Question: Can I use an NIR LED to help make weeds (or plants) stand out even more with an NIR camera?

robotfarms is asking a question about general
Follow this topic

by robotfarms | October 15, 2017 12:43 | #15052


I am playing around with a project that uses an openMV camera with the IR filter replaced with the infragram filter pack to detect weeds in a fallow (bare stubble) paddock of crops. Would it help if I added artificial lighting by using some NIR LED's to bounce some NIR light off the plants? Or am I wasting my time?

The goal is to make the plants stand out even more, and create a consistent lighting for the camera to view.

The light and camera module would only be about 50-60cm from the plant/ground.

There are quite a few different wavelength LED's available to choose from in the NIR range, is there a prefered wavelength or just something over the 750nm ish mark?

Any opinions or information would be greatly appreciated.

Thanks

Adam



5 Comments

Hola, yo creo que no va ha ser de ayuda, ya que todas las plantas pueden usar esa luz para hacer la fotosíntesis. He tratado de hacer una diferenciación entre malezas y arroz, y el agua y la reflectancia me juegan en contra, pero si se puede determinar las zonas donde son más pobladas mediante un mayor valor del index. Puedes probar una cámara sin modificar para ayudarte a identificar las malezas, y jugar con los index TGI y VARI, ya que te da la agrupación en falso color muy similar al NDVI (pero confía en las agrupaciones y no en los valores TGI y VARI). Saludos.

Reply to this comment...


The NIR illumination is probably not needed because daylight is a very consistent source of lots of NIR (although using NIR illumination at night could work well). If your goal is to distinguish healthy plants (weeds) from bare soil, your task is fairly easy. Healthy foliage reflects many times more NIR than most bare soil.

NIRweed_19800101-4516-3.JPG
Above: Pure NIR (>720 nm) photo of live dandelion surrounded by bare soil. It's easy to tell what is plant and what is not.

An index (e.g., NDVI) is not needed because a pure NIR photo probably produces more contrast between plants and soil. If you are using a good red filter which passes only wavelengths longer than 600 nm, then the blue channel of those photos could be a rather pure NIR photo. Exposure, focus, and processing might be easier with a pure NIR filter like Wratten 87.

If the soil is not bare, the task could be more difficult because old leaves reflect some NIR. In that case it could help to use an index which compares NIR with visible light.

NIRweed_19800101-4514-2.JPG
Above: Pure NIR photo (>720 nm) of the same dandelion before I removed the old dead leaves covering the soil. The contrast is not so great here.

NDVIdand.jpg
Above: NDVI image of the same dandelion on leaf-covered ground. This image was made from a single photo taken with a long pass filter at 620 nm (so the blue channel captured mostly NIR and the red channel captured both NIR and red).

Although an index like NDVI can be helpful, producing useful NDVI images requires a good filter, good exposure, custom white balance or calibration, and image processing. It has to be done carefully to get better results than you could get with pure NIR photos.

Chris

Thanks for the in depth answer Chris! If i needed to use this system in low light or night, do you think I would need to add NIR light, or is there some present in a regular led light? As I said above to William, by the looks of your examples above I might loose the weed in the image using NIR. I can play with both and see which one works best. Thanks again, Adam

Is this a question? Click here to post it to the Questions page.


I don't know how much LEDs vary in their NIR emission. But it's easy to add some NIR LEDs to the mix. There is an advantage to using only artificial lighting (doing the whole thing at night) because then the spectral quality of the illumination is constant. That can make computing indices (e.g., NDVI) more reliable.

One problem with lamps is that they could cast shadows which will add distinct patterns to the photos and also might compute to erroneous NDVI values (a common issue). Ideally, the lamps should be in a wide array behind a diffuser to reduce shadowing.

If you use lamps at night, the proportion of red to NIR might not be similar to daylight so standard indices will not give typical results. That is, legacy NDVI is based on the assumption that the foliage is being illuminated by sunlight which has a particular proportion of red and NIR. If you change this proportion, NDVI values could change and be out of the expected range for NDVI of healthy foliage. That won't be a problem, but you will have to redefine the range of NDVI values which signify healthy foliage.

Chris


Reply to this comment...


Thanks William, I was going to have a play with it in regular RGB color too, and find the best solution. I can have a play with all the variables too. As Chris showed below, the trash around the plant might also show up under NIR, so the unmodified might work better in my situation too. Thanks again, Adam

Reply to this comment...


Log in to comment