Title: Research on campus safety behaviour recognition based on data enhancement and multi-scale improved CNNs

Authors: Wei Zhao; Hui Xu

Addresses: Department of Information Security Technology, Jilin Police College, Changchun, 130117, China ' Changchun Humanities and Sciences College, Institute for Intelligent Elderly Care, Changchun, 130117, China

Abstract: This study tackles the issue of low accuracy in campus safety behaviour recognition methods by proposing a model that integrates a data enhancement algorithm with an improved multi-scale convolutional neural network (MSCNN). The newly developed data-enhanced MSCNN model was tested against two commonly used recognition models across various regions and identified safety behaviours with significantly greater accuracy. Remarkably, it achieved a 96% accuracy rate for recognising behaviours among students at a specific university and reached 97% for students across multiple colleges under similar conditions. The model demonstrated minimal errors in identifying diverse safety behaviours, highlighting its effectiveness in accurately recognising campus safety behaviours. Overall, the experimental results confirm that this model provides a robust tool for evaluating and analysing campus safety behaviours more effectively.

Keywords: data augmentation; CNN; multi-scale; campus safety behaviour; improved MSCNN model.

DOI: 10.1504/IJISTA.2026.153193

International Journal of Intelligent Systems Technologies and Applications, 2026 Vol.24 No.2, pp.119 - 135

Received: 28 Nov 2024
Accepted: 25 Jul 2025
Published online: 29 Apr 2026 *

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