Estimating Cannabis Flower Maturity in Greenhouse Conditions using Computer Vision

This repository contains the implementation of our research on automated cannabis flower maturity estimation using computer vision and deep learning techniques.

Publication

Estimating Cannabis Flower Maturity in Greenhouse Conditions Using Computer Vision Etay Lorberboym, Silit Lazare, Polina Golshmid, Guy Shani Published in Agriculture (MDPI, Q1), Volume 16, Issue 4, 2026

Read the full paper | DOI: 10.3390/agriculture16040460

Table of Contents

1. Motivation
2. Methodology
3. Results
4. Project Structure
5. Dataset
6. Installation

Motivation

As cannabis flowers mature, trichomes shift from clear to milky to amber — a key indicator of cannabinoid content. Conventional assessment relies on manual loupe inspection, which is subjective, time-consuming, and inconsistent across observers.

(Top) Schematic of a cannabis flower showing stigma and trichome locations. (Bottom) Trichome color progression: clear → milky → amber.

We replace manual inspection with smartphone-based macro imaging and computer vision to provide objective, reproducible maturity assessment.

(Left) Traditional loupe vs. smartphone macro imaging in greenhouse. (Right) Laboratory imaging setup.

Methodology

Our framework implements a dual-path pipeline for cannabis maturity assessment:

(Upper path) Trichome detection and classification. (Lower path) Stigma segmentation and color ratio computation. Both paths aggregate results to produce flower-level maturity indicators correlated with HPLC-measured cannabinoid concentrations.

Results

Trichome Detection Across Flowering Stages

Bounding boxes indicate detected trichomes with color-coded classification: blue (clear), orange (milky), and purple (amber).

Trichome detection results across the flowering period (Cultivar 805). Left: Early (DAF 51). Center: Mid (DAF 65). Right: Late (DAF 79).

Stigma Segmentation Across Flowering Stages

Blue overlays indicate detected stigma instances (bounding boxes and segmentation masks).

Stigma segmentation results across the flowering period (Cultivar 805). Left: Early (DAF 51). Center: Mid (DAF 65). Right: Late (DAF 79).

Stigma Color Classification

Each panel shows the original extracted stigma alongside its classified color map, where green pixels indicate immature tissue and orange pixels indicate mature tissue.

Stigma color classification using 1-NN pixel-level analysis (Cultivar 805). The progressive shift from green-dominant to orange-dominant reflects advancing maturity.

Project Structure

src/
├── common/                        # Shared utilities (detection, I/O, logging, metrics, visualization)
├── config/                        # Configuration (paths, settings)
├── classification/                # Trichome classification (FastAI-based training)
├── segmentation/                  # Detection & segmentation models
│   ├── handlers/                  # Detectron2, SAM, Ultralytics wrappers
│   ├── evaluation/                # Model evaluation (per-framework evaluators)
│   └── train_scripts/             # Training & tuning scripts
├── stigmas_segmentation/          # Stigma segmentation + color analysis
├── pipelines/                     # End-to-end processing pipelines
│   └── end_to_end/                # Full trichome & stigma analysis pipelines
├── data_preparation/              # Image I/O, patching, sharpness filtering
├── data_analysis/                 # Cannabis maturity correlation experiments
├── annotations/                   # Annotation handling & Segments.ai integration
├── app/                           # Streamlit web application
└── utils/                         # Additional utilities

Dataset

Over 14,000 images collected in partnership with RCK Science-Based Cannabis Genetics across the final weeks of cannabis flower growth using an iPhone 14 Pro with 10X macro lens, under both greenhouse and laboratory conditions. Annotation was performed via Segments.ai.

Installation

git clone https://github.com/elorberb/estimate-cannabis-maturity-with-cv.git
cd estimate-cannabis-maturity-with-cv
make setup

make setup runs uv sync and configures git hooks (branch name enforcement). See Makefile for other targets (lint, format, test).

Citation

If you use this code or our methodology in your research, please cite:

@Article{agriculture16040460,
AUTHOR = {Lorberboym, Etay and Lazare, Silit and Golshmid, Polina and Shani, Guy},
TITLE = {Estimating Cannabis Flower Maturity in Greenhouse Conditions Using Computer Vision},
JOURNAL = {Agriculture},
VOLUME = {16},
YEAR = {2026},
NUMBER = {4},
ARTICLE-NUMBER = {460},
URL = {https://www.mdpi.com/2077-0472/16/4/460},
ISSN = {2077-0472},
DOI = {10.3390/agriculture16040460}
}

License

This project is for research purposes. Please contact the authors for commercial use inquiries.