Acta Scientiarum Polonorum Technologia Alimentaria

Background. Accurate detection of wheat freshness is important for ensuring the quality and safety of wheat products, thereby protecting the health and interests of consumers.
Material and methods. This study integrates biophoton emission technology with advanced deep learning frameworks to transform the process of wheat freshness assessment. Leveraging the powerful feature extrac­tion capabilities of the ResNet architecture, we employ a multi-scale framework integrated with the Gaussian Context Transformer (GCT) attention mechanism. The MS-GCT-ResNet method presents a groundbreaking approach that not only enhances the accuracy and efficiency of wheat freshness discrimination but also dem­onstrates the potential for combining biophysical phenomena with cutting-edge Artificial Intelligence (AI) technologies for precision agriculture and food quality control.
Results. This model enhances detection accuracy and adaptability, offering a powerful technique for the rap­id and precise evaluation of wheat freshness. Its effectiveness is validated using years of wheat sample data. Based on the experimental results, MS-GCT-ResNet achieves a recognition accuracy of 93.6%. Compared with traditional CNN and ResNet models, the recognition accuracy increases by 1.9% and 1.1%, respectively.
Conclusion. MS-GCT-ResNet is a highly promising, non-invasive, and efficient technological advancement capable of quickly and accurately assessing wheat freshness. This method holds immense potential to revo­lutionize the agriculture and food processing industries.