• Journal of IKEEE
• /
• 제22권4호
• /
• pp.1079-1087
• /
• 2018

Recently, computer vision field based deep learning artificial intelligence has become a hot topic among various image analysis boundaries. In this study, flames are detected in fire images using the Faster R-CNN algorithm, which is used to detect objects within the image, among various image recognition algorithms based on deep learning. In order to improve fire detection accuracy through a small amount of data sets in the learning process, we use image augmentation techniques, and learn image augmentation by dividing into 6 types and compare accuracy, precision and detection rate. As a result, the detection rate increases as the type of image augmentation increases. However, as with the general accuracy and detection rate of other object detection models, the false detection rate is also increased from 10% to 30%.

• International Journal of Advanced Culture Technology
• /
• 제6권2호
• /
• pp.123-128
• /
• 2018

Learning data is composed of 100 characters with 10 different fonts, and test data is composed of 10 characters with a new font that is not used for the learning data. In order to consider the variety of learning data with several different fonts, 10 learning masks are constructed by accumulating pixel values of same characters with 10 different fonts. This process eliminates minute difference of characters with different fonts. After finding maximum values of learning masks, test data is expanded by multiplying these maximum values to the test data. The algorithm calculates sum of differences of two corresponding pixel values of the expanded test data and the learning masks. The learning mask with the smallest value among these 10 calculated sums is selected as the result of the recognition process for the test data. The proposed algorithm can recognize various types of fonts, and the learning data can be modified easily by adding a new font. Also, the recognition process is easy to understand, and the algorithm makes satisfactory results for character recognition.

• Progress in Medical Physics
• /
• 제30권2호
• /
• pp.39-48
• /
• 2019

Deep learning has been applied to various medical data. In particular, current deep learning models exhibit remarkable performance at specific tasks, sometimes offering higher accuracy than that of experts for discriminating specific diseases from medical images. The current status of deep learning applications to molecular imaging can be divided into a few subtypes in terms of their purposes: differential diagnostic classification, enhancement of image acquisition, and image-based quantification. As functional and pathophysiologic information is key to molecular imaging, this review will emphasize the need for accurate biomarker acquisition by deep learning in molecular imaging. Furthermore, this review addresses practical issues that include clinical validation, data distribution, labeling issues, and harmonization to achieve clinically feasible deep learning models. Eventually, deep learning will enhance the role of theranostics, which aims at precision targeting of pathophysiology by maximizing molecular imaging functional information.

• Journal of Korea Multimedia Society
• /
• 제21권12호
• /
• pp.1387-1395
• /
• 2018

Core algorithm of deep learning Convolutional Neural Network(CNN) shows better performance than other machine learning algorithms. However, if there is not sufficient data, CNN can not achieve satisfactory performance even if the classifier is excellent. In this situation, it has been proven that the use of transfer learning can have a great effect. In this paper, we apply two transition learning methods(freezing, retraining) to three CNN models(ResNet-50, Inception-V3, DenseNet-121) and compare and analyze how the classification performance of CNN changes according to the methods. As a result of statistical significance test using various evaluation indicators, ResNet-50, Inception-V3, and DenseNet-121 differed by 1.18 times, 1.09 times, and 1.17 times, respectively. Based on this, we concluded that the retraining method may be more effective than the freezing method in case of transition learning in image classification problem.

• Journal of the Korean Association of Geographic Information Studies
• /
• 제24권1호
• /
• pp.1-11
• /
• 2021

Detection of surface water features including river, wetland, reservoir from the satellite imagery can be utilized for sustainable management and survey of water resources. This research compared the water indices derived from the multispectral bands and the machine learning technique for detecting the surface water features from he Landsat-8 satellite image acquired in Daegu through the following steps. First, the NDWI(Normalized Difference Water Index) image and the MNDWI(Modified Normalized Difference Water Index) image were separately generated using the multispectral bands of the given Landsat-8 satellite image, and the two binary images were generated from these NDWI and MNDWI images, respectively. Then SVM(Support Vector Machine), the widely used machine learning techniques, were employed to generate the land cover image and the binary image was also generated from the generated land cover image. Finally the error matrices were used for measuring the accuracy of the three binary images for detecting the surface water features. The statistical results showed that the binary image generated from the MNDWI image(84%) had the relatively low accuracy than the binary image generated from the NDWI image(94%) and generated by SVM(96%). And some misclassification errors occurred in all three binary images where the land features were misclassified as the surface water features because of the shadow effects.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 한국정보통신학회 2021년도 추계학술대회
• /
• pp.551-552
• /
• 2021

In this paper, we present a multi-class image scene classification method based on transformation learning. ImageNet classifies multiple classes of natural scene images by relying on pre-trained network models on large image datasets. In the experiment, we obtained excellent results by classifying the optimized ResNet model on Kaggle's Intel Image Classification data set.

