Creating Predictive Weed Emergence Models Using Repeat Photography and Image Analysis.

Area of Science:

• Agricultural Science
• Plant Science
• Computer Vision

Background:

• Weed emergence models are crucial for automated weed control.
• Current data collection methods, like seedling counts, are labor-intensive and time-consuming.
• Developing more efficient data collection techniques is essential for improving weed management models.

Purpose of the Study:

• To investigate the potential of using image analysis for weed emergence data collection.
• To develop and validate weed emergence models based on pixel data from RGB images.
• To compare the accuracy of image-derived models with traditional seedling count methods.

Main Methods:

• RGB images were captured repeatedly during the emergence period of *Raphanus raphanistrum* and *Senna obtusifolia*.
• Pixel-based spectral classification was applied to the images.
• Relative cumulative pixel data over time was used to develop emergence models.
• Models were validated against actual seedling counts.

Main Results:

• The cumulative pixel models for *R. raphanistrum* and *S. obtusifolia* explained 92% of the variation in relative seedling emergence.
• Image analysis provided a reliable method for generating weed emergence data.
• The accuracy of the image-derived models was equivalent to those based on seedling counts.

Conclusions:

• A simple image analysis technique can effectively generate data for weed emergence predictive models.
• This method offers a low-cost and technologically simple alternative to manual data collection.
• The findings support the advancement of automated weed control strategies through improved data acquisition.