collinjroland
commented
7 months ago Update
I updated this to a less hardcoded solution, but I am still not sure if it matches the intent of the function. I think the intent of the function is quantify the fraction of valid pixels within the image, so the denominator in the image_extent calculation is supposed to the be the number of pixels in the image that intersect the AOI?
- Added these lines to the bottom of section 0) User Input Settings
import geopandas as gpdsettings['aoi_area'] = gpd.read_file(settings['aoi_geojson']).to_crs(settings['output_epsg']).area[0] -
Modify
get_file_extentdef get_file_extent(nan_path, settings): nan_mask = load_udm(nan_path) valid_area = np.where(nan_mask < 1)[0].size # aoi_area = nan_mask.shape[1]*nan_mask.shape[0] nan_area = np.sum(nan_mask) # image_extent = 100-(nan_area/aoi_area)*100 image_extent = ((valid_area / (settings['aoi_area'] / 9)) * 100) return image_extent
ydoherty
Description
When running cell
1.3) Pre-Processing - image coregistration and scene merging, I consistently receive 0 out of N scenes passed cloud and area criteria due to a failure to exceed the AOI area threshold, regardless of how low I set the AOI area threshold.Details
I attempted to debug the error and ended up at the function
get_file_extent.pre_process -> image_merge -> map_merge -> filter_aoi_cloud -> get_file_extentThe function calculates the
aoi_areaas the total number of pixels in thenan_maskarray, and thenan_areaas the sum of the cell values in thenan_maskarray. I think the nan values are supposed to be represented by a numeric 1 in the nan_array, while valid pixels are represented by a 0.I think this formulation works for rectangular images, but in my case I have clipped my imagery to an irregular polygon prior to download to minimize the areal extent of the download due to the total area ceiling on my Planet account. Thus, most or all of the nan pixels actually fall outside of my AOI and should not be included in the valid area calculation.
I modified the function to calculate the valid image extent by dividing the number of valid pixels in the nan_mask (those with a value of 0) by a hardcoded AOI area, converted to number of pixels by dividing by 9. This resolved the image filtering issue, but is not a robust / generalizable solution. More generally, I think that the udm2 -> udm conversion and general masking routine might be causing problems in the coregistration step as well.
Computing environment
OS: Windows 10 Enterprise Processor: 12th Gen Intel(R) Core(TM) i7-1265U Conda environment