Hopelessly lost on autogen trees

[crushnik]

I have spent weeks of reading and trial-and-error trying to get autogen trees to show up. I am using USGS 3.75-minute ortho tiles such as this one here from Monterey, Calif: https://prd-tnm.s3.amazonaws.com/St...36121/m_3612125_ne_10_h_20160712_20161004.jp2

According to the metadata: "32-bit pixels, 4 band color(RGBIR) values 0 - 255" I should have an IR band in what I am guessing to be band #3, with red being #0 as is the default setting in ScenProc's Texture Filter Editor. I try to replicate the steps from the tutorial section of the PDF but I am getting nowhere. When I converted the JP2 file to GeoTIFF in QGIS, is it possible that I somehow removed the IR channel?

The only steps I took in QGIS are exporting as a Rendered Image and changed CRS to WGS84. This may sound incredibly simple to someone that knows what they are doing, but I've spent way too much time getting absolutely nowhere.

Thanks,

 

[arno]

Hi,

QGIS should not drop the 4th band while editing normally. How does the output of the convert to ndvi step look?

 

[crushnik]

I would send a screenshot but the Texture Filter Editor runs very slow and frequently crashes (I'm using the latest development release). I get a greyscale-looking preview image if I click on the NDVI step, but it becomes blank once I click on the Threshold Binary step.

EDIT: Got it to work finally...

 

[arno]

How big is your sample image? The texture filter editor should be quite quick if you keep the sample around 1000x1000 pixels.

You might have to adjust the threshold size to match your ndvi values. If you move your mouse you can see the ndvi values of your image. You need to check that value your trees have and adjust the threshold.

 

[crushnik]

This TIFF is 11k x 16k so that must be the issue. Thanks, I'll try making adjustments!

Sorry if my issue is terribly basic... Perhaps something I couldn't make sense out of in the manual will become more clear now.

 

[arno]

Hi,

The idea is to use small sample files in the texture filter editor, so that you can tune your texture filter. Afterwards you can run it on bigger images in the script that does the actual vegetation detection.

The NDVI image looks quite normal, it's the vegetation that has lighter values. So I think if you tune the threshold it might be OK. Often a value between 140 and 160 for the threshold works OK, but it really depends on the source data.

 

[crushnik]

Spectacular!

 

[GaryGB]

Hi Arno:

Would I be correct that the ScenProc Feature Detection engine will automatically detect the source imagery linked above is EPSG:3857, and that ScenProc will next automatically re-project the source imagery to EPSG:4326 via another prior 'scripted' function in ScenProc or QGIS without pre-processing to re-project ...before attempting processing as Chris describes above ?

Linked file Metadata in Global Mapper:

FILENAME=[path]\m_3612125_ne_10_h_20160712_20161004.jp2
DESCRIPTION=m_3612125_ne_10_h_20160712_20161004.jp2
UPPER LEFT X=-13574023.351
UPPER LEFT Y=4386967.243
LOWER RIGHT X=-13567059.751
LOWER RIGHT Y=4378296.643
WEST LONGITUDE=121.93752643° W
NORTH LATITUDE=36.62501813° N
EAST LONGITUDE=121.87497135° W
SOUTH LATITUDE=36.56248216° N
UL CORNER LONGITUDE=121.93752643° W
UL CORNER LATITUDE=36.62501813° N
UR CORNER LONGITUDE=121.87497135° W
UR CORNER LATITUDE=36.62501813° N
LR CORNER LONGITUDE=121.87497135° W
LR CORNER LATITUDE=36.56248216° N
LL CORNER LONGITUDE=121.93752643° W
LL CORNER LATITUDE=36.56248216° N
PROJ_DESC=Mercator / GOOGLE / meters
PROJ_DATUM=GOOGLE MAPS (SPHERE RADIUS 6378137)
PROJ_UNITS=meters
EPSG_CODE=EPSG:3857
COVERED AREA=38.921 sq km
LOAD TIME=0.50 s
NUM COLUMNS=11606
NUM ROWS=14451
NUM BANDS=4
COLOR BANDS=0,1,2
PIXEL WIDTH=0.6 meters
PIXEL HEIGHT=0.6 meters
SAMPLE TYPE=Unsigned 8-bit Integer
BIT DEPTH=32

GaryGB

 

[arno]

Hi Gary,

Yes, when you load imagery that is not in WGS84 coordinate system, scenProc will automatically reproject. But it can be more efficient to make sure it is WGS84 beforehand.

 

[GaryGB]

Thanks, Arno.

GaryGB