Novel Two-Level Neural Network Architecture with Dual-Stage Feature Fusion for Human Activity Recognition

A new framework for human activity recognition employs a two-level network architecture with dual-stage feature fusion. This approach strategically integrates multiple neural networks to leverage their complementary strengths. Human activity recognition involves identifying human actions using sensor data. Neural networks like convolutional neural networks, long short-term memory networks, and their hybrid combinations have shown exceptional performance across research domains. The structural arrangement of these components critically influences overall performance. Developing multilevel individual or hybrid models for this purpose requires careful integration. The dual-stage fusion includes late fusion, which combines outputs from the first network level, and intermediate fusion, integrating features from both first and second levels. Researchers evaluated 15 different network architectures of CNN. This study explores how these integrations can enhance recognition capabilities.