• Journal of Internet Computing and Services
• /
• v.20 no.4
• /
• pp.91-102
• /
• 2019

In autonomous driving systems, the ability to classify pedestrians in images captured by cameras is very important for pedestrian safety. In the past, after extracting features of pedestrians with HOG(Histogram of Oriented Gradients) or SIFT(Scale-Invariant Feature Transform), people classified them using SVM(Support Vector Machine). However, extracting pedestrian characteristics in such a handcrafted manner has many limitations. Therefore, this paper proposes a method to classify pedestrians reliably and effectively using CNN's(Convolutional Neural Network) deep features and transfer learning. We have experimented with both the fixed feature extractor and the fine-tuning methods, which are two representative transfer learning techniques. Particularly, in the fine-tuning method, we have added a new scheme, called M-Fine(Modified Fine-tuning), which divideslayers into transferred parts and non-transferred parts in three different sizes, and adjusts weights only for layers belonging to non-transferred parts. Experiments on INRIA Person data set with five CNN models(VGGNet, DenseNet, Inception V3, Xception, and MobileNet) showed that CNN's deep features perform better than handcrafted features such as HOG and SIFT, and that the accuracy of Xception (threshold = 0.5) isthe highest at 99.61%. MobileNet, which achieved similar performance to Xception and learned 80% fewer parameters, was the best in terms of efficiency. Among the three transfer learning schemes tested above, the performance of the fine-tuning method was the best. The performance of the M-Fine method was comparable to or slightly lower than that of the fine-tuningmethod, but higher than that of the fixed feature extractor method.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.2
• /
• pp.276-284
• /
• 2020

This paper proposes a new convolutional neural network (CNN) based spectrum sensing technique for cognitive radio communications. The proposed technique determines the existence of the primary user (PU) by using energy detection without any prior knowledge of the PU's signal. In the proposed method, the received signal is high-rate sampled to sense the entire spectrum bands of interest. After that, fast Fourier transform (FFT) of the signal converts the time domain signal to frequency domain spectrum and by stacking those consecutive spectrums, a 2 dimensional signal is made. The 2 dimensional signal is cut by the sensing channel bandwidth and inputted to the CNN. The CNN determines the existence of the primary user. Since there are only two states (existence or non-existence), binary classification CNN is used. The performance of the proposed method is examined through computer simulation and indoor experiment. According to the results, the proposed method outperforms the conventional threshold-based method by over 2 dB.

• Korean Journal of Remote Sensing
• /
• v.38 no.6_3
• /
• pp.1777-1788
• /
• 2022

The reservoir is a major water supply source in the domestic agricultural environment, and the monitoring of water storage of reservoirs is important for the utilization and management of agricultural water resource. Remote sensing via satellite imagery can be an effective method for regular monitoring of widely distributed objects such as reservoirs, and in this study, image classification and image segmentation algorithms are applied to Sentinel-1 Synthetic Aperture Radar (SAR) imagery for water body detection in 53 reservoirs in South Korea. Six algorithms are used: Neural Network (NN), Support Vector Machine (SVM), Random Forest (RF), Otsu, Watershed (WS), and Chan-Vese (CV), and the results of water body detection are evaluated with in-situ images taken by drones. The correlations between the in-situ water surface area and detected water surface area from each algorithm are NN 0.9941, SVM 0.9942, RF 0.9940, Otsu 0.9922, WS 0.9709, and CV 0.9736, and the larger the scale of reservoir, the higher the linear correlation was. WS showed low recall due to the undetected water bodies, and NN, SVM, and RF showed low precision due to over-detection. For water body detection through SAR imagery, we found that aquatic plants and artificial structures can be the error factors causing undetection of water body.

