Understanding NDVI Cameras
NDVI is a widely used indicator to differentiate healthy vegetation from stressed vegetation. With the increasing use of UAVs in agriculture it is becoming popular to generate NDVI maps of agricultural fields using NDVI converted consumer cameras. Dronedeploy has already identified that the quality of NDVI camera conversions can significantly impact image quality. In this article I will explain both how consumer cameras can be converted into an NDVI camera and explain why the choice of filter also makes a significant impact on the resulting NDVI map.
What is NDVI?
Normalized Difference Vegetation Index (NDVI) was originally developed from satellite-based remote sensing imagery for classifying vegetation. Plants undergoing photosynthesis absorb the visible portion of the light spectrum from 400nm to about 700nm which is the same portion of the spectrum that humans can see. A little more green light is reflected compared to the blue and red portions of the spectrum which is why vegetation undergoing photosynthesis appears as green. Furthermore the plants reflect near-infrared wavelengths above 700nm. When plants are stressed the photosynthesis process slows down and less light is absorbed in the visible spectrum, and therefore more is reflected, and less light is reflected in the near-IR and more absorbed. This can be seen clearly in the figure below which shows the reflectivity of healthy and stressed vegetation.
NDVI is therefore calculated as the difference between the reflected light in the near-IR and the visible spectrum.
How digital cameras capture color
Digital cameras typically employ a repeating pattern of blue, green, and red filters over individual pixels called a bayer pattern. These filters modify the response of the pixels so that the blue pixels detect light in the spectrum between 400nm and 500nm, the green pixels detect light between 500nm and 600nm, and the red pixels detect light between 600nm and 700nm. However, these filters are not perfect and the blue, green and red pixels also detect light in the near-infrared above 800nm. The figure below shows an example of the frequency response for the Onsemi AR0134CS color image sensor, a common sensor used in industrial camera applications.
The response of the blue, green and red pixels to this near infrared light is problematic and without compensation images will typically be washed and have poor contrast. As a result it is common for camera manufacturers to put an IR-cutoff filter, also referred to as blue glass, between the lens and the image sensor to block wavelengths above 700nm. The transmittance curves for Schott BG38, a common IR-cutoff filter, are shown in the figure below.
Internal Transmittance is a measure of how much light passes through the filter. The Schott BG38 allows over 80% of the light in the visual range from 400nm to 650nm to pass through the filter before attenuating, or blocking, the flight beyond 650nm so that no near infrared light passes through the filter. When installed in front of the color image sensor, the blue, green, and red pixels now only detect light in their corresponding wavelengths as shown in the figure below.
The presence of the IR cutoff filter presents an opportunity to modify the camera’s optical response by replacing the IR cutoff filter with another filter with different properties. All NDVI conversion cameras employ this technique but the effectiveness varies significantly as I will discuss next.
Schott BG3 Filter Conversions
The earliest, simplest and most common NDVI conversion involves replacing the IR-cutoff filter with Schott BG3 colored glass. This filter absorbs, and therefore blocks, green and red light between 500nm and 700nm. The transmittance of Schott BG3 is given below.
Replacing the IR cutoff filter with Schott BG3 exposes the camera sensor to light in the blue (400nm-500nm) and near-IR (700nm-1000nm) portions of the spectrum. The result, shown in the figure below, is that the green and red pixels detect primarily near-IR light, while the blue pixels detect both blue and near-IR light.
To calculate NDVI the blue pixels are used as the value for the visible portion of the NDVI calculation, while the value for red, since it responds more to near-IR than the green pixel does, is used for the near-IR part of the equation. Therefore NDVI is calculated as:
Publiclab has several articles describing its camera conversions using Schott BG3 filters.
Green+Red Bandblock Filter Conversions
Event38 developed an improvement to the standard Schott BG3 filter by using a custom bandblock interference filter which improves the contrast of the NDVI calculation. The filter blocks roughly the same 500nm to 700nm green and red portion of the visible spectrum that Schott BG3 does but with significantly sharper ‘shoulders’ where the filter blocks the green and red light as shown below.
This custom filter has the advantage of transmitting more of the blue light below 500nm than Schott BG3. Since healthy vegetation reflects very little blue light, which is used as the visible portion of the NDVI calculation, contrast is improved by increasing the amount of blue light detected by the image sensor. The NDVI calculation is the same as for the Schott BG3 filter.
Midopt Blue+850nm IR Filter Conversions
More recently Midopt released a dual bandpass interference filter specifically designed for use in NDVI conversion cameras. The filter only transmits light in the blue spectrum between 460nm and 490nm, and in the near infrared between 830nm and 870nm. In the transmittance curve below, the filter transmits discrete wavelengths bands that align well with the visual spectrum absorption and near-IR reflection regions of healthy and stressed vegetation.
To fully evaluate this filter we have to look at it together with the image sensor as shown below.
With this filter the red pixels detect almost exclusively light around 850nm while the green pixels do the same but with slightly more detection of light in the blue visual spectrum. The blue pixels, on the other hand, detect blue light and light around 850nm. Therefore, with this filter we would calculate NDVI the same way as we would with the Schott BG3 filter.
With this filter, the user should consider the following points:
- The Midopt Blue+850nm NIR filter blocks large portions of the visual-near-IR spectrum compared to traditional NDVI filters and therefore requires longer exposure times, higher gains and larger apertures to obtain properly exposed images. This may require flying slower over the same area but it may also mean that you don’t have to fly with an ND filter.
- The amount of 850nm light detected by the blue pixels is almost equal to the amount of light that the red pixels detect so it may be worth investigating subtracting the red pixels from the blue pixels before calculating the NDVI index. This calculation can be added to the NDVI calculation which would change the NDVI calculation to:
Overall, I believe the Midopt Blue+850nm NIR filter has the potential to significantly improve NDVI performance compared to transitional Schott BG3 conversion. When you are shopping for an NDVI camera for your drone I suggest that you ask your potential vendor for the filter properties and ask yourself if that NDVI conversion will meet your needs.
