• Computers and Concrete
• /
• v.32 no.6
• /
• pp.615-623
• /
• 2023

Crack detection is an essential method to ensure the safety of dam concrete structures. Low-quality crack images of dam concrete structures limit the application of neural network methods in crack detection. This research proposes a modified attentional mechanism model to reduce the disturbance caused by uneven light, shadow, and water spots in crack images. Also, the focal loss function solves the small ratio of crack information. The dataset collects from the network, laboratory and actual inspection dataset of dam concrete structures. This research proposes a novel method for crack detection of dam concrete structures based on the U-Net neural network, namely AF-UNet. A mutual comparison of OTSU, Canny, region growing, DeepLab V3+, SegFormer, U-Net, and AF-UNet (proposed) verified the detection accuracy. A binocular camera detects cracks in the experimental scene. The smallest measurement width of the system is 0.27 mm. The potential goal is to achieve real-time detection and localization of cracks in dam concrete structures.

• Journal of the Semiconductor & Display Technology
• /
• v.23 no.1
• /
• pp.110-116
• /
• 2024

Despite the continuous advancement of science and technology, fire accidents continue to occur without decreasing over time, so there is a constant need for a system that can accurately detect fires at an early stage. However, because most existing fire detection systems detect fire in the early stage of combustion when smoke is generated, rapid fire prevention actions may be delayed. Therefore we propose an early fire detection system that can perform early fire detection at a reasonable cost using LSTM, a deep learning model based on multi-gas sensors with high selectivity in the early stage of decomposition rather than the smoke generation stage. This system combines multiple gas sensors to achieve faster detection speeds than traditional sensors. In addition, through window sliding techniques and model light-weighting, the false alarm rate is low while maintaining the same high accuracy as existing deep learning. This shows that the proposed fire early detection system is a meaningful research in the disaster and engineering fields.

• Journal of Korean Academy of Nursing
• /
• v.42 no.3
• /
• pp.424-433
• /
• 2012

Purpose: To investigate the accuracy, precision and validity of fever detection of tympanic membrane (TM), temporal artery (TA) and axillary temperature (AT) compared with pulmonary artery temperature (PA). Methods: Repeated-measures design was conducted for one year on 83 adult cardiac care unit patients with pulmonary artery catheters after open heart surgery. Sequential temperature measurements were taken three times at 20-minute intervals. Accuracy, precision, repeatability, and validity of fever detection were analyzed. Results: Mean pulmonary artery temperature was $37.04^{\circ}C$ (SD $0.70^{\circ}C$). The mean (SD) offsets from PA, with the mean reflecting accuracy and SD reflecting precision, were $-1.31^{\circ}C$ ($0.75^{\circ}C$) for TA, $-0.20^{\circ}C$ ($0.24^{\circ}C$) for TM, and $-0.97^{\circ}C$ ($0.64^{\circ}C$) for AT. Percentage of pairs with differences within ${\pm}0.5^{\circ}C$ was 9.6% for TA, 19.7% for AT, and 91.6% for TM. Repeated measurements with all three methods had mean SD values within $0.04^{\circ}C$. Sensitivity, specificity, and positive and negative predictive values of tympanic measurements were 0.76, 1.0, and 1.0, and 0.90, respectively. Conclusion: Results show that TM best reflects PA, and is most consistent, accurate, and precise. AT tends to underestimate PA, and TA is least accurate and precise. Therefore tympanic membrane measurement is a reliable alternative to other non-invasive methods of measuring temperatures.

• Journal of IKEEE
• /
• v.26 no.3
• /
• 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.

