Animal behavior classification via deep learning on embedded systems

We develop an end-to-end deep-neural-network-based algorithm for classifying animal behavior using accelerometry data on the embedded system of artificial intelligence things (AIoT) device installed in a wearable collar tag. The proposed jointly performs feature extraction and classification utilizing set infinite-impulse-response (IIR) finite-impulse-response (FIR) filters together with multilayer perceptron. utilized IIR FIR can be viewed as specific types recurrent convolutional neural network layers, respectively. evaluate performance via two real-world datasets collected from grazing cattle. results show that offers good intra- inter-dataset accuracy outperforms its closest contenders including state-of-the-art convolutional-neural-network-based time-series algorithms, which are significantly more complex. implement tag's AIoT to perform in-situ behavior. achieve real-time inference without imposing any strain available computational, memory, or energy resources system.