• Journal of IKEEE
• /
• v.26 no.4
• /
• pp.639-646
• /
• 2022

In this paper, to solve the large dataset problem, we collect images through an image collection method called web crawling and build datasets for use in image classification models through a data preprocessing process. We also propose a lightweight model that can automatically classify images by adding category values by incorporating transfer learning into the image classification model and an image classification model that reduces training time and achieves high accuracy.

• Journal of Computing Science and Engineering
• /
• v.11 no.1
• /
• pp.1-8
• /
• 2017

Naive Bayes nearest neighbor (NBNN) is a simple image classifier based on identifying nearest neighbors. NBNN uses original image descriptors (e.g., SIFTs) without vector quantization for preserving the discriminative power of descriptors and has a powerful generalization characteristic. However, it has a distinct disadvantage. Its memory requirement can be prohibitively high while processing a large amount of data. To deal with this problem, we apply a spherical hashing binary code embedding technique, to compactly encode data without significantly losing classification accuracy. We also propose using an inverted index to identify nearest neighbors among binarized image descriptors. To demonstrate the benefits of our method, we apply our method to two existing NBNN techniques with an image dataset. By using 64 bit length, we are able to reduce memory 16 times with higher runtime performance and no significant loss of classification accuracy. This result is achieved by our compact encoding scheme for image descriptors without losing much information from original image descriptors.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.35 no.4
• /
• pp.231-240
• /
• 2017

This paper aims to study a method to estimate precise carbon absorption by quantification of forest information that uses accurate LiDAR data, hyperspectral image. To estimate precise carbon absorption value by using spatial data, a problem was found out of carbon absorption value estimation method with statistical method, which is already existed method, and then offered optimized carbon absorption estimation method with spatial information by analyzing with methods of compare digital aerial photogrammetry and LiDAR data. It turned out possible Precise classification and quantification in case of using LiDAR and hyperspectral image. Various classification of tree species was possible with use of LiDAR and hyperspectral image. Classification of hyperspectral image was matched in general with field survey and Mahalanobis distance classification method. Precise forest resources could be extracted using high density LiDAR data. Compared with existing method, 19.7% in forest area, 19.2% in total carbon absorption, 0.9% in absorption per unit area of difference created, and improvement was found out to be estimated precisely in international code.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.5
• /
• pp.2590-2606
• /
• 2017

Generally, research of classical image classification algorithms assume that training data and testing data are derived from the same domain with the same distribution. Unfortunately, in practical applications, this assumption is rarely met. Aiming at the problem, a domain adaption image classification approach based on multi-sparse representation is proposed in this paper. The existences of intermediate domains are hypothesized between the source and target domains. And each intermediate subspace is modeled through online dictionary learning with target data updating. On the one hand, the reconstruction error of the target data is guaranteed, on the other, the transition from the source domain to the target domain is as smooth as possible. An augmented feature representation produced by invariant sparse codes across the source, intermediate and target domain dictionaries is employed for across domain recognition. Experimental results verify the effectiveness of the proposed algorithm.

• The Transactions of the Korean Institute of Electrical Engineers
• /
• v.43 no.7
• /
• pp.1193-1205
• /
• 1994

This paper presents a fuzzy system that adaptively classifies subimages to four classes according to image activity distribution. In adaptive transform image coding, subimage classification improves the compression performance by assigning different bit maps to different classes. A conventional classification method sorts subimages by their AC energy and divides them to classes with equal number of subimages. The fuzzy system provides more flexible classification to natural images with various distribution of image details than does the conventional method. Clustering of training data in the input-output product space generated the fuzzy rules for subimage classification. The fuzzy system of small number of fuzzy rules successfully classified subimages to improve the compression performance of the transform image coding without sorting of AC energies.

• Journal of Korea Multimedia Society
• /
• v.21 no.12
• /
• pp.1417-1424
• /
• 2018

In order to improve the performance of image classifications using Convolutional Neural Networks (CNN), applying a category hierarchy to the classification can be a useful idea. However, the visual separation of object categories is very different according to the upper and lower category levels and highly uneven in image classifications. Therefore, it is doubtable whether the use of category hierarchies for classification is effective in CNN. In this paper, we have clarified whether the image classification using category hierarchies improves classification performance, and found at which level of hierarchy classification is more effective. For experiments we divided the image classification task according to the upper and lower category levels and assigned image data to each CNN model. We identified and compared the results of three classification models and analyzed them. Through the experiments, we could confirm that classification effectiveness was not improved by reduction of number of categories in a classification model. And we found that only with the re-training method in the last network layer, the performance of lower category classification was not improved although that of higher category classification was improved.

