Multi-Stage UAV-Based System for Scalable and Accurate Crop Health Monitoring

This work presents a new approach to crop quality monitoring that makes use of deep learning vision algorithms and a Single Board Computer (SBC). Our approach consists of a multi-step image processing pipeline, wherein the ROI (Region of Interest) is first identified using a specialised Vegetation determination procedure (NDVI). Furthermore, geo-coordinates enabling accurate geospatial mapping of crop quality measurements. A 5-layer CNN classifier, as well as deep learning models for leaf classification and optimised picture filtering, augmentation, and segmentation, come next. The system's real-time capabilities were assessed using an Nvidia Jetson Nano Developer kit.