LLM-Assisted Strawberry Quality Assessment
Segmentation, quantitative image analysis, retrieval, and natural-language reporting for smart farms.
Problem
Conventional quality assessment stops at simple image classification or binary decisions, offering no detailed interpretation such as harvest criteria or disease-guidance references. Fixed classifiers also adapt poorly when criteria change across cultivars and environments.
Approach
Pipeline — YOLOv11-seg segmentation → OpenCV quantitative analysis → RAG document retrieval → LLM interpretation and report generation.
- Segmented individual strawberries with YOLOv11-seg to isolate analysis targets.
- Quantified ripeness (color), size, and potential disease indicators with OpenCV.
- Converted measurements into natural-language prompts and retrieved quality and disease-guidance documents through a RAG pipeline.
- An LLM interprets the retrieved references to generate a natural-language report on harvest readiness and recommended actions.
The goal was document-grounded interpretation that adapts by swapping reference documents — instead of retraining a fixed binary classifier.
YOLOv11-seg · OpenCV · RAG · LLM · PyTorch · Python
Presentation: In Gon Kim and Soo Young Shin, “An LLM-Based Automated Strawberry Quality Assessment System for Smart Farming Environments,” 35th Joint Conference on Communications and Information (JCCI 2025), 2025.