• Journal of radiological science and technology
• /
• v.45 no.1
• /
• pp.41-47
• /
• 2022

The purpose of this study is to apply a deep learning model that can distinguish lung perfusion and lung ventilation images in nuclear medicine, and to evaluate the image classification ability. Image data pre-processing was performed in the following order: image matrix size adjustment, min-max normalization, image center position adjustment, train/validation/test data set classification, and data augmentation. The convolutional neural network(CNN) structures of VGG-16, ResNet-18, Inception-ResNet-v2, and SE-ResNeXt-101 were used. For classification model evaluation, performance evaluation index of classification model, class activation map(CAM), and statistical image evaluation method were applied. As for the performance evaluation index of the classification model, SE-ResNeXt-101 and Inception-ResNet-v2 showed the highest performance with the same results. As a result of CAM, cardiac and right lung regions were highly activated in lung perfusion, and upper lung and neck regions were highly activated in lung ventilation. Statistical image evaluation showed a meaningful difference between SE-ResNeXt-101 and Inception-ResNet-v2. As a result of the study, the applicability of the CNN model for lung scintigraphy classification was confirmed. In the future, it is expected that it will be used as basic data for research on new artificial intelligence models and will help stable image management in clinical practice.

• The Journal of Korea Robotics Society
• /
• v.18 no.3
• /
• pp.293-298
• /
• 2023

The military is facing a continuous decrease in personnel, and in order to cope with potential accidents and challenges in operations, efforts are being made to reduce the direct involvement of personnel by utilizing the latest technologies. Recently, the use of various sensors related to Manned-Unmanned Teaming and artificial intelligence technologies has gained attention, emphasizing the need for flexible utilization methods. In this paper, we propose four dataset construction methods that can be used for effective training of robots that can be deployed in military operations, utilizing not only RGB image data but also data acquired from IR image sensors. Since there is no publicly available dataset that combines RGB and IR image data, we directly acquired the dataset within buildings. The input values were constructed by combining RGB and IR image sensor data, taking into account the field of view, resolution, and channel values of both sensors. We compared the proposed method with conventional RGB image data classification training using the same learning model. By employing the proposed image data fusion method, we observed improved stability in training loss and approximately 3% higher accuracy.

• Korean Journal of Remote Sensing
• /
• v.28 no.6
• /
• pp.611-622
• /
• 2012

This study proposed two multisensor fusion methods for segment-based image classification utilizing a region-growing segmentation. The proposed algorithms employ a Gaussian-PDF measure and an evidential measure respectively. In remote sensing application, segment-based approaches are used to extract more explicit information on spatial structure compared to pixel-based methods. Data from a single sensor may be insufficient to provide accurate description of a ground scene in image classification. Due to the redundant and complementary nature of multisensor data, a combination of information from multiple sensors can make reduce classification error rate. The Gaussian-PDF method defines a regional measure as the PDF average of pixels belonging to the region, and assigns a region into a class associated with the maximum of regional measure. The evidential fusion method uses two measures of plausibility and belief, which are derived from a mass function of the Beta distribution for the basic probability assignment of every hypothesis about region classes. The proposed methods were applied to the SPOT XS and ENVISAT data, which were acquired over Iksan area of of Korean peninsula. The experiment results showed that the segment-based method of evidential measure is greatly effective on improving the classification via multisensor fusion.

• Management & Information Systems Review
• /
• v.3
• /
• pp.15-45
• /
• 1999

Processing techniques of remote sensed image data using computer have been recognized very necessary techniques to all social fields, such as, environmental observation, land cultivation, resource investigation, military trend grasp and agricultural product estimation, etc. Especially, accurate classification and analysis to remote sensed image da are important elements that can determine reliability of remote sensed image data processing systems, and many researches have been processed to improve these accuracy of classification and analysis. Traditionally, remote sensed image data processing systems have been processed 2 or 3 selected bands in multiple bands, in this time, their selection criterions are statistical separability or wavelength properties. But, it have be bring up the necessity of bands selection method by data distribution characteristics than traditional bands selection by wavelength properties or statistical separability. Because data sensing environments change from multispectral environments to hyperspectral environments. In this paper for efficient data classification in multispectral bands environment, a band feature extraction method using the Rough sets theory is proposed. First, we make a look up table from training data, and analyze the properties of experimental multispectral image data, then select the efficient band using indiscernibility relation of Rough set theory from analysis results. Proposed method is applied to LANDSAT TM data on 2 June 1992. From this, we show clustering trends that similar to traditional band selection results by wavelength properties, from this, we verify that can use the proposed method that centered on data properties to select the efficient bands, though data sensing environment change to hyperspectral band environments.