Work

Improving food safety: Synthetic data augmentation for accurate mushroom species identification in complex environments

Year: 2025

Type: article

Abstract: Growing global demand for mushrooms in the agricultural, pharmaceutical, and food sectors raises pressing concerns for food safety, as inadequate identification can lead to the inadvertent inclusion of toxic species. Although deep learning offers promising solutions for large-scale classification and quality control, its effectiveness is often limited by the scarcity of high-quality training datasets and the inherently complex morphological traits of mushrooms. To address these challenges, this study proposes a diffusion-model-based synthetic augmentation strategy that preserves essential morphological features while diversifying training samples. By generating realistic images, this approach significantly improves classification accuracy, especially for underrepresented species, and achieves superior results across various image quality indicators. Furthermore, a Top-k recommendation mechanism ranks the most likely classifications, bolstering recognition of rarer species and reducing food safety risks linked to misidentification. Experiments on 110 mushroom species demonstrated a 13.51% increase in mean recall rate across eight deep learning models, with Top-3 and Top-5 recall rates further improved by 7.56% and 6.79%, respectively. These findings highlight the practical utility of diffusion-based augmentation in enhancing model robustness, ensuring more reliable monitoring of mushroom cultivation processes, and mitigating the likelihood of toxic mushrooms entering the food chain. Overall, this study underscores the critical importance of leveraging synthetic data generation to advance mushroom recognition, supporting safer and more sustainable food systems.

Source: Applied Food Research

DOI: 10.1016/j.afres.2025.101039