Working with 'find local extrema' algorithm in eCognition
Hi,
I made use of 'find local extrema' algorithm to identify highest elevation points from input SRTM 1 Arc Sec DEM (resampled 30m) of an AOI in eCognition v9. The chessboard segmentation was performed. The input parameter' values of chessboard segmentation and the algorithm can be seen below;
Subsequently, a grid covering same AOI was generated in ArcGIS, and for each grid cell (50 sq.km), the pixel having 'Maximum' value was identified making use of 'Zonal Statistics' tool and by supplying same DEM.
By comparing results from both approaches, it was observed that the 'find local extrema' algorithm did pretty good job. However, at many instances, it failed to capture maximum elevation value as seen below;
The red dots denote spots captured by the algorithm while blue circles are locations of Maximum value (elevation) within each grid cell identified utilizing Zonal Statistics tool.
Here is another example;
The Zonal Statistics tool effectively works to identify max elevation value within each grid cell. It is noticeable that location of extreme pixel value (blue circle upper right grid cell) was extracted, on the contrary the 'find local extrema' algorithm in eCognition didn't spotted anything such within the said area. I am wondering what went wrong in this case?
Was this article helpful?
2 comments
Hi Neelum,
Great that you are using this algorithm! Probably you will need to set up a similar structure like shown in this section of the video --> https://youtu.be/OvWRkh8cFOY?t=464
Key here is that you will need to loop through the large tiles and and apply a chessboard segmentation of 1 on the current image object, then check for the highest pixel (object in this case as you applied chessboard 1), then you could merge those small objects again (all but the highest one) and the loop will take the next tile. Have a look at the video, this more or less shows how you could implement this. If you need help, just drop a line here ;).
Cheers,
Matthias
Grateful for valuable insight.