Utilizing Band Ratios to Increase Contrast for Image Interpretation

Using band ratios to analyze remotely sensed data has become common in precision agriculture. The “Simple Ratio,” for instance, indicates the proportion of near-infrared radiation reflected from a leaf surface divided by the proportion of red radiation reflected from a surface. Every color image is created by “compositing” individual bands into a single image. A typical color image assigns Red to the red color gun (display), Green to the green color gun/display, and Blue to the blue color gun/display (Figure 1). However, we can manipulate the contrast in the image by assigning different bands or band ratios to different color guns/displays. 

Photo Credit:

David Kramar

True color composite showing emergence of corn. In this case, Red is assigned to the red color gun, Green to the green color gun, and Blue to the blue color gun. Just left of center and down is an area with weeds encroaching, making individual plant identification difficult.

Higher color contrast will aid in identifying each corn plant separately from the surrounding weeds, useful for obtaining a stand count or identifying weeds. By calculating a simple RGB-based vegetation index known as Excess Green we can remove background effects and focus solely on the plants (Figure 2).

Photo Credit:

David Kramar

Excess Green, defined as (2 * Green) – Red – Blue) does a good job identifying just the crops. This makes extraction relatively straightforward, and can be calculated from off-the-shelf consumer drones such as the Mavic Air, Mavic Pro, or Phantom 4.

Taking this one step further, we can use the Excess Green index as an input band into our composite imagery. By approaching image interpretation in this manner, we exploit the way that electromagnetic energy interacts with the plant. Exaggerating the spectral differences better identifies the plants themselves (Figure 3).

Photo Credit:

David Kramar

An image composite using the Excess Green index as the input for the red color gun, red for the blue color gun, and green for the green color gun. Adjusting the contrast of an input image aids interpretation of what we see in the image.

The next logical step is to extract the information from the image. To do this we need to identify only those pixels that represent corn. There are a few approaches, but object-based image analysis (OBIA) involves segmenting the image based on not only spectral characteristics, but also on texture, shape, orientation, and size and then extracting that information. The result is a layer that represents only the plants, and that can be used to obtain stand counts (Figure 4).

Photo Credit:

David Kramar

Extracted plants based off of the processes described above. In this image there are 900 corn plants identified. The area comprises approximately 0.04 acres.

There is much we can do with basic RGB imagery – from simply evaluating crop health, to obtaining accurate counts of plants.

David Kramar, PhD
David.Kramar@ndsu.edu
Extension Precision Agriculture Specialist