AI-driven deep learning framework for shelf life prediction of edible mushrooms

Abstract

Fresh edible mushrooms are highly perishable, prone to microbial contamination, and have a limited shelf life, making it essential to assess their freshness to ensure food safety and minimize waste. Additionally, the freshness of mushrooms has a significant impact on consumer health and market prices. This study introduces and evaluates a computer vision-based application utilizing convolutional neural networks (CNNs) to estimate the shelf life of three widely consumed mushroom varieties: white button, shiitake, and oyster, stored at 4 ± 1 °C. To improve classification accuracy and reduce training costs, transfer learning was employed to fine-tune CNN models, including EfficientNet, NASNetLarge, ResNet-50, Inception-V3, and MobileNet-V2. ResNet-50 demonstrated the highest performance for white button and oyster mushrooms, with overall accuracies of 94.10 % and 89.11 %, respectively, processing 1960 images in 6.39 min for the former and 8.70 min for the latter. For shiitake mushrooms, MobileNet-V2 showed superior performance with an accuracy of 86.36 % and a processing time of 5.73 min. Integrating digital imaging with CNN methods offers a reliable and efficient approach for rapid and precise evaluation of mushroom freshness. This technology can significantly improve monitoring during the postharvest storage and distribution stages, facilitating informed decision-making and timely actions to reduce spoilage.