• Proceedings of the Korean Environmental Sciences Society Conference
• /
• 2003.11a
• /
• pp.33-36
• /
• 2003

Principal component analysis (PCA) was used to improve image classification by the unsupervised classification techniques, the K-means. To do this, I selected a Landsat TM scene of Jeju Island, Korea and proposed two methods for PCA： unstandardized PCA (UPCA) and standardized PCA (SPCA). The estimated accuracy of the image classification of Jeju area was computed by error matrix. The error matrix was derived from three unsupervised classification methods. Error matrices indicated that classifications done on the first three principal components for UPCA and SPCA of the scene were more accurate than those done on the seven bands of TM data and that also the results of UPCA and SPCA were better than those of the raw Landsat TM data. The classification of TM data by the K-means algorithm was particularly poor at distinguishing different land covers on the island. From the classification results, we also found that the principal component based classifications had characteristics independent of the unsupervised techniques (numerical algorithms) while the TM data based classifications were very dependent upon the techniques. This means that PCA data has uniform characteristics for image classification that are less affected by choice of classification scheme. In the results, we also found that UPCA results are better than SPCA since UPCA has wider range of digital number of an image.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.6
• /
• pp.2806-2825
• /
• 2018

Unsupervised learning has shown good performance on image, video and audio classification tasks, and much progress has been made so far. It studies how systems can learn to represent particular input patterns in a way that reflects the statistical structure of the overall collection of input patterns. Many promising deep learning systems are commonly trained by the greedy layerwise unsupervised learning manner. The performance of these deep learning architectures benefits from the unsupervised learning ability to disentangling the abstractions and picking out the useful features. However, the existing unsupervised learning algorithms are often difficult to train partly because of the requirement of extensive hyperparameters. The tuning of these hyperparameters is a laborious task that requires expert knowledge, rules of thumb or extensive search. In this paper, we propose a simple and effective unsupervised feature learning algorithm for image classification, which exploits an explicit optimizing way for population and lifetime sparsity. Firstly, a sparse target matrix is built by the competitive rules. Then, the sparse features are optimized by means of minimizing the Euclidean norm ($L_2$) error between the sparse target and the competitive layer outputs. Finally, a classifier is trained using the obtained sparse features. Experimental results show that the proposed method achieves good performance for image classification, and provides discriminative features that generalize well.

• Korean Journal of Remote Sensing
• /
• v.22 no.2
• /
• pp.111-121
• /
• 2006

The classification using spectral angle is a new approach based on the fact that the spectra of the same type of surface objects in RS data are approximately linearly scaled variations of one another due to atmospheric and topographic effects. There are many researches on the unsupervised classification using spectral angle recently. Nevertheless, there are only a few which consider the characteristics of Hyperspectral data. On this study, we propose the ISOMUSAC(Iterative Self-Organizing Modified Unsupervised Spectral Angle Classification) which can supplement the defects of previous unsupervised spectral angle classification. ISOMUSAC uses the Angle Division for the selection of seed points and calculates the center of clusters using spectral angle. In addition, ISOMUSAC perform the iterative merging and splitting clusters. As a result, the proposed algorithm can reduce the time of processing and generate better classification result than previous unsupervised classification algorithms by visual and quantitative analysis. For the comparison with previous unsupervised spectral angle classification by quantitative analysis, we propose Validity Index using spectral angle.

• Korean Journal of Remote Sensing
• /
• v.38 no.2
• /
• pp.199-213
• /
• 2022

The unsupervised domain adaptation can solve the impractical issue of repeatedly collecting high-quality training data every year for annual crop classification. This study evaluates the applicability of deep learning-based unsupervised domain adaptation models for crop classification. Three unsupervised domain adaptation models including a deep adaptation network (DAN), a deep reconstruction-classification network, and a domain adversarial neural network (DANN) are quantitatively compared via a crop classification experiment using unmanned aerial vehicle images in Hapcheon-gun and Changnyeong-gun, the major garlic and onion cultivation areas in Korea. As source baseline and target baseline models, convolutional neural networks (CNNs) are additionally applied to evaluate the classification performance of the unsupervised domain adaptation models. The three unsupervised domain adaptation models outperformed the source baseline CNN, but the different classification performances were observed depending on the degree of inconsistency between data distributions in source and target images. The classification accuracy of DAN was higher than that of the other two models when the inconsistency between source and target images was low, whereas DANN has the best classification performance when the inconsistency between source and target images was high. Therefore, the extent to which data distributions of the source and target images match should be considered to select the best unsupervised domain adaptation model to generate reliable classification results.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.3 no.2
• /
• pp.134-146
• /
• 2009

