Clip Level 1 images to plot, analyze, combine

[dlebauer]

Given limitations of the full field stitching algorithm for downstream analyses ( #326 (comment),
#326 (comment), terraref/reference-data#183 (comment)), we need to have a protocol that supports the following workflow:

1. for a given plot, query all images of one or more types
2. compute metrics
3. combine into plot-level phenotypes

This approach should be consistent with the experimental design. For example, we treat the plots as our units of replication from which we can expect iid ('identical and identically distributed') samples. Thus we need to account for different sizes of the clipped images as well as regions of image overlap. Two approaches include 1) weighting by the area of the image and 2) identifying and keeping track of unique features.

Notes from a discussion on this topic https://docs.google.com/document/d/1CYxI03rZS6kZRff4RTXQ5ZJDApMdclc4T2PbiMftMXg/edit

Completion Criteria (this is an epic)

• develop terrautils function to query all image fragments of type given bounding box
• develop protocol(s) for aggregating results from image-level analysis to plot level summaries
• implement, starting with the panicle counting algorithm

[max-zilla]

ACTION ITEMS

• Max will soon share commands to go from a plot & time to a set of clipped GeoTIFF images for that plot
• We identify initial target extractors - RGB canopy cover, FLIR mean temperature, panicle counts - and try current approach + alternative approaches w/ manual visual inspection to attempt to quantify the relative accuracy of these approaches
• reconvene to determine integrating best approaches into extractors

[max-zilla]

Further discussion occurring here before proceeding: #422