A self-developed electronic nose system combines data enhancement and multi-branch kernels channel attention to identify the gas information of industrial polypropylene

With growing concerns regarding the air quality inside vehicles, the emission of pungent gases from polypropylene (PP), the primary material utilized in car interiors, at high temperatures is becoming increasingly worrisome. In this study, combining a self-developed electronic nose (e-nose) system with proposed gas information classification method to identify the gas information of PP. First, based on the self-developed e-nose system, PP volatile gases under different temperature gradients are detected. Second, given the limited number of samples in the gas information acquired by the e-nose, we propose a data augmentation method to expand the dataset. Third, a multi-branch kernels channel attention (MBKCA) is proposed to selectively incorporate multi-scale information, enabling effective adaptive feature attention. Finally, the combination of data augmentation with a multi-branch kernel channel attention network (MBKCA-Net) improves the classification accuracy and stability of industrial PP gas. The results demonstrate that the method achieves the best classification accuracy of 98.70%, and the best F1-score of 0.9868.