Edge-Based Fruit Quality Assessment Using an AI-Driven Electronic Nose

Abstract

Ensuring food safety and minimizing waste remain key challenges across domestic and industrial contexts. This paper presents a smart fruit spoilage detection system that combines advanced gas sensing with deep learning for real-time monitoring and automated decision-making. The system employs an electronic nose (E-nose) with nine gas sensors to identify early indicators of fruit spoilage, thereby improving hygiene and reducing health risks. To achieve accurate and efficient classification, three deep learning models Artificial Neural Networks (ANN), Long Short-Term Memory networks (LSTM) and hybrid CNN-LSTM architectures were evaluated. The LSTM achieved the best results with 97.03% accuracy, a 2.57 MB model size,and an inference time of 133.2 ms. To overcome the constraints of edge computing, several model compression techniques were explored. Among them, the knowledge distillation approach achieved the best trade-off, reducing the model size by 78.3% while incurring only a 0.71% accuracy loss. The optimized model was subsequently deployed on an ARM-based microcontroller integrated with nine gas sensors, achieving an overall accuracy of 90% and enabling low-latency, real-time inference in IoT environments. Compact, efficient and scalable, the proposed system provides a complete end-to-end solution from sensing to embedded deployment for intelligent food safety monitoring applications.