Per-Pixel Analysis to Detect Almond Bloom

In pursuit of the most accurate yield prediction methods, growers often look to early indicators of fruit set, such as bloom density, to provide insights. This is usually a lengthy, labor-intensive, and error-prone process with a narrow bloom window during which data can be collected. With multispectral imaging, targeted drone data captures, and advanced pixel-level analytics, Pollen Systems is making bloom detection faster and more consistent on a per-plant basis to help growers improve their yield prediction results.

In a nutshell, the analysis uses the color characteristics of almond blooms to target specific pixels in our imagery and calculate the exposed bloom surface area. For example, the ratio between the red and green light reflectance from an almond bloom stays relatively consistent when viewed from above and can be used to select pixels that contain blooms.

Almond Bloom Detection Composite RGB Image

Our model works within a defined threshold of red and green color levels to filter out any pixels that don’t fit within the correct red-green ratio range for the bloom. In the result below, all white pixels are within the correct color range, while the black pixels are omitted from further analysis.

Almond Bloom Detection Red-Green Ratio Mask

These steps are then repeated for the other wavelengths using sampled bloom pixel color values. The resulting mask layers are each used to filter out non-bloom pixels later in the process.

Not only do we use color readings in the visible light (red, green, and blue) wavelengths, but we also use near-infrared wavelengths invisible to the human eye to screen out soil pixels that could interfere with our results.

Almond Bloom Detection NDVI Mask

Finally, all mask layers are combined into a single mask that takes all bloom identification factors into account.

Almond Bloom Detection Combined Bloom Mask

This layer then has our individual plant map placed on top so each tree’s bloom data has a unique identifier. From here, we can calculate the exposed surface area of all the blooms on each tree and report bloom uniformity across an entire orchard.

Almond Bloom Detection Targeted Bloom Pixels

After ground truthing calibrations from representative trees in the orchard, growers can use this model to apply bloom protection measures and refine their yield estimates down to the individual tree. Drone data collection can be performed over an entire orchard in one day, dramatically reducing labor costs while optimizing sample size. These and many more actionable insights are how Pollen Systems can help growers produce higher yields and better crops. Reach out to our team today to learn how this technology can be applied in your field this season!