Poplar Optimization Algorithm-Driven Hybrid Siamese Top-Down Neural Networks for Accurate Human Activity Recognition in IoT Networks
S. Prabagaran, Ashok Kumar Bandla, R J Venkatesh, M. Jeba Malar · 2024
Human Activity Recognition (HAR) recognizes human actions including sitting, running, and walking using sensor data. This data is gathered in real-time from several linked devices when it is integrated with the Internet of Things (IoT), enabling ongoing monitoring and analysis. The lot of techniques is implemented to recognize human activities. But the existing methods have lot of disadvantages such as high error rate and low accuracy. To overcome the before mentioned problem, Hybrid Siamese Top-Down Neural Networks using Poplar Optimization Algorithm (Hyb-STDNN-POA) is proposed for recognizing human activity with high accuracy. In this input data is taken from Extrasensory dataset.. To reduce noise in the input data, Shape-Aware Mesh Normal Filtering (SAMNF) is proposed. Following that, the pre-processed images undergo feature extraction using Quantized Discrete Haar Wavelet Transform (QDHWT). After that feature fusion, segmentation and classification are done using Hybrid Siamese Top-Down Neural Networks (Hyb-STDNN) and optimized using Poplar Optimization Algorithm (POA) for human activity recognition demonstrating superior efficiency and accuracy. The efficiency of the proposed Hyb-STDNN-POA is analyzed using Extrasensory dataset and attains 99.5% accuracy, 99.4 % recall and attains better results in comparison with the existing techniques. The outcomes of the proposed technique showed that it could improve the evaluations ability of the computerized human activity recognition method.