frames. The region of interest (RoI) parameters and mask coordinates of each pig, from which eigenvectors were extracted, could be obtained. Subsequently, the eigenvectors were classified with a kernel extreme learning machine (KELM) to determine whether mounting behaviour has occurred. The pig segmentation presented considerable accuracy and mean pixel accuracy (MPA) with 94.92% and 0.8383 respectively. The presented method showed high accuracy, sensitivity, specificity and Matthews correlation coefficient with 91.47%, 95.2%, 88.34% and 0.8324 respectively. This method can be an efficient way of solving the problem of segmentation difficulty caused by partial occlusion and adhesion of pig bodies, even if the pig body colour was similar to the background, in recognition of mounting behaviour.

The success of deep learning-based bearing fault diagnosis depends on (a) training data and test data following the same data distribution; (b) a mass of labeled fault data being available. However, the working conditions of the bearings is changing, which leads to a difference in data distribution, and it is very laborious to obtain a mass of labeled data with fault information. To address these issues, a domain adaptation-based deep feature learning method with a mixture of distance measures for bearing fault diagnosis is proposed. First, the noise in vibration signal is filtered by wavelet packet decomposition and reconstruction. Then, frequency slice wavelet transform is used to transform the reconstructed signal into a two-dimensional time–frequency image. Furthermore, a domain-adaptative deep neural network based on ResNet50 is used for bearing fault diagnosis under different working conditions. A mixture of distance measures is used to minimize the distribution discrepancy between source and target domains. The bearing datasets provided by Case Western Reserve University and Paderborn University verify the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

The statistical data of different kinds of behaviors of pigs can reflect their health status. However, the traditional behavior statistics of pigs were obtained and then recorded from the videos through human eyes. In order to reduce labor and time consumption, this paper proposed a pig behavior recognition network with a spatiotemporal convolutional network based on the SlowFast network architecture for behavior classification of five categories. Firstly, a pig behavior recognition video dataset (PBVD-5) was built by cutting short clips from 3-month non-stop shooting videos, which was composed of five categories of pig's behavior: feeding, lying, motoring, scratching and mounting. Subsequently, a SlowFast network based spatiotemporal convolutional network for the pig's multi-behavior recognition (PMB-SCN) was proposed. The results of the networks with variant architectures of the PMB-SCN were implemented and the optimal architecture was compared with the state-of-the-art single stream 3D convolutional network in our dataset. Our 3D pig behavior recognition network showed a top-1 accuracy of 97.63% and a views accuracy of 96.35% on the test set of PBVD and a top-1 accuracy of 91.87% and a views accuracy of 84.47% on a new test set collected from a completely different pigsty. The experimental results showed that this network provided remarkable ability of generalization and possibility for the subsequent pig detection and behavior recognition simultaneously. [ABSTRACT FROM AUTHOR]

Wind power forecast (WPF) plays a crucial role in the reliable and safe operation of power systems. Existing wind power prediction methods mainly focus on the overall trend information, often neglecting the performance of some typical intervals that can significantly impact the reserve requirements of power system, e.g. , unexpected ramp-up and ramp-down, local peak interval (LPI), and so on. Therefore, the primary focus of this paper is to explore techniques for enhancing the accuracy of wind power during the LPI. Considering that the points during the LPI are influenced by dynamic ramp characteristics, with the steepness and magnitude of ramps influencing the range of LPI, this paper proposes a novel focused-LPI model by incorporating dynamic ramp considerations. Specifically, beyond the tradition forecasting model, the proposed model encodes the information of ramp occurrence in wind power data based on a positional encoder; subsequently, the proposed model not only outputs standard predictions within a prediction range but also identifies the presence of ramps within these predictions. In addition, this paper employs a dynamically weighted loss function to optimize the proposed multi-task model. The proposed model is applied for various network architectures, and the results confirm that the proposed model is widely suitable for various basis network architectures. Specifically, the R-squared (R2) of the proposed model for the overall trend can exist a little error compared to the initial basic network architectures; the Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Relative Error (MRE) of the proposed model for predicting the LPI points can outperform the initial basis network architecture by average enhancements of 7.10%, 2.66%, and 3.46%, respectively. In addition, compared to existing state-of-the-art (SoTA) models in the LPI points, the proposed model obtains improved performance with 14.43% for MAE and 11.19% for RMSE, respectively, presenting its substantial performance enhancement. • Multi-tasks wind power forecast model enhances the accuracy of local peak interval. • Dynamic weighted loss function with learnable parameter balance the multi-tasks model. • Accuracy of local peak interval is improved by an average of 12.8% than SoTA models. • The proposed model outperforms the initial basis network architecture by 4.4%. [ABSTRACT FROM AUTHOR]