Traffic classification seeks to assign packet flows to an appropriate quality of service(QoS) class based on flow statistics without the need to examine packet payloads. Classification proceeds in two steps. Classification rules are first built by analyzing traffic traces, and then the classification rules are evaluated using test data. In this paper, we use self-organizing map and K-means clustering as unsupervised machine learning methods to identify the inherent classes in traffic traces. Three clusters were discovered, corresponding to transactional, bulk data transfer, and interactive applications. The K-nearest neighbor classifier was found to be highly accurate for the traffic data and significantly better compared to a minimum mean distance classifier.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2016.07a
• /
• pp.51-54
• /
• 2016

In computer vision especially in image processing, it has become popular to apply deep convolutional networks for supervised learning. Convolutional networks have shown a state of the art results in classification, object recognition, detection as well as semantic segmentation. However, supervised learning has two major disadvantages. One is it requires huge amount of labeled data to get high accuracy, the second one is to train so much data takes quite a bit long time. On the other hand, unsupervised learning can handle these problems more cheaper way. In this paper we show efficient way to learn features for classification in an unsupervised way. The network trained layer-wise, used backpropagation and our network learns features from unlabeled data. Our approach shows better results on Caltech-256 and STL-10 dataset.

• Korean Journal of Cognitive Science
• /
• v.10 no.1
• /
• pp.13-24
• /
• 1999

In this paper, we propose a classifier based on modular networks using an unsupervised learning method. The structure of each module is designed through stochastic analysis of input data and each module classifier data independently. The result of independent classification of each module and a measure of the nearest distance are integrated during the final data classification phase to allow more precise c classification. Computation time is decreased by deleting modules that have been classified to be incorrect during the final classification phase. Using this method. a neural network sharing the best performance was implemented without considering. lots of of variables which can affect the performance of the neural network.

• Proceedings of the KSRS Conference
• /
• 2005.10a
• /
• pp.134-137
• /
• 2005

Unsupervised spectral angle classification (USAC) is the algorithm that can extract ground object information with the minimum 'Spectral Angle' operation on behalf of 'Spectral Euclidian Distance' in the clustering process. In this study, our algorithm uses the unit vector instead of the spectral distance to compute the mean of cluster in the unsupervised classification. The proposed algorithm (MUSAC) is applied to the Hyperion and ETM+ data and the results are compared with K-Meails and former USAC algorithm (FUSAC). USAC is capable of clearly classifying water and dark forest area and produces more accurate results than K-Means. Atmospheric correction for more accurate results was adapted on the Hyperion data (Hyperion-FLAASH) but the results did not have any effect on the accuracy. Thus we anticipate that the 'Spectral Angle' can be one of the most accurate classifiers of not only multispectral images but also hyperspectral images. Furthermore the cluster unit vector can be an efficient technique for determination of each cluster mean in the USAC.

• Korean Journal of Remote Sensing
• /
• v.24 no.5
• /
• pp.517-524
• /
• 2008

In order to classify an satellite imagery into geospatial features of interest, the supervised classification needs to be trained to distinguish these features through training sampling. However, even though an imagery is classified, different results of classification could be generated according to operator's experience and expertise in training process. Users who practically exploit an classification result to their applications need the research accomplishment for the consistent result as well as the accuracy improvement. The experiment includes the classification results for training process used VITD polygons as a prior probability and training parameter, instead of manual sampling. As results, classification accuracy using VITD polygons as prior probabilities shows the highest results in several methods. The training using unsupervised classification with VITD have produced similar classification results as manual training and/or with prior probability.

• Korean Journal of Remote Sensing
• /
• v.19 no.4
• /
• pp.329-339
• /
• 2003

In this study, an approach of image fusion in decision level has been proposed for unsupervised image classification using the images acquired from multiple sensors with different characteristics. The proposed method applies separately for each sensor the unsupervised image classification scheme based on spatial region growing segmentation, which makes use of hierarchical clustering, and computes iteratively the maximum likelihood estimates of fuzzy class vectors for the segmented regions by EM(expected maximization) algorithm. The fuzzy class vector is considered as an indicator vector whose elements represent the probabilities that the region belongs to the classes existed. Then, it combines the classification results of each sensor using the fuzzy class vectors. This approach does not require such a high precision in spatial coregistration between the images of different sensors as the image fusion scheme of pixel level does. In this study, the proposed method has been applied to multispectral SPOT and AIRSAR data observed over north-eastern area of Jeollabuk-do, and the experimental results show that it provides more correct information for the classification than the scheme using an augmented vector technique, which is the most conventional approach of image fusion in pixel level.