• Journal of the Institute of Convergence Signal Processing
• /
• v.24 no.1
• /
• pp.21-26
• /
• 2023

Knowledge distillation (KD) is a model lightening technology that transfers the knowledge of deep models to light models. Most KD methods have been developed for classification models, and there have been few KD studies in the field of super-resolution (SR). In this paper, various KD methods are applied to an SR model and their performance is compared. Specifically, we modified the loss function to apply each KD method to the SR model and conducted an experiment to learn a student model that was about 27 times lighter than the teacher model and to double the image resolution. Through the experiment, it was confirmed that some KD methods were not valid when applied to SR models, and that the performance was the highest when the relational KD and the traditional KD methods were combined.

• Journal of Korean Tunnelling and Underground Space Association
• /
• v.19 no.6
• /
• pp.915-936
• /
• 2017

In this study, current status of Korean hazard mitigation guideline for tunnel operation is summarized. It shows that requirement for CCTV installation has been gradually stricted and needs for tunnel incident detection system in conjunction with the CCTV in tunnels have been highly increased. Despite of this, it is noticed that mathematical algorithm based incident detection system, which are commonly applied in current tunnel operation, show very low detectable rates by less than 50%. The putative major reasons seem to be (1) very weak intensity of illumination (2) dust in tunnel (3) low installation height of CCTV to about 3.5 m, etc. Therefore, an attempt in this study is made to develop an deep-learning based tunnel incident detection system, which is relatively insensitive to very poor visibility conditions. Its theoretical background is given and validating investigation are undertaken focused on the moving vehicles and person out of vehicle in tunnel, which are the official major objects to be detected. Two scenarios are set up: (1) training and prediction in the same tunnel (2) training in a tunnel and prediction in the other tunnel. From the both cases, targeted object detection in prediction mode are achieved to detectable rate to higher than 80% in case of similar time period between training and prediction but it shows a bit low detectable rate to 40% when the prediction times are far from the training time without further training taking place. However, it is believed that the AI based system would be enhanced in its predictability automatically as further training are followed with accumulated CCTV BigData without any revision or calibration of the incident detection system.

• Journal of Digital Convergence
• /
• v.15 no.2
• /
• pp.173-181
• /
• 2017

In this paper, we have studied tracking as a training stage of considering the position and the scale of a person given its previous position, scale, as well as next and forward image fraction. Unlike other learning methods, CNN is thereby learning combines both time and spatial features from the image for the two consecutive frames. We introduce multiple path ways in CNN to better fuse local and global information. A creative shift-variant CNN architecture is designed so as to alleviate the drift problem when the distracting objects are similar to the target in cluttered environment. Furthermore, we employ CNNs to estimate the scale through the accurate localization of some key points. These techniques are object-independent so that the proposed method can be applied to track other types of object. The capability of the tracker of handling complex situations is demonstrated in many testing sequences. The accuracy of the SVM classifier using the features learnt by the CNN is equivalent to the accuracy of the CNN. This fact confirms the importance of automatically optimized features. However, the computation time for the classification of a person using the convolutional neural network classifier is less than approximately 1/40 of the SVM computation time, regardless of the type of the used features.

• Journal of Internet Computing and Services
• /
• v.23 no.3
• /
• pp.35-45
• /
• 2022

Even in recent years, treatment of first-aid patients is still often delayed due to a shortage of medical resources in marginalized areas. Research on automating the analysis of medical data to solve the problems of inaccessibility for medical services and shortage of medical personnel is ongoing. Computer vision-based medical inspection automation requires a lot of cost in data collection and labeling for training purposes. These problems stand out in the works of classifying lesion that are rare, or pathological features and pathogenesis that are difficult to clearly define visually. Anomaly detection is attracting as a method that can significantly reduce the cost of data collection by adopting an unsupervised learning strategy. In this paper, we propose methods for detecting abnormal images on chest X-RAY images as follows based on existing anomaly detection techniques. (1) Normalize the brightness range of medical images resampled as optimal resolution. (2) Some feature vectors with high representative power are selected in set of patch features extracted as intermediate-level from lesion-free images. (3) Measure the difference from the feature vectors of lesion-free data selected based on the nearest neighbor search algorithm. The proposed system can simultaneously perform anomaly classification and localization for each image. In this paper, the anomaly detection performance of the proposed system for chest X-RAY images of PA projection is measured and presented by detailed conditions. We demonstrate effect of anomaly detection for medical images by showing 0.705 classification AUROC for random subset extracted from the PadChest dataset. The proposed system can be usefully used to improve the clinical diagnosis workflow of medical institutions, and can effectively support early diagnosis in medically poor area.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.12
• /
• pp.29-40
• /
• 2022

In this paper, we studied a system that detects and analyzes the pathological features of diabetic retinopathy using Mask R-CNN and a Random Forest classifier. Those are one of the deep learning techniques and automatically diagnoses diabetic retinopathy. Diabetic retinopathy can be diagnosed through fundus images taken with special equipment. Brightness, color tone, and contrast may vary depending on the device. Research and development of an automatic diagnosis system using artificial intelligence to help ophthalmologists make medical judgments possible. This system detects pathological features such as microvascular perfusion and retinal hemorrhage using the Mask R-CNN technique. It also diagnoses normal and abnormal conditions of the eye by using a Random Forest classifier after pre-processing. In order to improve the detection performance of the Mask R-CNN algorithm, image augmentation was performed and learning procedure was conducted. Dice similarity coefficients and mean accuracy were used as evaluation indicators to measure detection accuracy. The Faster R-CNN method was used as a control group, and the detection performance of the Mask R-CNN method through this study showed an average of 90% accuracy through Dice coefficients. In the case of mean accuracy it showed 91% accuracy. When diabetic retinopathy was diagnosed by learning a Random Forest classifier based on the detected pathological symptoms, the accuracy was 99%.

