Enhancing human activity recognition with lightweight CNN models and integrated blocks

Human activity recognition (HAR) is crucial for health tracking, fitness monitoring, and fall detection systems. Recently, convolutional neural network (CNN) models have been proven to be highly effective for HAR tasks. This study aimed to enhance HAR performance by integrating specific architectural improvements, namely identity, convolutional, and bottleneck blocks, into lightweight CNN models. To evaluate the effectiveness of these enhancements, two data sets were utilized: HAR using smartphones data set version 1.0 (UCI-HAR) and wireless sensor data mining activity prediction data set version 1.1. The results indicated that the convolutional and identity block models outperformed the original lightweight CNN model on both data sets. The proposed models strike a balance between high performance and computational complexity, thereby making them suitable for real-world applications. The findings of this study contribute to the field of HAR and provide valuable insights for improving the recognition and classification of human activities.