• Journal of the Korean Society of Radiology
• /
• v.16 no.4
• /
• pp.453-462
• /
• 2022

In this paper, the artificial intelligence (AI) technology used in the medical image analysis field was analyzed through a literature review. Literature searches were conducted on PubMed, ResearchGate, Google and Cochrane Review using the key word. Through literature search, 114 abstracts were searched, and 98 abstracts were reviewed, excluding 16 duplicates. In the reviewed literature, AI is applied in classification, localization, disease detection, disease segmentation, and fit degree of registration images. In machine learning (ML), prior feature extraction and inputting the extracted feature values into the neural network have disappeared. Instead, it appears that the neural network is changing to a deep learning (DL) method with multiple hidden layers. The reason is thought to be that feature extraction is processed in the DL process due to the increase in the amount of memory of the computer, the improvement of the calculation speed, and the construction of big data. In order to apply the analysis of medical images using AI to medical care, the role of physicians is important. Physicians must be able to interpret and analyze the predictions of AI algorithms. Additional medical education and professional development for existing physicians is needed to understand AI. Also, it seems that a revised curriculum for learners in medical school is needed.

• Nuclear Engineering and Technology
• /
• v.55 no.1
• /
• pp.339-352
• /
• 2023

Prediction of the time-related traits of pebble flow inside pebble-bed HTGRs is of great significance for reactor operation and design. In this work, an image-driven approach with the aid of a convolutional neural network (CNN) is proposed to predict the remaining time of initially loaded pebbles and the time interval of paired flow images of the pebble bed. Two types of strategies are put forward: one is adding FC layers to the classic classification CNN models and using regression training, and the other is CNN-based deep expectation (DEX) by regarding the time prediction as a deep classification task followed by softmax expected value refinements. The current dataset is obtained from the discrete element method (DEM) simulations. Results show that the CNN-aided models generally make satisfactory predictions on the remaining time with the determination coefficient larger than 0.99. Among these models, the VGG19+DEX performs the best and its CumScore (proportion of test set with prediction error within 0.5s) can reach 0.939. Besides, the remaining time of additional test sets and new cases can also be well predicted, indicating good generalization ability of the model. In the task of predicting the time interval of image pairs, the VGG19+DEX model has also generated satisfactory results. Particularly, the trained model, with promising generalization ability, has demonstrated great potential in accurately and instantaneously predicting the traits of interest, without the need for additional computational intensive DEM simulations. Nevertheless, the issues of data diversity and model optimization need to be improved to achieve the full potential of the CNN-aided prediction tool.

• Journal of the Korean Geotechnical Society
• /
• v.39 no.10
• /
• pp.41-48
• /
• 2023

The presence of abnormalities in the subgrade of roads poses safety risks to users and results in significant maintenance costs. In this study, we aimed to experimentally evaluate the temperature distributions in abnormal areas of subgrade materials using infrared cameras and analyze the data with machine learning techniques. The experimental site was configured as a cubic shape measuring 50 cm in width, length, and depth, with abnormal areas designated for water and air. Concrete blocks covered the upper part of the site to simulate the pavement layer. Temperature distribution was monitored over 23 h, from 4 PM to 3 PM the following day, resulting in image data and numerical temperature values extracted from the middle of the abnormal area. The temperature difference between the maximum and minimum values measured 34.8℃ for water, 34.2℃ for air, and 28.6℃ for the original subgrade. To classify conditions in the measured images, we employed the image analysis method of a convolutional neural network (CNN), utilizing ResNet-101 and SqueezeNet networks. The classification accuracies of ResNet-101 for water, air, and the original subgrade were 70%, 50%, and 80%, respectively. SqueezeNet achieved classification accuracies of 60% for water, 30% for air, and 70% for the original subgrade. This study highlights the effectiveness of CNN algorithms in analyzing subgrade properties and predicting subsurface conditions.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.40 no.1
• /
• pp.19-29
• /
• 2003

This paper presents a novel hybrid classifier for face recognition with artificially generated virtual training samples. We utilize both the nearest neighbor approach in feature angle space and a connectionist model to obtain a synergy effect by combining the results of two heterogeneous classifiers. First, a classifier called the nearest feature angle (NFA), based on angular information, finds the most similar feature to the query from a given training set. Second, a classifier has been developed based on the recall of stored frontal projection of the query feature. It uses a frontal recall network (FRN) that finds the most similar frontal one among the stored frontal feature set. For FRN, we used an ensemble neural network consisting of multiple multiplayer perceptrons (MLPs), each of which is trained independently to enhance generalization capability. Further, both classifiers used the virtual training set generated adaptively, according to the spatial distribution of each person's training samples. Finally, the results of the two classifiers are combined to comprise the best matching class, and a corresponding similarit measure is used to make the final decision. The proposed classifier achieved an average classification rate of 96.33% against a large group of different test sets of images, and its average error rate is 61.5% that of the nearest feature line (NFL) method, and achieves a more robust classification performance.

