Title: CNN-based hybrid deep learning framework for human activity classification

Authors: Naeem Ahmad; Sunit Ghosh; Jitendra Kumar Rout

Addresses: Department of Computer Applications, National Institute of Technology Raipur, Raipur, India ' Department of Institute of Technology, Sikkim Manipal Institute of Technology, Sikkim, India ' Department of Computer Science and Engineering, National Institute of Technology Raipur, Raipur, India

Abstract: One of the primary goals of many defence applications is the detection and immediate response to human activities. Human activity detection has been proposed using various technologies, such as surveillance cameras and sensor-equipped wearable devices. However, these technologies are not widely adopted because of issues with accessibility, privacy, cost, and convenience. In recent years, the ability to identify and detect human actions based on the properties of wireless signals has been seriously considered. This paper proposes a CNN-based hybrid approach for detecting human activities using Wi-Fi sensors. This hybrid model combines Shallow CNN and SqueezeNet, which uses transfer learning to SVM classification model. This new model reduces the number of false negatives as much as possible. The designed hybrid model has an average of 98.67% accuracy for ten-fold cross-validation.

Keywords: convolutional neural network; CNN; deep learning; human activity; signal classification; neural networks.

DOI: 10.1504/IJSNET.2024.136697

International Journal of Sensor Networks, 2024 Vol.44 No.2, pp.74 - 83

Received: 22 Aug 2023
Accepted: 04 Nov 2023
Published online: 16 Feb 2024 *

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