Enhanced Performance in Real-Time Fruit Classification Using CatBoost on Embedded Systems

This paper presents an innovative edge-based fruit ripeness detection system that leverages ambient sensor data and machine learning algorithms to accurately classify fruit ripeness in real-time. The proposed system utilizes a Raspberry Pi 4B as the edge device, equipped with temperature, humidity, volatile organic compounds (VOCs), and carbon dioxide (CO2) sensors. The collected sensor data undergoes preprocessing and is then fed into a CatBoost machine learning model, optimized for efficient operation on the Raspberry Pi. The trained model achieves an impressive overall accuracy of 98.9% in classifying fruit ripeness into categories such as unripe, ripe, and spoiled.The system's performance is validated through real-time data collection, model training, and prediction on the Raspberry Pi, confirming its effectiveness in providing precise fruit classification. Additionally, the system incorporates an RGB LED for visual feedback and integrates with AWS IoT Core for seamless data transmission and alert notifications. This research addresses the critical challenges of food waste in agriculture by offering a sustainable and efficient approach to fruit quality assessment. The proposed edge-based system demonstrates the potential for automated fruit ripeness detection, enhancing agricultural productivity and promoting sustainable practices through efficient resource utilization and waste reduction.