Application of hyperspectral imaging for detection and mapping of small patch clubroot infestations in commercial canola fields

Term: 3 years, ending September 2024
Status: Complete
Researcher(s): David Halstead, Saskatchewan Polytechnic; Bruce Gossen, AAFC; Mary Ruth McDonald, Guelph University
SaskCanola Investment: $51,226
Total Project Cost: $140,900
Funding Partners: Agriculture Development Fund

Grower Benefits:

  • Remotely piloted aircraft systems (RPAS) equipped with hyperspectral cameras offer an ideal platform for clubroot detection especially during canola flowering when field surveys are otherwise difficult.

  • Research results indicate clubroot affected canola offers a distinct spectral signature for small patch detection with successful classification of clubroot affected areas in >90% of the small patches tested in producer’s fields and research facilities throughout Alberta and Saskatchewan.

  • The strength of the analysis portion of this study is owing to an adaptive approach to supervised classification model development, the power of machine learning, and the abundance and diversity of data provided by research collaborators.

  • Additional refinement of these research results are possible.

  • This research should help facilitate early detection of clubroot and contribute to significant cost savings before the disease is permitted to spread in individual farmers fields and more broadly throughout the prairies.

Project Summary

Researchers had a couple of objectives to complete during this project; the first was to identify readily applied diagnostic features for mapping small patch clubroot distributions using hyperspectral data, and to develop a diagnostic tool. Second, they wanted to refine and validate the diagnostic tool for identifying small patches of clubroot infestations.

Twenty-three RPAS hyperspectral flights took place in Alberta and Saskatchewan over 21 sites between 2021 and 2023. Each flight involved a rotary-wing DJI M600 Pro with a Headwall Nano-hyperspec camera (figure 1). Each survey targeted an area of about one hectare at entrances or high traffic areas within the field. Camera calibration was done ahead of each flight. Flight altitudes were maintained at 40 to 50 m and ground sampling distances were calculated between 14 and 15 cm.

 

Follow-up ground sampling of imaged canola fields (concentrated to the targeted areas within the field) was conducted by researchers after swathing to determine the extent of clubroot infestations. Severity of positive infections were assessed by applying a modified disease severity index for 10 randomly selected plants within a 3 x 3 m area in the clubroot patch.

Once the data was collected and processed, 450-pixel regions of interest were extracted for further analysis. The goal of this analysis was to isolate spectral regions of interest and identify spectral bands likely to have predictive power and apply a classification approach to differentiate between clubroot infected and non-infected pixels. Several different machine learning models were tested. Datasets from the best model were used to create percent matches of clubroot infected or non-infected pixels. Variation in clubroot image classification was then validated at two locations in Alberta and Saskatchewan.

 

This research showed that drones equipped with hyperspectral cameras are valuable for disease detection during flowering when disease surveys are difficult. This project only looked at flowering canola, so further research is needed to expand detectability to earlier and later times within the growing season. Cross validation will also remain important to ensure accuracy.

Figure 1. Rotary-wing DJI M600 Pro with a Headwall Nano-hyperspec camera.

Previous
Previous

A Sustainable Future for the Saskatchewan Soil Information System (SKSIS)

Next
Next

SCAP CCC Canola AgriScience Cluster 2023-2028