Optimizing cereal genetic selection

Client

Sollio Agriculture

Collaborator

Agrinova

Funding

Research and Transfer Assistance Program (RTAP) – Technological Innovation Component

Objectives

This project is testing a methodology for collecting and processing aerial images to evaluate genetic selection criteria specific to the cultivation of oats (Avena sativa L.). The aim is to create an automated platform for rapid and accurate analysis of crop growth characteristics, with reproducible protocols and artificial intelligence models capable of automatically interpreting drone data.

More specifically, the project aims to:

1. Experiment with a methodology for collecting images of experimental plots by drone (flight height, precision and characteristics of sensors, climatic conditions and site instrumentation required).
2. Experiment with an image processing methodology for collecting data specific to genetic selection (height and maturity date of oats and maturity date of soybeans).
3. Validate the correlation between conventionally collected data and data collected using the technological tool.
4. Ensure technology transfer to the industrial partner by automating image processing.

Methodology

The methodology consists of three key stages:

1. Acquisition of imagery by drone, coordinated with in situ measurements.
2. Image processing and analysis using deep learning.
3. Process automation.

Image acquisition by drone is carried out at four strategic periods (bare ground, post-emergence, mid-season and pre-harvest), using high-resolution multispectral and RGB cameras.

Field validation is carried out simultaneously with the drone flights, with precise manual measurements of plant height, lodging (0-9 scale) and maturity date (90% of ears/yellow pods). Data processing involves the creation of orthomosaics, digital surface models and the calculation of vegetation indices.

Analyses use advanced machine learning techniques, including convolutional neural networks (CNNs), random forests and gradient boosting, to develop powerful and robust detection models.

Equipment

• Drone: DJI Matrice 600 Pro
• Cameras: Multispectral MicaSense RedEdge-P and high-resolution RGB
• Phase One IXM-50
• GNSS: Leica GS18I
• Deep learning server: Nvidia DGX A100
• Photogrammetric and GIS software: Agisoft Metashape, Open Drone Map, QGIS
  Deep learning technologies: PyTorch and Raster Vision

Field

Resources management