DTLN: Deep Transformer Learning Network for Multiwavelength Strain Classification in Sports Wearables

Wearable Piezoresistive pressure sensors measure the electrical resistance change under mechanical strain, but existing methods struggle with heterogeneous wavelength classification due to amplitude overlaps and intensity variations. Additionally, changes in electrical resistance caused by mechanical strain are utilized for detecting and mapping Stress and strain during physical activity, as well as identifying surface roughness and its impact on the athlete. This article introduces a Deep Transformer Learning Network (DTLN) based solution that combines transformer-based self-attention mechanisms with deep learning classifications to enhance strain detection accuracy. The DTLN addresses wavelength signal differences based on ascending intensities to compute the maximum difference between the impacting outputs and also classifies negative and positive differences. These outputs are correlated with the maximum strain acceptance factors to ensure that identified impacts on the sportsperson are accurately assessed. Therefore, the transformer learning classification is updated based on the previously trained difference to identify various wavelength intensities. Thus, the proposed method detects strain impacts over the sportsperson’s practice sessions. The proposed method improved strain detection by 13.2% and difference classification by 14.01%, reducing the error factor by 14.53% for average pressure. This work advances wearable sensor analytics by providing a scalable and accurate method for real-time strain impact assessment in athletic applications.