• Korean Journal of Remote Sensing
• /
• v.39 no.4
• /
• pp.425-440
• /
• 2023

Monitoring water surface has become one of the most prominent areas of research in addressing environmental challenges.Accurate and automated detection of watersurface in remote sensing imagesis crucial for disaster prevention, urban planning, and water resource management, particularly for a country where water plays a vital role in human life. However, achieving precise detection poses challenges. Previous studies have explored different approaches,such as analyzing water indexes, like normalized difference water index (NDWI) derived from satellite imagery's visible or infrared bands and using k-means clustering analysis to identify land cover patterns and segment regions based on similar attributes. Nonetheless, challenges persist, notably distinguishing between waterspectralsignatures and cloud shadow or terrain shadow. In thisstudy, our objective is to enhance the precision of water surface detection by constructing a comprehensive water database (DB) using existing digital and land cover maps. This database serves as an initial assumption for automated water index analysis. We utilized 1:5,000 and 1:25,000 digital maps of Korea to extract water surface, specifically rivers, lakes, and reservoirs. Additionally, the 1:50,000 and 1:5,000 land cover maps of Korea aided in the extraction process. Our research demonstrates the effectiveness of utilizing a water DB product as our first approach for efficient water surface extraction from satellite images, complemented by our second and third approachesinvolving NDWI analysis and k-means analysis. The image segmentation and binary mask methods were employed for image analysis during the water extraction process. To evaluate the accuracy of our approach, we conducted two assessments using reference and ground truth data that we made during this research. Visual interpretation involved comparing our results with the global surface water (GSW) mask 60 m resolution, revealing significant improvements in quality and resolution. Additionally, accuracy assessment measures, including an overall accuracy of 90% and kappa values exceeding 0.8, further support the efficacy of our methodology. In conclusion, thisstudy'sresults demonstrate enhanced extraction quality and resolution. Through comprehensive assessment, our approach proves effective in achieving high accuracy in delineating watersurfaces from satellite images.

• Proceedings of the Korean Society of Near Infrared Spectroscopy Conference
• /
• 2001.06a
• /
• pp.1248-1248
• /
• 2001

Mastitis is a major problem for the global dairy industry and causes substantial economic losses from decreasing milk production and considerable compositional changes in milk, reducing milk quality. The potential of near infrared (NIR) spectroscopy in the region from 1100 to 2500nm and chemometric method for classification to detect milk from mastitic cows was investigated. A total of 189 milk samples from 7 Holstein cows were collected for 27 days, consecutively, and analyzed for somatic cells (SCC). Three of the cows were healthy, and the rest had mastitis periods during the experiment. NIR transflectance milk spectra were obtained by the InfraAlyzer 500 spectrophotometer in the spectral range from 1100 to 2500nm. All samples were divided into calibration set and test set. Class variable was assigned for each sample as follow: healthy (class 1) and mastitic (class 2), based on milk SCC content. The classification of the samples was performed using soft independent modeling of class analogy (SIMCA) and different spectral data pretreatment. Two concentration of SCC - 200 000 cells/ml and 300 000 cells/ml, respectively, were used as thresholds fer separation of healthy and mastitis cows. The best detection accuracy was found for models, obtained using 200 000 cells/ml as threshold and smoothed absorbance data - 98.41% from samples in the calibration set and 87.30% from the samples in the independent test set were correctly classified. SIMCA results for classes, based on 300 000 cells/ml threshold, showed a little lower accuracy of classification. The analysis of changes in the loading of first PC factor for group of healthy milk and group of mastitic milk showed, that separation between classes was indirect and based on influence of mastitis on the milk components. The accuracy of mastitis detection by SIMCA method, based on NIR spectra of milk would allow health screening of cows and differentiation between healthy and mastitic milk samples. Having SIMCA models, mastitis detection would be possible by using only DIR spectra of milk, without any other analyses.

