Title: Real-time emotion recognition in visual art using mobile sensing terminals

Authors: Ran Wei

Addresses: School of Animation and Digital Arts, Communication University of China, Nanjing, 211172, China

Abstract: Visual art emotion recognition enables intelligent guiding services on mobile sensing devices in museums. Deploying complex recognition models on resource-limited sensing devices, however, remains challenging due to the trade-off between accuracy and efficiency. This paper proposes a lightweight dual-path neural network for efficient on-device processing. The architecture employs two parallel branches to separately analyse stylistic elements (e.g., colour, texture) and semantic content, effectively capturing artistic emotions. It integrates depthwise separable convolutions and channel attention to reduce computational cost. Evaluations on two public art datasets, ArtEmis and Metropolitan Art, show that a lightweight dual-path neural network achieves accuracies of 68.7% and 65.4%, surpassing the lightweight baseline MobileNetV3 by over 5% while containing only 0.79 million parameters. The model achieves an inference delay of 23.4 milliseconds per image on a mobile platform, demonstrating its suitability for real-time sensing applications. This work facilitates the deployment of sophisticated art analysis capabilities directly on sensor network nodes.

Keywords: art emotion recognition; mobile sensing terminal; lightweight neural network; dual-path architecture; real-time inference.

DOI: 10.1504/IJSNET.2026.154345

International Journal of Sensor Networks, 2026 Vol.51 No.2, pp.101 - 114

Received: 03 Dec 2025
Accepted: 11 Dec 2025
Published online: 23 Jun 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article