EVALUATING SITTING POSTURE USING PRESSURE SENSORS: A LIGHTWEIGHT CNN APPROACH WITH PSYCHOMETRIC AND ERGONOMIC PERSPECTIVES

The seated posture is quite an important element of ergonomics at work and exercise. An accurate category can help in injury prevention and work-related health programs. Samples of 30 subjects in four sitting positions were collected, and data from 7200 samples were used to train a lightweight CNN. In the training epochs (20-50) and batch sizes (16-32), a systematic search was done. The model has been contrasted with the MobileNetV2 in terms of accuracy, precision, recall, training time and size. The custom CNN achieved a higher accuracy across all the batch sizes (93.83%-99.63%) than MobileNetV2. Training time per iteration reduced (4.12-13.37 seconds vs. 162.88-493.59 seconds), and storage requirements were also minimal 0.03 MB vs. 9.87 MB). The data collected was the same between trials. The findings indicate that the parameter tuning enhances psychometric robustness in classification. The compact CNN can be used to monitor sitting behaviour in real-time, direct ergonomic product design, prevent injuries and conduct psychologically relevant studies.