Learning objectives of this topic
- Explain the concepts of image space and feature space
- Summarize the different types of spectral classification
- Discuss training and validation strategies for image classification
- Calculate and interpret a confusion matrix in the context of classification accuracies
The objective of digital image classification is to assign all pixels in the image to particular classes or themes (e.g., water, coniferous forest, deciduous forest, corn, wheat). The resulting classified image comprises a mosaic of pixels, each of which belong to a particular theme, and is essentially a thematic “map” of the original image. There is a variety of approaches to perform digital image classification. Spectral image classification uses the spectral information represented by digital numbers (pixel values) in one or more spectral bands, and attempts to classify each individual pixel based on this spectral information. This type of classification is termed spectral pattern recognition.
When talking about classes, we need to distinguish between information classes and spectral classes. Information classes are those categories of interest that the analyst is actually trying to identify in the imagery, such as different kinds of crops, different forest types or tree species, different geologic units or rock types, etc. Spectral classes are groups of pixels that are uniform (or near-similar) with respect to their brightness values in the different spectral bands of the data. The objective is to match the spectral classes in the data to the information classes of interest. Rarely is there a simple one-to-one match between these two types of classes. Rather, unique spectral classes may appear which do not necessarily correspond to any information class of particular use or interest to the analyst. Alternatively, a broad information class (e.g., forest) may contain a number of spectral subclasses with unique spectral variations. Using the forest example, spectral subclasses may be due to variations in age, species, and density, or perhaps as a result of shadowing or variations in scene illumination. Also keep in mind that the appearance of classes through time might not be homogenous (e.g., seasonal effects). It is the analyst’s job to decide on the utility of the different spectral classes and their correspondence to useful information classes.
Part I Feature Space
We begin first by covering the concept of image space versus feature space and how this is used in classifying images.
The following presentation summarizes the main points of the knowledge clip above. You can use it to review what you’ve just learned.
Part II Classification Types
There are many classification algorithms available, in this clip you will be introduced to some of them as well as things you should consider when conducting a classification on your remote sensing images.
The presentation below reviews some main topics and the classification algorithms discussed.
Part III Training Data Collection
Another important factor to consider is the training data collections including how this data is gathered.
The following book reviews the main concepts of the knowledge clip. You can use it to supplement your learning.
Sources & further reading
Belgiu, M., & Drăguţ, L. (2016). Random forest in remote sensing: A review of applications and future directions. ISPRS Journal of Photogrammetry and Remote Sensing, 114, 24-31
Buchhorn, M., Lesiv, M., Tsendbazar, N.-E., Herold, M., Bertels, L., & Smets, B. (2020). Copernicus Global Land Cover Layers—Collection 2. Remote Sensing, 12, 1044
Chen, D., & Wei, H. (2009). The effect of spatial autocorrelation and class proportion on the accuracy measures from different sampling designs. ISPRS Journal of Photogrammetry and Remote Sensing, 64, 140-150
Elmes, A et al (2020) Accounting for Training Data Error in Machine Learning Applied to Earth Observations. Remote Sensing, 12, 1034
Foody, G.M., & Mathur, A. (2004). Toward intelligent training of supervised image classifications: directing training data acquisition for SVM classification. Remote Sensing of Environment, 93, 107-117
Giles M. Foody (2013) Ground reference data error and the mis-estimation of the area of land cover change as a function of its abundance, Remote Sensing Letters, 4:8, 783-792, DOI: 10.1080/2150704X.2013.798708
Global Land Analysis and Discovery Lab. (2017, March 6). Estimating Tree Cover Area and Change Using Sample-based Analysis [Slides]. GLAD University of Maryland. https://glad.umd.edu/Potapov/Madagascar_2017/Documents/03_GLAD_Sampling.pdf
Mellor, A et al (2015) Exploring issues of training data imbalance and mislabelling on random forest performance for large area land cover classification using the ensemble margin. ISPRS J.Photogramm. Remote Sens.
Olofsson, P., Foody, G. M., Herold, M., Stehman, S. V., Woodcock, C. E., & Wulder, M. A. (2014). Good practices for estimating area and assessing accuracy of land change. Remote Sensing of Environment, 148, 42–57. https://doi.org/10.1016/j.rse.2014.02.015
Stehman S.V. and Czaplewski R.L. Design and analysis for thematic map accuracy assessment: fundamental principles. Remote Sensing of Environment 64, 331-334 (1998)
Stehman S.V. Sampling designs for accuracy assessment of land cover. International Journal of Remote Sensing 30 (20), 5243-5272 (2009)
Tarko, A., Tsendbazar, N.-E., de Bruin, S., & Bregt, A.K. (2020). Producing consistent visually interpreted land cover reference data: learning from feedback. International Journal of Digital Earth, 1-19