• International Journal of Computer Science & Network Security
• /
• v.24 no.8
• /
• pp.14-20
• /
• 2024

Traffic sign recognition is an essential feature of intelligent transportation systems and Advanced Driver Assistance Systems (ADAS), which are necessary for improving road safety and advancing the development of autonomous cars. This research investigates the incorporation of the YOLOv9 model into traffic sign recognition systems, utilizing its sophisticated functionalities such as Programmable Gradient Information (PGI) and Generalized Efficient Layer Aggregation Network (GELAN) to tackle enduring difficulties in object detection. We employed a publically accessible dataset obtained from Roboflow, which consisted of 3130 images classified into five distinct categories: speed_40, speed_60, stop, green, and red. The dataset was separated into training (68%), validation (21%), and testing (12%) subsets in a methodical manner to ensure a thorough examination. Our comprehensive trials have shown that YOLOv9 obtains a mean Average Precision (mAP@0.5) of 0.959, suggesting exceptional precision and recall for the majority of traffic sign classes. However, there is still potential for improvement specifically in the red traffic sign class. An analysis was conducted on the distribution of instances among different traffic sign categories and the differences in size within the dataset. This analysis aimed to guarantee that the model would perform well in real-world circumstances. The findings validate that YOLOv9 substantially improves the precision and dependability of traffic sign identification, establishing it as a dependable option for implementation in intelligent transportation systems and ADAS. The incorporation of YOLOv9 in real-world traffic sign recognition and classification tasks demonstrates its promise in making roadways safer and more efficient.

• Journal of the Korean Society for Nondestructive Testing
• /
• v.21 no.5
• /
• pp.510-517
• /
• 2001

This paper describes an initial study for the application of eddy current pattern recognition approaches to more realistic flaw characterization in steam generator tubes. For this purpose, finite-element model-based theoretical eddy current testing (ECT) signals are simulated from 5 types of OD flaws with the variation in flaw size parameters and testing frequency. In addition, three kinds of software are developed for the convenience in the application of steps in pattern recognition approaches such as feature extraction feature selection and classification by probabilistic neural networks (PNNs). The cross point of the ECT signals simulated from flaws with non-symmetric cross-sections shows the deviation from the origin of the impedance plane. New features taking advantages of this phenomenon are added to complete the feature set with a total of 18 features. Then, classification with PNNs are performed based on this feature set. The PNN classifiers show high performance for the identification of symmetry in the cross-section of a flaw. However, they show very limited success in the interrogation of the sharpness of flaw tips.

• Journal of KIISE:Databases
• /
• v.28 no.3
• /
• pp.279-300
• /
• 2001

Data mining plays an important role in knowledge discovery process and usually various existing algorithms are selected for the specific purpose of the mining. Currently, data mining techniques are actively to the statistics, business, electronic commerce, biology, and medical area and currently numerous algorithms are being researched and developed for these applications. However, in a long run, only a few algorithms, which are well-suited to specific applications with excellent performance in large database, will survive. So it is reasonable to focus our effort on those selected algorithms in the future. This paper classifies about 30 existing algorithms into 7 categories - association rule, clustering, neural network, decision tree, genetic algorithm, memory-based reasoning, and bayesian network. First of all, this work analyzes systematic hierarchy and characteristics of algorithms and we present 14 criteria for classifying the algorithms and the results based on this criteria. Finally, we propose the best algorithms among some comparable algorithms with different features and performances. The result of this paper can be used as a guideline for data mining researches as well as field applications of data mining.