• The Journal of the Korea institute of electronic communication sciences
• /
• v.18 no.3
• /
• pp.449-458
• /
• 2023

Many people think that sign language is only for people who are deaf and cannot speak, but of course it is necessary for people who want to talk with them. One of the biggest challenges in ASL(American Sign Language) alphabet recognition is the high inter-class similarities and high intra-class variance. In this paper, we proposed an architecture that can overcome these two problems, which performs similarity learning to reduces inter-class similarities and intra-class variance between images. The proposed architecture consists of the same convolutional neural network with a double configuration that shares parameters (weights and biases) and also applies the Keras API to reduce similarity learning and variance through this pathway. The similarity learning results the use of the dual CNN shows that the accuracy is improved by reducing the similarity and variability between classes by not including the poor results of the two classes.

• Journal of Korea Multimedia Society
• /
• v.21 no.5
• /
• pp.571-582
• /
• 2018

Recently, development of deep learning has shown better image classification result than human. According to recent research, a hidden layer of deep learning is deeper, and a preservation of extracted features shows good results. However, in the case of general images, the extracted features are clear and easy to sort. This study aims to classify plant leaf images. This plant leaf image has high similarity in each image. Since plant leaf images have high similarity not only between images of different species but also within the same species, classification accuracy is not increased by simply extending the hidden layer or connecting the layers. Therefore, in this paper, we tried to improve the hidden layer of the algorithm called DenseNet which shows the recent excellent classification results, and compare the results of several different modified layers. The proposed method makes it possible to classify plant leaf images collected in a natural environment more easily and accurately than conventional methods. This results in good classification of plant leaf image data including unnecessary noise obtained in a natural environment.

• Journal of Korea Multimedia Society
• /
• v.21 no.12
• /
• pp.1407-1416
• /
• 2018

The purpose of this study was to train a model for the ulna and radius bone segmentation based on Convolutional Neural Networks and to verify the segmentation model. The data consisted of 840 training data, 210 tuning data, and 200 verification data. The learning model for the ulna and radius bone bwas based on U-Net (19 convolutional and 8 maximum pooling) and trained with 8 batch sizes, 0.0001 learning rate, and 200 epochs. As a result, the average sensitivity of the training data was 0.998, the specificity was 0.972, the accuracy was 0.979, and the Dice's similarity coefficient was 0.968. In the validation data, the average sensitivity was 0.961, specificity was 0.978, accuracy was 0.972, and Dice's similarity coefficient was 0.961. The performance of deep convolutional neural network based models for the segmentation was good for ulna and radius bone.

• Journal of the Korea Institute of Military Science and Technology
• /
• v.25 no.5
• /
• pp.462-475
• /
• 2022

Different from the field of electro-optical(EO) image analysis, there has been less interest in similarity metrics between synthetic aperture radar(SAR) target images. A reliable and objective similarity analysis for SAR target images is expected to enable the verification of the SAR measurement process or provide the guidelines of target CAD modeling that can be used for simulating realistic SAR target images. For this purpose, this paper presents a similarity analysis method based on the siamese network that quantifies the subjective assessment through the distance learning of similar and dissimilar SAR target image pairs. The proposed method is applied to MSTAR SAR target images of slightly different depression angles and the resultant metrics are compared and analyzed with qualitative evaluation. Since the image similarity is somewhat related to recognition performance, the capacity of the proposed method for target recognition is further checked experimentally with the confusion matrix.

• The Journal of the Acoustical Society of Korea
• /
• v.37 no.4
• /
• pp.256-261
• /
• 2018

There are three types of noise generated inside the vehicle: BSR (Buzz, Squeak, Rattle). In this paper, we propose a classifier that automatically classifies automotive BSR noise by using features extracted from deep convolutional neural networks. In the preprocessing process, the features of above three noises are represented as noise-map using STFT (Short-time Fourier Transform) algorithm. In order to cope with the problem that the position of the actual noise is unknown in the part of the generated noise map, the noise map is divided using the sliding window method. In this paper, internal parameter of the deep convolutional neural networks is visualized using the t-SNE (t-Stochastic Neighbor Embedding) algorithm, and the misclassified data is analyzed in a qualitative way. In order to analyze the classified data, the similarity of the noise type was quantified by SSIM (Structural Similarity Index) value, and it was found that the retractor tremble sound is most similar to the normal travel sound. The classifier of the proposed method compared with other classifiers of machine learning method recorded the highest classification accuracy (99.15 %).

