• Korean Journal of Remote Sensing
• /
• v.25 no.3
• /
• pp.255-261
• /
• 2009

A new polarimetric SAR image classification technique based on the degree of polarization (DoP) and the co-polarized phase-difference (CPD) is presented in this paper. Since the DoP and the CPD of a scattered wave provide information on the randomness of the scattering and the type of scattering mechanisms, at first, the statistics of the DoP and CPD are examined with measured polarimetric SAR image data. Then, a DoP-CPD diagram with appropriate boundaries between six different classes is developed based on the SAR image. The classification technique is verified using the JPL AirSAR and ALOS PALSAR polarimetric data. The technique may have capability to classify an SAR image into six major classes; a bare surface, a village, a crown-layer short vegetation canopy, a trunk-layer short vegetation canopy, a crown-layer forest, and a trunk-dominated forest.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.16 no.2
• /
• pp.187-194
• /
• 1998

In this study, effective multispectral Landsat TM band combinations for a merging with the high resolution IRS-1C PAN data using the IHS method to improve landuse accuracy is discussed. From the pre-classified image using the merged images with TM all six band images(with the exception of band 6 image) and PAN image, a sample data which has ten classes was generated. An evaluation of the overall classification accuracy for the representative seven merged images which were merged using each TM three-band images and IRS-1C PAN image by IHS method for the sample area. The increase in classification accuracy is most significant with the inclusion of two of TM4, TM5 and TM7 infrared band images. Especially, the largest increase(11.8 percent) in landuse classification accuracy were investigated when Landsat TM247 bands were merged with IRS-1C PAN data. The classification accuracy when TM three band image and PAN image were used without merging is higher than result of the case of using the merged images.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.2 no.2
• /
• pp.42-60
• /
• 1992

An adaptive fuzzy system can efficiently classify subimages into four categories according to image activity level for image data compression. The system estimates fuzzy rules by clustering input-output data generated from a given adaptive transform image coding process. The system encodes different images without modification and reduces side information when encoding multiple images. In the second part, a fuzzy system estimates optimal bit maps for the four subimage classes in noisy channels assuming a Gauss-Markov image model. The fuzzy systems respectively estimate the sampled subimage classification and the bit-allocation processes without a mathematical model of how outputs depend on inputs and without rules articulated by experts.

• Proceedings of the KSRS Conference
• /
• 2005.10a
• /
• pp.221-224
• /
• 2005

Hyperspectral image data have great potential to depict more detailed information on biophysical characteristics of surface materials, which are not usually available with multispectral data. This study aims to test the potential of hyperspectral data for classifying five soil classes defined by the vector product interim terrain data (VITD). In this study, we try to classify surface materials of bare soil over the study area in Korea using both hyperspectral and multispectral image data. Training and test samples for classification are selected with using VITD vector map. The spectral angle mapper (SAM) method is applied to the EO-I Hyperion data and Landsat ETM+ data, that has been radiometrically corrected and geo-rectified. Higher classification accuracy is obtained with the hyperspectral data for classifying five soil classes of gravel, evaporites, inorganic silt and sand.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2019.05a
• /
• pp.151-152
• /
• 2019

In this paper, we propose an improved learning method based on a small data set for animal image classification. First, CNN creates a training model for a small data set and uses the data set to expand the data set of the training set Second, a bottleneck of a small data set is extracted using a pre-trained network for a large data set such as VGG16 and stored in two NumPy files as a new training data set and a test data set, finally, learn the fully connected network as a new data set.

• Proceedings of the KSRS Conference
• /
• 2002.10a
• /
• pp.81-81
• /
• 2002

Classification is considered as one of the processes of extracting attributes from satellite imagery and is one of the usual functions in the commercial satellite image processing software. Accuracy of classification plays a key role in deciding the usage of its results. Many tremendous efforts far the higher accuracy have been done in such fields; training area selection, classification algorithm. Our research is one of these effort in different manners. In this research, we conduct classification using multiple satellite image data and evidential approach. We statistically consider the posterior probabilities and certainty in maximum likelihood classification and methodologically Dempster's orthogonal sums. Unfortunately, accuracy for the whole data sets has not assessed yet, but accuracy assessments in training fields and check fields shows accuracy improvement over 10% in overall accuracy and over 0.1 in kappa index.

