automatic detection of anomalies in elevation data using deep learning

How to detect anomalies (sinks and spikes) in elevation data

[Daniel Pollard]

I wish Mathworks had a downvote button like Stack Overflow does.

1. Start here: https://www.mathworks.com/matlabcentral/answers/6200-tutorial-how-to-ask-a-question-on-answers-and-get-a-fast-answer
2. Deep learning is a buzzword. Machine learning is a process whereby you train a computer to recognise patterns in order to either make a decision about the classification of a dataset, or optimise parameters of fitting a regression model to a dataset. This is a vast field with present day research being done in a variety of areas - one of which is a subfield known as "deep learning". I took a Masters level course on ML and we only briefly touched on deep learning.
3. Why do you think ML is even applicable to your situation? Identifying anomalies is different to every single situation you could imagine. What do you define as an outlier? Why would deep learning be better at reliably finding anomalies than basic statistics, such as using the standard deviation?
4. This website is about MATLAB. What methods have you tried so far to identify anomalies using Matlab? Why didn't they work?

[Guy Boussougou]

Thank you for your answer.

I would like to use this method to detect artifacts on elevation data, because I work on a global scale (open access data like srtm). The classical methods do not allow to optimize the treatment of anomalies on a large volume of data. The anomalies in our case are deformations of the terrain characterized by peaks and troughs (Spike & Well).

The goal is to be able to identify them, then to correct them with classical methods of interpolation or replacement.

[Daniel Pollard]

Excellent and concise reply! You should have written that in your original question, it's much much clearer what you're after now.

All my experience with this sort of thing is in Python, not Matlab, so I'm not sure how useful I can be. The MATLAB documentation is always a great first port of call for any questions and that link has information about finding outliers in a dataset using regression lines. I don't know what form your data takes but the statistics and machine learning toolbox is almost certainly what you're after.

[Guy Boussougou]

Thank you

My data is elevation data such as SRTM, so there is a z (elevation) component that needs to be considered in the outlier detection process. This is what makes my problem complex.