Aurora-thesis

This repo contains a variety of Matlab scripts and functions used during my thesis work. It's been cleaned up somewhat but can be messy in parts. It is not intended for random people to be able to run this code, as they won't have the data needed.

Notes for using the code

It's easiest if you add all the directories to your MATLAB runtime path. For example, you could run the following:

addpath(genpath('aurora-thesis'))

3rd Party

A variety of functions from other developers. The blobs directory contains the multi scale Laplacian code and the C++ code for 3D extrema search, which needs to be compiled to a mex file.

Functions

cartesian2polar - used for resampling images using interpolation
findclosest - utility for searching vectors
fit_ellipse - function for fitting ellipses to polar projections of aurora (uses optimisation functions)
gaussian2d - utility for Gaussian function
gaussian2dtilt - utility for tilted 2D Gaussian function (given an angle of rotation)
generatepixelcoords - generates matrices with the MLAT and MLT of each pixel for a polar projection
getAllMatPaths - given the path to a directory containing multiple events, and the name of a subdirectory, this function creates a data structure with all the filenames in the matching subdirectories (matching the VISE name pattern)
loadNSPolar - loads a 3D matrix of size 512x512xN given a path to a folder of files matching the pattern *NS_Polar.mat. Also returns normalised version of matrix and a cell array of their filenames
loadPeakImages - improved version of above, used for thesis work where there are far more files to deal with
loadpngto256sq - gets raw pixels from PNG images and converts to matlab format - very unlikely to need this!
mypca - do not use this PCA function
removedaylight - pre-thesis - probably should avoid using as it cuts the top of the image off
rmserror - compute RMS error of 2 images
rmsthresh - as above, with a threshold on the error to avoid skewed result
robustmse - robust mean squared error
timestamp_to_date - unix to matlab date conversion
vectorAngle
vise_fname_to_datenum converts filenames to matlab timestamps
magnetic_to_pixels and pixels_to_magnetic are for general coordinate conversion

Expectation Maximisation

In the em folder there are files from my 2013 project on extracting features using the E-M algorithm to represent images as a Gaussian Mixture Model.

For the thesis work, only update_mixture.m was used - this contains the algorithm for iterating the mixture model.

Clustering

The clustering folder contains code used to running clustering experiments. Experiments involve creating a feature vector for each event, and these are generated from pre-computed features (saved to disk as MAT files).

AbstractFeatures is essentially a definition of an (object oriented) interface for a set of clustering features.
BlobFeatures is a class for loading features for clustering using multi scale blobs. Instantiating an object of this class allows specifying which features are put into the feature vectors.
EMComponentsFeatures - load features from Gaussian Mixture Models
EllipseFeatures - load ellipse parameters
EllipseBlobFeatures - load both ellipse and blob features
clusteranalysis and clusterdensityGaussian are used for running some quick experiments and analysis of clustering. You probably can’t use them as is, but they demonstrate how to use the BlobFeatures class.

Laplacianblobs

This folder contains a lot of functions and other code relating to detecting multi-scale blobs. Only listing interesting ones here:

detectBlobs is the main function for extracting blobs given an image, and scale parameters
filterDayside is an example of a way to filter irrelevant blobs. We assume dayside is between 06:00 and 18:00 MLT and the interesting range of latitudes is between 50 and 70 MLAT.
saveAllBlobsOnlyPeak - this is a function for detecting blobs for peak images, and saving the results to disk. It has hard-coded parameters for the output directories (subdirectory name is the ‘blobsdirname’ parameter) and blob scales - it should be self-explanatory.

Ovalfit

Functions for fitting ovals to the images. This could still be improved - the basic approach is Jesper’s, but probably simplified in places. The main issue is the image quality…

fitEllipse - this method involves generating several guess ellipses and choosing the best of them using the eval_ellipse_binary function.
fitEllipsesAllNew is a script function which uses fitEllipse to detect ellipses and save the parameters to disk. Note it uses hard-coded paths. The intensity-scale is a multiplicative factor for image intensity (basically a unit conversion).

Experiments

This contains files for running clustering experiments. The main file for running a batch of clustering experiments is allexperiments.m. The clusterBlobsPeak etc. files do similar things, all producing plots which are saved to disk.

doClustering does the actual work using feature objects (see the Clustering folder above).
plotClustering is a helper function for producing figures about cluster statistics
replotClusters will reload cluster data and generate new plots/figures

Data Processing

Scripts for extracting features over the entire data set.

batchProcessEvents takes a function handle and performs a ‘map’ over all events in a directory
detectBlobs… - these files scan over all events in a directory and detect multi scale blobs, saving the data to subdirectories for each event. This is where the scales and blurring etc. are changed.
processEvent2mat is another higher-level mapping function for processing all events using a function handle

jensraaby / aurora-thesis