Lightweight Edge-AI Frameworks for Coffee Bean Defect Identification and Classification

The process of examination of the quality of coffee beans is a manual process, which is also subjective and mostly inconsistent, resulting in change in grading and market value. The current study suggests the Lightweight Edge-AI Framework of in-situ coffee bean defect detection and classification due to an urgent requirement of real-time, low-cost, and dependable quality measures at the farm/ warehouse levels. The suggested solution combines MobileNetV3 in order to perform feature extraction efficiently, as well as Tiny-YOLOv8 to detect defects quickly, and optimize the solution with quantization and pruning to run on edge devices like Raspberry Pi and Jetson Nano. Experimental outcomes prove the detection accuracy of 96.8, the precision of 95.2, and the speed of inference of less than 150 ms, which proves the ability of the model to perform in real time with insufficient computation resources. The results underline the potential application of deep learning models in embedded systems, which is encouraging energy-efficient and decentralized agricultural automation based on the use of AI. This system helps smallholder farmers to do field-based bean grading, which guarantees uniformity, minimizes losses lost after harvesting, and improves transparency in the supply chain. The strategy helps to produce coffee in a sustainable way and is a scalable solution to a larger implementation of smart agriculture