• Journal of Korea Multimedia Society
• /
• 제9권5호
• /
• pp.658-670
• /
• 2006

As the educational media tends to be more digitalized and individualized, the learning paradigm is dramatically changing into e-learning. Existing on-line courseware gives a learner more chances to learn when they are home with their own PCs. However, it is of little use when they are away from their digital media. Also, it is very labor-intensive to convert the original off-line contents to on-line contents. This paper proposes education mobile contents(EMC) that can supply the learners with dynamic interactions using various multimedia information by recognizing real images of off-line contents using mobile devices. Content-based image retrieval based on object shapes is used to recognize the real image, and shapes are represented by differential chain code with estimated new starting points to obtain rotation-invariant representation, which is fitted to computational resources of mobile devices with low resolution camera. Moreover we use a dynamic time warping method to recognize the object shape, which compensates scale variations of an object. The EMC can provide learners with quick and accurate on-line contents on off-line ones using mobile devices without limitations of space.

• Journal of IKEEE
• /
• 제26권3호
• /
• pp.468-474
• /
• 2022

In this paper, we propose a method to implement a real-time fire alarm system using deep learning. The deep learning image dataset for fire alarms acquired 1,500 sheets through the Internet. If various images acquired in a daily environment are learned as they are, there is a disadvantage that the learning accuracy is not high. In this paper, we propose a fire image data expansion method to improve learning accuracy. The data augmentation method learned a total of 2,100 sheets by adding 600 pieces of learning data using brightness control, blurring, and flame photo synthesis. The expanded data using the flame image synthesis method had a great influence on the accuracy improvement. A real-time fire detection system is a system that detects fires by applying deep learning to image data and transmits notifications to users. An app was developed to detect fires by analyzing images in real time using a model custom-learned from the YOLO V4 TINY model suitable for the Edge AI system and to inform users of the results. Approximately 10% accuracy improvement can be obtained compared to conventional methods when using the proposed data.

• Journal of the Semiconductor & Display Technology
• /
• 제22권1호
• /
• pp.11-17
• /
• 2023

This paper proposes a lightweight deep learning network for detecting an image splicing forgery. The research on image forgery detection using CNN, a deep learning network, and research on detecting and localizing forgery in pixel units are in progress. Among them, CAT-Net, which learns the discrete cosine transform coefficients of images together with images, was released in 2022. The DCT coefficients presented by CAT-Net are combined with the JPEG artifact learning module and the backbone model as pre-learning, and the weights are fixed. The dataset used for pre-training is not included in the public dataset, and the backbone model has a relatively large number of network parameters, which causes overfitting in a small dataset, hindering generalization performance. In this paper, this learning module is designed to learn the characterization depending on the DCT domain in real-time during network training without pre-training. The DCT RRU-Net proposed in this paper is a network that combines RRU-Net which detects forgery by learning only images and JPEG artifact learning module. It is confirmed that the network parameters are less than those of CAT-Net, the detection performance of forgery is better than that of RRU-Net, and the generalization performance for various datasets improves through the network architecture and training method of DCT RRU-Net.

• Journal of Advanced Navigation Technology
• /
• 제25권6호
• /
• pp.569-575
• /
• 2021

Deep learning shows differences in prediction performance depending on data quality and model. This study uses various input data and multiple deep learning models to build an optimal deep learning model for predicting solar radiation, which has the most influence on power generation prediction. did. As the input data, the weather data of the Korea Meteorological Administration and the clairvoyant meteorological image were used by segmenting the image of the Korea Meteorological Agency. , comparative evaluation, and predicting solar radiation by constructing multiple deep learning models connecting the models with the best error rate in each model. As an experimental result, the RMSE of model A, which is a multiple deep learning model, was 0.0637, the RMSE of model B was 0.07062, and the RMSE of model C was 0.06052, so the error rate of model A and model C was better than that of a single model. In this study, the model that connected two or more models through experiments showed improved prediction rates and stable learning results.