• The Journal of Bigdata
• /
• v.7 no.1
• /
• pp.15-27
• /
• 2022

In the past five years, the fatality rate of single-vehicle accidents has been 4.7 times higher than that of all accidents, so it is necessary to establish a system that can detect and respond to single-vehicle accidents immediately. The IoT(Internet of Thing)-based real-time traffic accident recognition system proposed in this study is as following. By attaching an IoT sensor which detects the impact and vehicle ingress to the guardrail, when an impact occurs to the guardrail, the image of the accident site is analyzed through artificial intelligence technology and transmitted to a rescue organization to perform quick rescue operations to damage minimization. An IoT sensor module that recognizes vehicles entering the monitoring area and detects the impact of a guardrail and an AI-based object detection module based on vehicle image data learning were implemented. In addition, a monitoring and operation module that imanages sensor information and image data in integrate was also implemented. For the validation of the system, it was confirmed that the target values were all met by measuring the shock detection transmission speed, the object detection accuracy of vehicles and people, and the sensor failure detection accuracy. In the future, we plan to apply it to actual roads to verify the validity using real data and to commercialize it. This system will contribute to improving road safety.

• Journal of Broadcast Engineering
• /
• v.25 no.2
• /
• pp.192-199
• /
• 2020

Recently, the utilization of the object tracking algorithm based on the deep learning model is increasing. A system for tracking multiple objects in an image is typically composed of a chain form of an object detection algorithm and an object tracking algorithm. However, chain-type systems composed of several modules require a high performance computing environment and have limitations in their application to actual applications. In this paper, we propose a method that enables real-time operation in low-performance computing environment by adjusting the computational process of object detection module in the object detection-tracking chain type system.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2016.06a
• /
• pp.9-11
• /
• 2016

최근 UHDTV(ultra high definition television) 등의 고해상도 디스플레이가 시장에 등장하면서, 기존의 저해상도 FHD(full high definition) 영상을 고해상도 영상으로 변환할 수 있는 초해상화(super-resolution, SR) 기법들이 각광을 받고 있다. 그 중, 선형 매핑(linear mapping)을 사용하여 저해상도 패치(patch)로부터 고해상도 패치를 복원하는 초해상화 기법은 상대적으로 낮은 복잡도로 좋은 품질의 고해상도 영상을 생성한다. 그러나 이러한 기법은 단순한 선형 매핑을 기반으로 하기 때문에 복잡한 비선형적(nonlinear) 저해상도-고해상도 관계를 예측하기 힘든 단점이 있다. 최근 각광받는 딥러닝(deep learning) 기술은 다층(multi-layer) 네트워크를 쌓아 입력과 출력 간의 복잡한 비선형 관계를 훈련시켜 좋은 성능을 보이는데, 이를 바탕으로 본 논문에서는 다중의 레이어로 구성된 다층 선형 매핑(multi-layer linear mappings, MLLM)을 기반으로 하는 초해상화 기법을 새롭게 제안한다. 제안하는 다층 선형 매핑은 기존 선형 매핑보다 비선형적 관계를 더 잘 예측하여 높은 품질의 고해상도 영상을 생성할 수 있게 한다. 제안된 초해상화 기법은 딥러닝 기반 초해상화 기법과 필적하는 품질의 고해상도 영상을 생성하면서도 더 낮은 복잡도를 지니는 것을 확인하였다.

• Journal of Korea Multimedia Society
• /
• v.21 no.10
• /
• pp.1150-1161
• /
• 2018

Classification is widely used in medical images to categorize patients and non-patients. However, conventional classification requires a complex procedure, including some rigid steps such as pre-processing, segmentation, feature extraction, detection, and classification. In this paper, we propose a novel convolutional neural network (CNN), called LeukemiaNet, to specifically classify two different types of leukemia, including acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML), and non-cancerous patients. To extend the limited dataset, a PCA color augmentation process is utilized before images are input into the LeukemiaNet. This augmentation method enhances the accuracy of our proposed CNN architecture from 96.9% to 97.2% for distinguishing ALL, AML, and normal cell images.

• Korean Journal of Artificial Intelligence
• /
• v.5 no.2
• /
• pp.15-25
• /
• 2017

Recently, technologies are being developed to recognize and authenticate users using bioinformatics to solve information security issues. Biometric information includes face, fingerprint, iris, voice, and vein. Among them, face recognition technology occupies a large part. Face recognition technology is applied in various fields. For example, it can be used for identity verification, such as a personal identification card, passport, credit card, security system, and personnel data. In addition, it can be used for security, including crime suspect search, unsafe zone monitoring, vehicle tracking crime.In this thesis, we conducted a study to recognize faces by detecting the areas of the face through a computer webcam. The purpose of this study was to contribute to the improvement in the accuracy of Recognition of Face Based on CNN Algorithms. For this purpose, We used data files provided by github to build a face recognition model. We also created data using CNN algorithms, which are widely used for image recognition. Various photos were learned by CNN algorithm. The study found that the accuracy of face recognition based on CNN algorithms was 77%. Based on the results of the study, We carried out recognition of the face according to the distance. Research findings may be useful if face recognition is required in a variety of situations. Research based on this study is also expected to improve the accuracy of face recognition.