AI / COMPUTER VISION

Drone Phenomics

Precision crop & weed segmentation workflow using YOLOv11 and canopy analysis models.

01 // Overview & Problem Statement

An automated computer vision script designed to processes high-resolution drone photography to extract canopy coverage and map crop vs. weed density in fields.

THE CHALLENGE:

Traditional field monitoring requires walking row-by-row, which is labor-intensive and fails to scale. Immediate weed alerts are necessary to optimize pesticide usage.

02 // Methodology & Workflow Pipeline

Analyzed drone imaging data sets by annotating crop zones in Roboflow, training custom YOLOv11 models, and running inference scripts to calculate agricultural indices.

Pipeline Stages:
1UAV Image Stitching
2Roboflow Data Augmentation
3YOLOv11 Object Localization
4Canopy Extraction
5Density Heatmapping

03 // The Solution & My Contribution

Developed the image annotation workflows, configured Databricks pipelines for model execution, and wrote post-processing scripts to calculate canopy cover percentages.

MY SPECIFIC ROLE:

Wrote python parser to convert YOLO output bounding boxes into vegetation density maps.

Technical Roadblock & Resolution

Varying sunlight shadows skewed early greenness index masks. Solved by transforming color bands to HSV space and applying adaptive threshold filters.

LESSONS LEARNED

Data labeling quality dominates model performance. Standardizing annotations is vital in variable outdoor fields.

FUTURE ROADMAP

  • Incorporate multispectral image band processing (NDVI extraction).
  • Deploy model on localized edge hardware for drone-mounted real-time spraying trigger signals.

Project Dossier

Domain DomainAI / COMPUTER VISION
My ContributionSystem Architecture & Dev
Repository & Live Link

Evaluation Results

Identified crop rows and weed infestations from high-altitude views.

Accurately extracted vegetation canopy coverage fractions for water-use estimation.