Comparison of support vector machine, random forest and neural network classifiers for tree species classification on airborne hyperspectral APEX images

Abstract

Knowledge of tree species composition in a forest is an important topic in forest management. Accurate tree species maps allow for much more detailed and in-depth analysis of biophysical forest variables. The paper presents a comparison of three classification algorithms: support vector machines (SVM), random forest (RF) and artificial neural networks (ANN) for tree species classification using airborne hyperspectral data from the Airborne Prism EXperiment sensor. The aim of this paper is to evaluate the three nonparametric classification algorithms (SVM, RF and ANN) in an attempt to classify the five most common tree species of the Szklarska Poręba area: spruce (Picea alba L. Karst), larch (Larix decidua Mill.), alder (Alnus Mill), beech (Fagus sylvatica L.) and birch (Betula pendula Roth). To avoid human introduced biases a 0.632 bootstrap procedure was used during evaluation of each compared classifier. Of all compared classification results, ANN achieved the highest median overall classification accuracy (77%) followed by SVM with 68% and RF with 62%. Analysis of the stability of results concluded that RF and SVM had the lowest variance of overall accuracy and kappa coefficient (12 percentage points) while ANN had 15 percentage points variance in results.

Classification and Mapping of Paddy Rice by Combining Landsat and SAR Time Series Data

“…The performance of SVM was better than that of RF at both sites, which is consistent with Pouteau et al [70] and Raczko & Zagajewski [71]. In this study, classifications were conducted in a 10-dimensional feature space for scheme 1, and more than 60-dimensional feature space for the other schemes.…”

“…As SVM is known to work well in a large dimensional feature space, it could be better for SVM to classify paddy rice than RF [70]. Raczko and Zagajewski [71] stated that the number of training samples has an impact on the performance of a classifier, and Kavzoglu & Mather [72] stated that more than 400 pixels per class are needed for robust classification. In this study, a relatively small number of training samples (less than 200 pixels per class) were used, which might explain the better performance of SVM than RF because SVM works well with a small sample size as well as mixed pixels [56,73].…”

Support vector machines on the D-Wave quantum annealer

“…The field of supervised machine learning deals with the problem of learning model parameters from a set of labeled training data in order to make predictions about test data. SVMs in particular are known for their stability (in comparison to decision trees or deep neural networks [11][12][13][14]), in the sense that small differences in the training data do not generally produce huge differences in the resulting classifiers. Moreover, kernel-based SVMs profit from the kernel trick, effectively maneuvering around the "curse of dimensionality" [9,15].…”

The Impact of Tourist Traffic on the Condition and Cell Structures of Alpine Swards

“…Non-invasive methods should play a leading role in highly protected areas but have to be supported by biophysical variables [2,3]. For example, remote sensing enables the identification of plants and vegetation communities in mountainous areas [4,5], and the mapping or monitoring of vegetation can be carried out at various spatial scales depending on the sensors used [6]. The state of cell structures, photosynthetically active pigments, water, lignin and cellulose content determines the spectral reflectance curve in different parts of the electromagnetic spectrum [7].…”