• Journal of Korea Multimedia Society
• /
• v.25 no.1
• /
• pp.29-38
• /
• 2022

Although image compression is one of the essential technologies to transmit image data on a variety of surveillance and mobile healthcare applications, it causes unnecessary compression artifacts such as blocking and ringing artifacts by the lossy compression in the limited network bandwidth. Recently, image restoration methods using convolutional neural network (CNN) show the significant improvement of image quality from the compressed images. In this paper, we propose Image Denoising Convolutional Neural Networks (IDCNN) to reduce the compression artifacts for the purpose of improving the performance of object classification. In order to evaluate the classification accuracy, we used the ImageNet test dataset consisting of 50,000 natural images and measured the classification performance in terms of Top-1 and Top-5 accuracy. Experimental results show that the proposed IDCNN can improve Top-1 and Top-5 accuracy as high as 2.46% and 2.42%, respectively.

• Korean Journal of Remote Sensing
• /
• v.22 no.1
• /
• pp.75-85
• /
• 2006

This paper has dealt with a data fusion for the problem of land-cover classification using multisensor imagery. Dempster-Shafer evidence theory has been employed to combine the information extracted from the multiple data of same site. The Dempster-Shafer's approach has two important advantages for remote sensing application: one is that it enables to consider a compound class which consists of several land-cover types and the other is that the incompleteness of each sensor data due to cloud-cover can be modeled for the fusion process. The image classification based on the Dempster-Shafer theory usually assumes that each sensor is represented by a single channel. The evidential approach to image classification, which utilizes a mass function obtained under the assumption of class-independent beta distribution, has been discussed for the multiple sets of mutichannel data acquired from different sensors. The proposed method has applied to the KOMPSAT-1 EOC panchromatic imagery and LANDSAT ETM+ data, which were acquired over Yongin/Nuengpyung area of Korean peninsula. The experiment has shown that it is greatly effective on the applications in which it is hard to find homogeneous regions represented by a single land-cover type in training process.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.11a
• /
• pp.112-115
• /
• 2022

The process of deep learning usually needs to deal with massive data which has greatly limited the development of deep learning technologies today. Convolutional Neural Network (CNN) structure is often used to solve image classification problems. However, a large number of images may be required in order to train an image in CNN, which is a heavy burden for existing computer systems to handle. If the image data can be compressed under the premise that the computer hardware system remains unchanged, it is possible to train more datasets in deep learning. However, image compression usually adopts the form of lossy compression, which will lose part of the image information. If the lost information is key information, it may affect learning performance. In this paper, we will analyze the effect of image compression on deep learning performance on CNN-based cat and dog classification. Through the experiment results, we conclude that the compression of images does not have a significant impact on the accuracy of deep learning.

• Journal of information and communication convergence engineering
• /
• v.19 no.2
• /
• pp.79-83
• /
• 2021

This paper proposes a novel image classification method based on few-shot learning, which is mainly used to solve model overfitting and non-convergence in image classification tasks of small datasets and improve the accuracy of classification. This method uses model structure optimization to extend the basic convolutional neural network (CNN) model and extracts more image features by adding convolutional layers, thereby improving the classification accuracy. We incorporated certain measures to improve the performance of the model. First, we used general methods such as setting a lower learning rate and shuffling to promote the rapid convergence of the model. Second, we used the data expansion technology to preprocess small datasets to increase the number of training data sets and suppress over-fitting. We applied the model to 10 monkey species and achieved outstanding performances. Experiments indicated that our proposed method achieved an accuracy of 87.92%, which is 26.1% higher than that of the traditional CNN method and 1.1% higher than that of the deep convolutional neural network ResNet50.