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.22 no.10
• /
• pp.1300-1306
• /
• 2018

The purpose of this study is to see possibility of Char2Vec as alternative of Word2Vec that most famous word embedding model in Sentence Similarity Measure Problem by Deep-Learning. In experiment, we used the Siamese Ma-LSTM recurrent neural network architecture for measure similarity two random sentences. Siamese Ma-LSTM model was implemented with tensorflow. We train each model with 200 epoch on gpu environment and it took about 20 hours. Then we compared Word2Vec based model training result with Char2Vec based model training result. as a result, model of based with Char2Vec that initialized random weight record 75.1% validation dataset accuracy and model of based with Word2Vec that pretrained with 3 million words and phrase record 71.6% validation dataset accuracy. so Char2Vec is suitable alternate of Word2Vec to optimize high system memory requirements problem.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.23 no.3
• /
• pp.193-198
• /
• 2023

We study a learner-customized lecture recommendation project using deep learning. Recommendation systems can be easily found on the web and apps, and examples using this feature include recommending feature videos by clicking users and advertising items in areas of interest to users on SNS. In this study, the sentence similarity Word2Vec was mainly used to filter twice, and the course was recommended through the Surprise library. With this system, it provides users with the desired classification of course data conveniently and conveniently. Surprise Library is a Python scikit-learn-based library that is conveniently used in recommendation systems. By analyzing the data, the system is implemented at a high speed, and deeper learning is used to implement more precise results through course steps. When a user enters a keyword of interest, similarity between the keyword and the course title is executed, and similarity with the extracted video data and voice text is executed, and the highest ranking video data is recommended through the Surprise Library.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.9
• /
• pp.1-12
• /
• 2022

Traditional methods of measuring corneal ulcers were difficult to present objective basis for diagnosis because of the subjective judgment of the medical staff through photographs taken with special equipment. In this paper, we propose a method to detect the ulcer area on a pixel basis in corneal ulcer images using a semantic segmentation model. In order to solve this problem, we performed the experiment to detect the ulcer area based on the DeepLab model which has the highest performance in semantic segmentation model. For the experiment, the training and test data were selected and the backbone network of DeepLab model which set as Xception and ResNet, respectively were evaluated and compared the performances. We used Dice similarity coefficient and IoU value as an indicator to evaluate the performances. Experimental results show that when 'crop & resized' images are added to the dataset, it segment the ulcer area with an average accuracy about 93% of Dice similarity coefficient on the DeepLab model with ResNet101 as the backbone network. This study shows that the semantic segmentation model used for object detection also has an ability to make significant results when classifying objects with irregular shapes such as corneal ulcers. Ultimately, we will perform the extension of datasets and experiment with adaptive learning methods through future studies so that they can be implemented in real medical diagnosis environment.

• Ocean and Polar Research
• /
• v.36 no.4
• /
• pp.423-435
• /
• 2014

Verifying the similarity of environmental characteristics between an artificial impact site and a preserved or reference site is necessary to quantitatively and qualitatively evaluate the environmental impact of mining activity. Although an impact site (BIS station) and a preserved site (called KOMO station) that have been selected in the Korea manganese nodule contract area may share similar environmental characteristics, similarities in terms of the water column environment between both sites has not been investigated. In this study, we compared the chemical properties of the water columns and sinking particle fluxes between BIS and KOMO stations through two observations (August 2011 and September 2012). Additionally, we observed particle fluxes at the KOMO station for five years (July 2003~July 2008) to understand long-term natural variability. Vertical distributions of water column properties such as dissolved oxygen, inorganic nutrients (N, P, Si), total organic carbon below surface layer (within the depth range of 200 m) were not considerably different between the two sites. Especially, values of water column parameters in the abyssopelagic zone from 4000 m to bottom layer (~5000 m) were very similar between the BIS and KOMO sites. Sinking particle fluxes from the two sites also showed similar seasonality. However, natural variation of particle flux at the KOMO site varied from 3.5 to $129.9mg\;m^{-2}day^{-1}$, with a distinct temporal variation originating from ENSO events (almost forty times higher than a minimum value). These results could provide valuable information to more exactly evaluate the environmental impact of mining activity on water columns.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.2
• /
• pp.19-25
• /
• 2023

In this paper, we propose a method for improving raw images captured in a low light condition based on deep learning considering the image signal processing. In the case of a smart phone camera, compared to a DSLR camera, the size of a lens or sensor is limited, so the noise increases and the reduces the quality of images in low light conditions. Existing deep learning-based low-light image processing methods create unnatural images in some cases since they do not consider the lens shading effect and white balance, which are major factors in the image signal processing. In this paper, pixel distances from the image center and channel average values are used to consider the lens shading effect and white balance with a deep learning model. Experiments with low-light images taken with a smart phone demonstrate that the proposed method achieves a higher peak signal to noise ratio and structural similarity index measure than the existing method by creating high-quality low-light images.