Crop Quality Assessment Using Image Processing and Artificial Intelligence (Guardians of Farm Project)

This university capstone team developed an experimental crop-quality assessment pipeline for disease detection and leaf-health evaluation in smart-farm settings.

We first explored unsupervised visual grounding with GroundingDINO and SAM (Segment Anything Model) for leaf segmentation and RGB‑based classification. Under class imbalance, average‑RGB features proved biased and produced false classifications.

The team then combined Kaggle sources with field photographs, labelled and augmented the data in Roboflow, and replaced the initial pipeline with a supervised YOLOv8 detector.

The experimental system overlays detections and maps the ratio of healthy to diseased leaves to five project-defined health levels. In the project evaluation, the trained model achieved mAP@0.5 of 0.81 and AP of 0.846 for diseased leaves.

IoT-based monitoring remains a potential extension; it was not presented as an implemented operational integration.