Resampling and Data Augmentation For Equines’ Behaviour Classification Based on Wearable Sensor Accelerometer Data Using a Convolutional Neural Network

Eerdekens, Anniek; Deruyck, Margot; Fontaine, Jaron; Martens, Luc; Poorter, Eli De; David, Paul A.; Joseph, Wout

doi:10.1109/coins49042.2020.9191639

Cited by 10 publications

(13 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The comparison results of our CMI-Net with three machine learning methods (i.e., NB, DT, and SVM) and two deep learning methods (i.e., CNN and ConvNet7) [ 14 , 15 ] are illustrated in Table 1 . The results revealed that the CMI-Net with softmax CE loss outperformed the machine learning algorithms with higher precision, recall, F1-score, and accuracy of 79.74%, 79.57%, 79.02%, and 93.37%, respectively.…”

Section: Resultsmentioning

confidence: 99%

“…The reason for this superior performance was the convolution and pooling operations in CNN, which could achieve automated feature learning and aggregate more complex and general patterns without any domain knowledge [ 38 ]. The other CNN-based method [ 15 ] obtained inferior precision of 72.07% and accuracy of 82.94% compared to DT and SVM. This result is consistent with the “No Free Lunch” theorem [ 39 ] because this CNN-based method [ 15 ] was developed using leg-mounted sensor data.…”

Section: Resultsmentioning

confidence: 99%

“…The other CNN-based method [ 15 ] obtained inferior precision of 72.07% and accuracy of 82.94% compared to DT and SVM. This result is consistent with the “No Free Lunch” theorem [ 39 ] because this CNN-based method [ 15 ] was developed using leg-mounted sensor data. In addition, our CMI-Net with softmax CE loss performed better than ConvNet7 [ 14 ], which obtained lower precision, recall, F1-score, and accuracy of 79.03%, 77.79%, 77.90%, and 91.27%, respectively.…”

Section: Resultsmentioning

confidence: 99%

“…The best model with the highest validation accuracy was saved and verified using test data. To evaluate the classification performance of our CMI-Net, we compared it against various existing methods, including three machine learning methods (i.e., NB, DT, and SVM) and two deep learning methods used in equine activity recognition (i.e., CNN and ConvNet7) [14,15], based on the same public dataset. Specifically, the hand-crafted features used in machine learning were the same as those used by Kamminga et al [7].…”

Section: Implementation Detailsmentioning

confidence: 99%

“…For example, feed-forward neural networks (FNNs) and long shortterm memory (LSTM) models were applied to automatically recognize cattle behaviors (e.g., feeding, lying, and ruminating) using data collected from inertial measurement units (IMUs) [12,13]. Convolutional neural networks (CNNs), which accurately capture local temporal dependency and scale invariance in signals, were developed in automated equine activity classification based on triaxial accelerometer and gyroscope data [1,14,15]. FilterNet, presented based on CNN and LSTM architectures, was adopted to classify important health-related canine behaviors (e.g., drinking, eating, and scratching) using a collar-mounted accelerometer [16].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Cross-Modality Interaction Network for Equine Activity Recognition Using Imbalanced Multi-Modal Data

Mao

Huang

Gan

et al. 2021

Sensors

View full text Add to dashboard Cite

With the recent advances in deep learning, wearable sensors have increasingly been used in automated animal activity recognition. However, there are two major challenges in improving recognition performance—multi-modal feature fusion and imbalanced data modeling. In this study, to improve classification performance for equine activities while tackling these two challenges, we developed a cross-modality interaction network (CMI-Net) involving a dual convolution neural network architecture and a cross-modality interaction module (CMIM). The CMIM adaptively recalibrated the temporal- and axis-wise features in each modality by leveraging multi-modal information to achieve deep intermodality interaction. A class-balanced (CB) focal loss was adopted to supervise the training of CMI-Net to alleviate the class imbalance problem. Motion data was acquired from six neck-attached inertial measurement units from six horses. The CMI-Net was trained and verified with leave-one-out cross-validation. The results demonstrated that our CMI-Net outperformed the existing algorithms with high precision (79.74%), recall (79.57%), F1-score (79.02%), and accuracy (93.37%). The adoption of CB focal loss improved the performance of CMI-Net, with increases of 2.76%, 4.16%, and 3.92% in precision, recall, and F1-score, respectively. In conclusion, CMI-Net and CB focal loss effectively enhanced the equine activity classification performance using imbalanced multi-modal sensor data.

show abstract

Section: Resultsmentioning

confidence: 99%