Fuzzy Neural Network Enhanced Information Fusion for Multimodel Action Recognition

With the development of hardware and communication technology, human action recognition (HAR) in the Internet of Things (IoT) environment is gradually becoming the solution to problems such as long-term healthcare, security surveillance, etc. However, HAR in IoT faces challenges due to heterogeneous, uncertain, and resource-constrained sensing conditions. To address this, we propose FIFIAR, a lightweight adaptive fuzzy neural network (FNN)-based decision fusion framework designed for image-based HAR in IoT systems. FIFIAR learns fuzzy relationships across multiple modalities (RGB, depth, IR, skeleton) to reduce decision uncertainty and support efficient, real-time inference on edge devices. Experiments on MSR Daily Activity and NTU RGB+D 120 datasets show that FIFIAR achieves 99.8% and 96.36% accuracy, respectively, outperforming conventional fusion methods and demonstrating strong potential for real-world IoT deployments.