Novel hybrid optimization based adaptive deep convolution neural network approach for human activity recognition system

Human Activity Recognition (HAR) has emerged as a crucial area of research in computer vision, signal processing, and machine learning, enabling to automatically identify human activities from sensor data. Although various models are developed for automatic HAR, they face challenges like limited accuracy, large training time, etc. To resolve these issues, we developed an innovative hybrid strategy using the combination of hybrid meta-heuristic optimization algorithms and deep learning. The developed work commences with the collection of HAR database and the collected database was pre-processed to enhance image quality. Further, segmentation was performed using optimized Otsu’s methodology to extract the most significant features. Then, an innovative classification model was developed using the combination of Improved Spotted Hyena algorithm with Seagull Optimization Algorithm (ISHO-SOA) and Adaptive Deep Convolutional Neural Networks (ADCNN) for identifying and classifying human activities. The novelty of the algorithm lies in the seamless integration of different meta-heuristic optimization algorithms and deep learning for HAR. In the proposed algorithm, the ADCNN was trained using the database to classify human activities, while the ISHO-SOA for refining the hyperparameters of ADCNN to optimize its training and improve classification performances. The proposed algorithm was modeled in MATLAB software, and validated using HAR dataset from the Kaggle database. The experimental results depict that the developed algorithm achieved higher accuracy of 99.84%, and greater precision of 99.90%. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.