AI-Powered X-Ray Image Analysis for Automated Detection of Insect Damage and Hollow Defects in Coffee Beans

Coffee ranks among the most economically important agricultural commodities traded globally. Despite this significance, its quality is routinely affected by internal defects – particularly insect damage and hollow beans -that are undetectable through conventional surface inspection. This paper describes a quality inspection system built on Google Gemini’s multimodal vision capabilities, designed to automatically analyze X-ray images of coffee beans and identify hidden structural defects. The system runs though a Next.js web application, allowing users to upload images in real time and receive structured quality assessment reports. X-ray images were gathered, preprocessed, and sent to the Gemini Vision API with carefully designed prompts, which returned classification outputs indicating defect type, confidence level, and corrective recommendations. Testing on 60-image dataset produced an overall classification accuracy of 88.3%, with processing times under four second per image. These results show that combining large multimodal AI model with web-based deployment provides a practical, affordable, and scalable alternative to both manual destructive inspection and costly specialized equipment. This work presents a reproducible pipeline for AI-driven agricultural quality control and lays groundwork for future deep learning improvements in food safety applications.