DeePet: A Deep Learning Based Pet Comforting and Feeding System Using Internet of Things Technologies

This paper designs and implements a deep learning based pet comforting and feeding system, called DeePet, to achieve comprehensive pet care services through IoT technologies. The DeePet system provides functions of activity recognition, instant feeding, music/sound playing, pet soothing, and exercise encouragement. The pet activity recognition model is developed based on densely connected neural networks to accurately identify pet status (e.g., eating, anxious, drinking, or agitated). In addition, the pet can be conditionally feeded as it performs sufficient roller rotations for encouraging exercise. Furthermore, the scent of the owner can be extracted beforehand and released to alleviate the anxiety of the pet as necessary. To validate the feasibility and performance of DeePet, an Android-based prototype with the IoT device is developed. Experimental results show that DeePet outperforms existing methods and can significantly improve the accuracy of pet activity recognition.