Active Semi-Supervised Random Forest for Hyperspectral Image Classification

Abstract

Random forest (RF) has obtained great success in hyperspectral image (HSI) classification. However, RF cannot leverage its full potential in the case of limited labeled samples. To address this issue, we propose a unified framework that embeds active learning (AL) and semi-supervised learning (SSL) into RF (ASSRF). Our aim is to utilize AL and SSL simultaneously to improve the performance of RF. The objective of the proposed method is to use a small number of manually labeled samples to train classifiers with relative high classification accuracy. To achieve this goal, a new query function is designed to query the most informative samples for manual labeling, and a new pseudolabeling strategy is introduced to select some samples for pseudolabeling. Compared with other AL-and SSL-based methods, the proposed method has several advantages. First, ASSRF utilizes the spatial information to construct a query function for AL, which can select more informative samples. Second, in addition to providing more labeled samples for SSL, the proposed pseudolabeling method avoids bias caused by AL-labeled samples. Finally, the proposed model retains the advantages of RF. To demonstrate the effectiveness of ASSRF, we conducted experiments on three real hyperspectral data sets. The experimental results have shown that our proposed method outperforms other state-of-the-art methods.Random forest, proposed by Breiman [16], is one of the most important machine learning methods. Compared with other machine learning methods, RF has the following advantages [9][10][11][17][18][19][20][21][22][23][24]. First, RF is suitable for high-dimensional data, which can alleviate the influence of curse of dimensionality [20]. Second, RF is very fast because it is implemented in parallel [9]. Third, RF is effective to handle data with imbalanced class labels or missing values [21]. Fourth, RF is not easy to fall into overfitting compared with the Boosting method [18]. Moreover, RF measures the importance of variables automatically [25]. Finally, RF can obtain a higher classification accuracy compared to other well-known classifiers such as SVM [5,6] and maximum likelihood (ML) [17,23], with fewer parameters.Active learning is a kind of iterative method that queries the most informative samples for manual labeling at each iteration [14]. Initially, the data set consists of two parts: a small number of labeled samples and a large number of unlabeled samples. At each iteration, a query function is adopted to investigate the unlabeled samples and select the most informative samples for manual labeling. After multiple iterations, the performance of the classifier improves along with the increase in the number of manually labeled samples. The query function plays an important role in determining the samples selected for manual labeling, which directly affects the classification performance [15].Semi-supervised learning, another way to tackle the issue of limited samples, trains a classifier with the labeled and unlabeled samples together, which does not r...

Hyperspectral Image Classification—Traditional to Deep Models: A Survey for Future Prospects

“…[299] proposed a feature-driven AL framework to define a wellconstructed feature space for HSIC. [300] proposed a Random Forest-based semi-supervised AL method that exploits spectral-spatial features to define a query function to select the most informative samples as target candidates for the training set.…”

A Novel Query Strategy-Based Rank Batch-Mode Active Learning Method for High-Resolution Remote Sensing Image Classification

“…Generally, the similarity function in RBMAL adopts the Euclidean distance and does not consider the spectral characteristics of samples [38]. The purpose of spectral similarity measurement in the combined query strategy in this study is to determine the similarity between the unknown spectrum and a known spectrum according to a spectral similarity measurement function and then divide the attributes of unknown categories according to the similarity results [72][73][74]. This approach is consistent with the description of the similarity function in RBMAL.…”

Improved Active Deep Learning for Semi-Supervised Classification of Hyperspectral Image

“…A new semi-supervised framework has been proposed to cope with high-dimensional and large-scale data, which combined the randomness with anchor graphs [32]. The advantages of random forest [33] include not over-fitting due to randomness and featuring good generalization performance. Moreover, the advantage of the anchor graph algorithm [34] is that it can linearly scale the size of the dataset, which is very suitable for HSI characteristics.…”

“…where F Θ is a nonlinear function of the CNN using parameters Θ, L u is the unary term that predicts the samples by using CNN, and L p can make adjacent samples to have the similar label. Here, the detailed explanation of label Y and the parameter Θ refers to the ADL classification method from [33]. The data input to the convolutional neural network is provided in the form of a three-dimensional cubic block with a three-dimensional convolutional kernel for convolution and pooling.…”