• Imaging Science in Dentistry
• /
• v.44 no.2
• /
• pp.89-94
• /
• 2014

Purpose: Vertical root fracture (VRF) is a common complication in endodontically treated teeth. Considering the poor prognosis of VRF, a reliable and valid detection method is necessary. Cone beam computed tomography (CBCT) has been reported to be a reliable tool for the detection of VRF; however, the presence of metallic intracanal posts can decrease the diagnostic values of CBCT systems. This study evaluated and compared the effects of intracanal stainless steel or titanium posts on the sensitivity, specificity, and accuracy of VRF detection using a NewTom VG CBCT system. Materials and Methods: Eighty extracted single-rooted teeth were selected and sectioned at the cemento-enamel junction. The roots were divided into two groups of 40. Root fracture was induced in the test group by using an Instron machine, while the control group was kept intact. Roots were randomly embedded in acrylic blocks and radiographed with the NewTom VG, both with titanium and stainless steel posts and also without posts. Sensitivity, specificity, and accuracy values were calculated as compared to the gold standard. Results: The sensitivity, specificity, and accuracy of VRF diagnosis were significantly lower in teeth with stainless steel and titanium posts than in those without posts. Interobserver agreement was the highest in teeth without posts, followed by stainless steel posts, and then titanium posts. Conclusion: Intracanal posts significantly decreased the VRF diagnostic values of CBCT. The stainless steel posts decreased the diagnostic values more than the titanium posts.

• Journal of Korean Society of Forest Science
• /
• v.109 no.1
• /
• pp.23-30
• /
• 2020

The use of increased computing power, machine learning, and deep learning techniques have dramatically increased in various sectors. In particular, image detection algorithms are broadly used in forestry and remote sensing areas to identify forest types and tree species. However, in South Korea, machine learning has rarely, if ever, been applied in forestry image detection, especially to classify tree species. This study integrates the application of machine learning and forest image detection; specifically, we compared the ability of two machine learning data collection methods, namely image data captured by forest experts (D1) and web-crawling (D2), to automate the classification of five trees species. In addition, two methods of characterization to train/test the system were investigated. The results indicated a significant difference in classification accuracy between D1 and D2: the classification accuracy of D1 was higher than that of D2. In order to increase the classification accuracy of D2, additional data filtering techniques were required to reduce the noise of uncensored image data.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.1
• /
• pp.69-74
• /
• 2021

Internet access is also increasing as online activities increase due to the influence of Corona 19. However, network attacks are also diversifying by malicious users, and DDoS among the attacks are increasing year by year. These attacks are detected by intrusion detection systems and can be prevented at an early stage. Various data sets are used to verify intrusion detection algorithms, but in this paper, CICIDS2017, the latest traffic, is used. DDoS attack traffic was analyzed using the decision tree. In this paper, we analyzed the traffic by using the decision tree. Through the analysis, a decisive feature was found, and the accuracy of the decisive feature was confirmed by proceeding the decision tree to prove the accuracy of detection. And the contents of false positive and false negative traffic were analyzed. As a result, learning the feature and the two features showed that the accuracy was 98% and 99.8% respectively.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.5
• /
• pp.1310-1338
• /
• 2023

As Internet of Things (IoT) applications and devices rapidly grow, cyber-attacks on IoT networks/systems also have an increasing trend, thus increasing the threat to security and privacy. Botnet is one of the threats that dominate the attacks as it can easily compromise devices attached to an IoT networks/systems. The compromised devices will behave like the normal ones, thus it is difficult to recognize them. Several intelligent approaches have been introduced to improve the detection accuracy of this type of cyber-attack, including deep learning and machine learning techniques. Moreover, dimensionality reduction methods are implemented during the preprocessing stage. This research work proposes deep Autoencoder dimensionality reduction method combined with Artificial Neural Network (ANN) classifier as botnet detection system for IoT networks/systems. Experiments were carried out using 3- layer, 4-layer and 5-layer pre-processing data from the MedBIoT dataset. Experimental results show that using a 5-layer Autoencoder has better results, with details of accuracy value of 99.72%, Precision of 99.82%, Sensitivity of 99.82%, Specificity of 99.31%, and F1-score value of 99.82%. On the other hand, the 5-layer Autoencoder model succeeded in reducing the dataset size from 152 MB to 12.6 MB (equivalent to a reduction of 91.2%). Besides that, experiments on the N_BaIoT dataset also have a very high level of accuracy, up to 99.99%.