The work described here is discussed in Kemeny and Kim (2023). A large AI model was trained on a variety of different kinds of edges, as described in Xie and Tu (2015). The AI technique is called Holistically-Nested Edge Detection (HED), and considers holistic image training and prediction similar to Convolutional Neural Nets, and nested multi-scale learning, as shown in the first image below. We have tested the HED pre-trained network on many images of rock outcrops containing discontinuities and found that it works surprisingly well compared with traditional edge detection. We have adapted the technique described in Bhattiprolu (2022) that involves the following steps: 1) image capture, 2) preprocessing, 3) applying the pre-trained HED net, 4) thresholding, and 5) block segmentation. The figure below shows these steps.

There are several important applications to geological engineering and rock mechanics. Fracture spacing and/or block size are very important parameters in classifying rock masses (e.g., GSI, RMR, Q). Accurate discontinuity delineation can also be used to monitor slope loosening and slope movement. Images can be scaled if using a stereo camera (see previous blog post on the Oak-D stereo camera with Myriad X neural vision chip), and 2D edges can be used in 3D characterization when applied with photogrammetry or Gaussian Splats (see next blog post on gaussian splats).

The figure below shows HED applied to a number of different rock outcrops, demonstrating the usefulness and accuracy of this technique. Note that adjusting the threshold level and removing very small and very large blocks can improve accuracy, which was not undertaken here.

If you are interested in this topic and want to work with us on further developing it (as research or a potential commercial product) please contact and let’s discuss.

Xie, S., Z. Tu. 2023. Holistically-Nested Edge Detection, arXiv:1504.06375 [cs.CV]

Bhattiprolu. 2022. Object segmentation using Deep Learning based edge detection (HED)​, github.com/bnsreenu/python_for_microscopists, number 291.

Kemeny, J. and K. Kim. 2023. Simple Sensor Solutions (With the Help of AI) for Geologic and Hydrologic Hazards Associated with Climate Change, Poster presentation at the 2023 Fall Meeting of the American Geophysical Union, San Francisco, CA.