Episode #125 of the Stack Overflow podcast is here. We talk Tilde Club and mechanical keyboards. Listen now
34

I would have to say that the most complete software environment for Machine Learning and nonparametric modeling is R. This is a big field in statistics, spanning K-NN, Kernel smoothing, General Additive Models, weak learners, support vectors, neural nets, semi-parametric spline regression, imputation, etc... I would highly recommend reading: Hastie, T., R. ...


28

I am not sure that I understand what you mean by "collect" data. If you are referring to heads-up digitizing and assignment of classes, this is best done in a GIS. There are many free options that would be suitable (i..e, QGIS, GRASS). Ideally you would have field data to train your classification. The procedure for classification using Random Forests is ...


23

I'd strongly recommend scikits-learn for Python. It supports supervised and unsupervised classification and the documentation is excellent (particularly check out the Machine Learning for Astronomical Data Analysis tutorial and the accompanying YouTube video (note: this is 3 hours long)). The project is under active development, with the last version being ...


11

A good overview of machine learning techniques in R is the machine learning taskview. It offers a host of different algorithms, recommended by the experts.


7

Classification algorithms such as Maximum Liklihood, random forests, and SVM are statistical methods for grouping data. These data may be words, colors, sounds or anything you can imagine. In a remote sensing context, these algorithms are used to group pixels or image objects (segments) based on statistical properties, or spectral profiles. To answer ...


6

Your question assumes that machine learning algorithms for land classification are somehow distinct from software used for other machine learning applications. There are some applications that require special treatment because of unusual characteristics, but there is no reason I know of to think that land use needs special treatment. If land use data can ...


6

In general, there are two approaches to classification: pixel-based and object-based: Pixel-based: Each spatial pixel is evaluated by itself against a set classification parameters. In this case, pansharpening the image will not help you at all. Object-based / Segmentation: In this approach pixels are evaluated as groups and segmented into groups based on ...


5

I do not really see the applicability of Machine Learning with GPS data. However, I utilize these type of modeling techniques on spatial/GIS databases for a large variety of applications. I would imagine that, in fact, these skills would be highly valued indeed. Methods encompassing supervised and unsupervised data mining, machine learning and ...


5

I know that this thread is a little old, but for anyone wanting to try classification of remote sensing data in R, a very promising new package has been released. install.packages("RSToolbox") It comes with functions for both unsupervised and supervised classification (using random forests). More information can be found here - http://bleutner.github.io/...


5

Did you have a look at eCognition? With their new Version (8.9) they provide Random Forests algorithm within a GUI environment. You can create nice process trees and include object features.


5

Yes, image classification is generally improved when you remove topographic illumination effects (same goes with atmospheric effects). However, as with anything like this, there exists a wide range of techniques for accomplishing this task and the effectiveness of it will depend both on the sophistication of the technique (its ability to model the physical ...


5

You can perform a land cover classification on a single Landsat scene without performing spectral and radiometric corrections. You will only need to do those corrections if you're trying to apply reference spectra to your classification, performing a classification that covers multiple scenes or performing a classification over a time series of the same ...


5

Suppose two LULC rasters with 6 classes each one: library(raster) library(rasterVis) r <- raster() set.seed(123) lc1 <- setValues(r, sample(1:6, 64800, replace = T)) lc2 <- setValues(r, sample(1:6, 64800, replace = T)) To detect landcover changes, the basic approach is to using logical tests: changeDet1 <- lc1 != lc2 The result is 1 when ...


4

The statistics that you highlight do not care if the data is a set of discrete "objects" or individual pixel based. I would also point out that it is quite incorrect to assert that "Random Forests" or "Support Vectors" are object-based and "Maximum Likelihood" pixel based classification algorithms. The model specification is dependent on a response vector [y]...


4

A classifier, any classifier, can classify any kind of data. These objects, as Aaron correctly states, can be pixels, objects, superpixels, bananas, sounds, DNA, etc. The main differences which, in my opinion, is really relevant between superpixel- and pixel-based classification are as follows: pixel based : The resolution of the prediction is maximal, ...


4

I had some success by changing the target directory to which I saved the signature file. Rather than using the server space I had been alloted, I chose the local desktop, and it worked without a hitch. Not sure why.


4

You can also do land classification with DTclassifier (Decision Tree classifier) plugin for QGIS. It provides simple interface for classification of raster data using decision trees, to perform within QGIS.


4

There is a group out of Duke University that have developed some interesting script tools for ArcGIS, including random forest models. Marine Geospatial Ecology Tools


4

ATCOR for Erdas Imagine will convert DN to true reflectance--this step is critical for change detection analyses. Then you can use DeltaCue add-on in Erdas Imagine to detect land cover change. More details on the DeltaCue add-on can be found here. Additionally, there is a fairly good instructional video on how to use DeltaCue to get you started. I would ...


4

The WWF ecoregions are the original Olsen et al., (2001) classification units. If this is what you are after I would recommend using The Nature Conservancy modification to Olsen that deals with some of the known issues with the original classification. My preference is the Brown et al., (1998) classification and its Rehfeldt et al., (2012) modification but,...


4

First of all you cannot create a spectral signature from a point. You will need polygons (training areas). Since it seems that you very exactly know your study area, you could add a column to your gcp's with a radius in wich the observed category certainly exists (e.g. primary forest around 200m of a point), this should be manageable for the mentioned 100+ ...


4

Welcome to the trenches of classification, probability and statistics ;) . Assuming you used sklearn, it has a detailed user guide on what metrics it provides to evaluate your classifications: I'll quote the user guide on this: Intuitively, precision is the ability of the classifier not to label as positive a sample that is negative, and recall is the ...


3

Here and here are tutorials on supervised classification / regression with R, which includes a RandomForest examples.


3

You should provide more details about the task. In general, extracting features from the data image heavily depends on what you are trying to detect/classify, and how are you trying to do it. Here's an example. If you are interested in classifying roads from an urban scene, you may be interested in evaluating large linear filter responses over the whole ...


3

I'm not 100% sure, but I'll give a tentative answer. Ideally, if you have an image with "pure" and well defined classes from the spectral point of view, the use of parametric classifiers is safe. You can safely assume that your distribution follows e.g. a multivariate normal, with Gaussian additive noise as in all optical images, and then fit a fixed (...


3

1. where should I begin? Do you know what Image Classification is? If not here's an intro article ESRI wrote about for arcgis. You don' need arcgis to read it. Read it, and in the end you'll understand what you should need. Keep in mind that image classification is about creating classes. To do that should well defined classes beforehand (how many, ...


3

You may want to check out the USGS Land Cover Institute. Multiple land cover datasets and projects are referenced in the link.


3

You could try an image segmentation approach but, I would not hold my breath on usable results. As far as application of a classification algorithm to panchromatic imagery, it is quite doubtful that you will get usable results because of the lack of any spectral separability associated with your target vegetation class. The only relevant information that you ...


3

In my opinion, the coarser resolution of the thermal band is not necessarily the reason why it gets excluded in many applications, after all, you can always resample the thermal band to unify the resolutions. As @Radar has suggested in the comment, whether the thermal band needs to be included or not needs to be determined based on the specific relationship ...


Only top voted, non community-wiki answers of a minimum length are eligible