It can be useful to detect objects on images within documents. I labelled boreholes/well objects on 30 public domain images to illustrate what results can be achieved in less than an hour on unseen data. There are many other use cases in the subsurface such as objects on borehole logs, satellite imagery, remote sensing, thin sections, outcrops etc. There are of course many examples already in the literature of using AI in this way, such as sinkhole detection from aerial photography and satellite imagery.
The building of custom deep learning models on top of ImageNet is very easy with no code solutions. This is a brief foray away from my usual NLP topic into image analysis; my last (and only other) post on this topic was in 2017 showing how subsurface related images in documents can be easily classified with very few examples per category.
Classifying subsurface/geoscience images extracted from documents is quite commonplace now, but back then (to my knowledge) it had not been attempted before. I have 5,000 labelled subsurface images that I’m happy to give freely to non-profits should anyone need training data.
Link to 2017 article here: https://paulhcleverley.com/2017/11/15/applying-deep-learning-to-geoscience-image-type-classification-in-literature-some-early-research-findings/
#maps #borehole #artificialintelligence #unstructureddata #geosciences #subsurface
A Geodyssey - Geoscience Text Analytics and Enterprise Search Research 
Leave a comment