Integrating Criss-Cross Attention Mechanism for Origin Traceability of Peanuts Using Electronic Nose Technology

Significant variations in quality are observed among peanuts sourced from different regions. This study proposes a rapid, non-destructive method for assessing peanut quality. Gas information from peanuts sourced from seven different regions is collected using an electronic nose (e-nose) system. A method combining Criss-Cross Attention (CCA) and convolutional computation is designed to extract features of gas information in terms of time series and cross-sensitivity. Lastly, global features are constructed through Recurrent Criss-Cross Attention (RCCA) to address the limitation of convolution in representing global information. The RCCA-Net achieves an accuracy of 97.85%, precision of 97.92%, and recall of 97.78%. In summary, RCCA-Net offers a practical technical solution for quality control in the food industry by successfully identifying quality variations between peanuts sourced from various geographical areas.