ADQS-YOLO: Automatic Dragon Fruit Quality Classification and Sorting Mechanism Using YOLOv8n Model

The growing demand for high-quality horticultural produce, particularly dragon fruit, has emphasized the need for efficient post-harvest quality assessment systems. Manual sorting methods remain prevalent in many agricultural regions, but they are time-consuming, inconsistent, and highly dependent on human expertise. These limitations result in reduced market value, increased post-harvest losses, and difficulty in meeting export standards. To address this gap, this paper introduces ADQS-YOLO, an automatic dragon fruit quality classification and sorting mechanism powered by a fine-tuned YOLOv8n object detection model. The system was trained using a standard benchmark dataset containing labeled images of fresh and defective dragon fruits. YOLOv8n is selected for its lightweight architecture and high inference speed, making it ideal for resource-constrained environments. The model was deployed on a Raspberry Pi 4B, enabling real-time, on-device decision-making without reliance on external servers or cloud infrastructure. A working prototype has been developed, featuring a conveyor belt and mechanical sorting unit that autonomously classifies and separates fruits into fresh and defective bins. The proposed system achieved a better quality and sorting classification accuracy, demonstrating high reliability in operational conditions. Thus, ADQS-YOLO presented a cost-effective, scalable, and portable solution for automating fruit sorting, particularly suited for small- to medium-scale post-harvest facilities.