Posted by
Mario Emmenlauer-3 on
Dec 02, 2020; 7:28pm
URL: http://imagej.273.s1.nabble.com/Is-imaging-technology-advanced-enough-to-do-this-tp5024265p5024267.html
Dear Jorge,
On 02.12.20 19:14, Jorge A. Santiago-Blay wrote:
> Dear ImageJ members:
>
>
> Is technology advanced enough now to do what I describe in the next few
> lines? Given that Google Maps stores the data and images captured from the
> roads, is there a way to automatically combine artificial intelligence
> software to id the obvious objects (e.g., plants) all over the planet and
> whenever one is detected, dump the image into a folder of images along with
> the all-important lat, long, and the date? Once the software learns that
> such and such shape is whatever is of interest to us, it can
> *automatically* keep identifying the object and dumping images with the
> necessary ancillary data. In other words, like the folks who advocated
> phenetics in the 1960's, can machines do the grunt work currently done by
> mortals? Thanks for any constructive feedback to
[hidden email] .
The question is what precision you would need. What false positive and
false negatives can you accept? All the steps of the method you describe
are nowadays possible. But the output may still be completely useless for
you, for example because the AI may have a hard time to discriminate
individual plants in a vegetation, or because too many non-plant objects
look like plants, etc...
So you would really need to be very specific in what errors you can accept,
and then you would need to work with machine learning specialists and see
if they can achieve this precision. If people have done this before (and I
would assume they have tried) you may be best of in a very geography-
centric community or in a very machine-learning centric community. ImageJ
is a great tool but its not predominantly used for machine learning.
All the best,
Mario
--
ImageJ mailing list:
http://imagej.nih.gov/ij/list.html