A Deep Learning and IoT-Based Dual-Stage System for Detecting Fruit and Vegetable Spoilage in Smart Refrigerators

It is critical to develop sustainable production and consumption methods to ensure a sustainable future and a livable world. Economic development and sustainable living can be achieved by minimizing environmental and household waste and by using food resources efficiently. Artificial intelligence, computer vision, data processing, and integrated systems offer the opportunity to develop such smart solutions. In this study, a Raspberry Pi-based smart refrigerator module was designed and implemented for the early detection of spoilage in fruits and vegetables. Fruits and vegetables that start to rot release various gases into the surrounding environment. Based on this, the proposed system uses a two-stage verification method. In the first stage, the spoilage of fruits and vegetables in refrigerators is detected by gas sensors. When the gas sensors detect spoilage, the second stage is triggered; images of fruits and vegetables are classified using CNN-based models, including ResNet50, DenseNet201, InceptionV3, and VGG16. If spoilage is confirmed, a notification is sent to the designated user. The integration of gas sensing with deep learning–based image classification constitutes the main novelty of the proposed system, enabling more reliable and early detection compared to single-stage approaches. Moreover, extensive classification experiments were carried out on a benchmark dataset containing 12,000 images across 20 classes. Fine-tuning and hyperparameter optimization were performed on all CNN models, with ResNet50 achieving the highest accuracy of 98.00%. This performance surpasses results reported in some of the earlier studies on the same dataset. Given its capabilities, the proposed prototype could be widely implemented in both existing and next-generation refrigerators.