2

I'm struggling with a multispectral drone imagery composition in QGIS. Each drone's camera shot creates 5 images: RGB (don't caring right now), NIR, green, red edge and red. I need to do a composite of the 4 spectral images, but separately. I mean, first image with 4 bands; second with 4 bands; third with 4...

I could do this process manually but there's like 500 shots for each drone flight and that would mean 500 manual composites, so I want to explore more solutions.

How could I do this?

0

2 Answers 2

2

If stacking the individual multi-spectal images into a single 4-band image is your desired outcome then I think programming is your only option. Are your images already aligned? Parrot Sequoias and MicaSense camera lenses are offset. The rig-relative values (the measure of offset for each lens) is a published values and also should be included in the image EXIF or XMP data (read those with ExifTool). The cameras are offset, the images are offset. Any map algebras from those images will not be valid because the pixels are not aligned. One would think a simple trimming of the images based on the rig-relative values would be an easy fix but the multi-spectral sensors heat and expand at different rates and this warps the images in unpredictable ways. Those rig-relative values do not relate to actual image offsets for sets of images. This is quite frustrating.

Here is a web browser search link for aligning Micasense images to get you started on aligning multi-spectral images. If I remember there is some final stacking code in there as well.

A simpler solution would be to use Agisoft Metashape to create a single orthomosaic from all of the images from your camera. This is a 'paid-to-play' solution but Agisoft does handle the alignments nicely and it will output some plant-health indices layers upon request. Unfortunately, Agisoft (as with any Structure from Motion software) will have a difficult time with mono cultures, row crops, or heavily forested areas where there is little correspondence between adjacent and overlapping images the area if interest. In these cases the software cannot determine image alignment and will not include those images in the reconstruction.

If you are hardy but want a freeware solution I see that there now some multi-spectral camera support in WebODM (or ODM).

Can you post a screen shot of the files in your directory? Can you include the camera type in your post?

1

Build a model. (The band inputs are Raster Layer inputs). I use Build virtual raster, to create actual rasters you can use Merge instead.

enter image description here

Run it F5

At the bottom select "Run as batch process"

For each band, press Autofill - Add files by pattern.

InE:\gisdata\Sentinel2\Relativt_molnfria_spridda_bilder\Lov_pa\2020 there are three subfolders each with Sentinel 2 satellite image bands in them.

For Band 1 I find the red bands by the pattern *B04_10m.jp2: enter image description here

Keep autofilling the different bands.

For the outputs I use Fill by expression: concat('C:/Users/bera/Desktop/GIStemp/', base_file_name( @band_1 ), '.vrt'), (dont use backslashes) to concatenate the output folder, the band 1 filename without the extension, and .vrt:

enter image description here

Result: enter image description here

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.