• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.5
• /
• pp.217-228
• /
• 2023

Video instance segmentation is an intelligent visual task with high complexity because it not only requires object instance segmentation for each image frame constituting a video, but also requires accurate tracking of instances throughout the frame sequence of the video. In special, human instance segmentation in drama videos has an unique characteristic that requires accurate tracking of several main characters interacting in various places and times. Also, it is also characterized by a kind of the class imbalance problem because there is a significant difference between the frequency of main characters and that of supporting or auxiliary characters in drama videos. In this paper, we introduce a new human instance datatset called MHIS, which is built upon drama videos, Miseang, and then propose a novel video data augmentation method, CDVA, in order to overcome the data imbalance problem between character classes. Different from the previous video data augmentation methods, the proposed CDVA generates more realistic augmented videos by deciding the optimal location within the background clip for a target human instance to be inserted with taking rich spatio-temporal context embedded in videos into account. Therefore, the proposed augmentation method, CDVA, can improve the performance of a deep neural network model for video instance segmentation. Conducting both quantitative and qualitative experiments using the MHIS dataset, we prove the usefulness and effectiveness of the proposed video data augmentation method.

• Journal of Korean Society on Water Environment
• /
• v.39 no.1
• /
• pp.1-8
• /
• 2023

Monitoring and prediction of water quality are essential for effective river pollution prevention and water quality management. In this study, a multi-classification model was developed to predict chlorophyll-a (Chl-a) level in rivers. A model was developed using CatBoost, a novel ensemble machine learning algorithm. The model was developed using hourly field monitoring data collected from January 1 to December 31, 2015. For model development, chl-a was classified into class 1 (Chl-a≤10 ㎍/L), class 2 (10<Chl-a≤50 ㎍/L), and class 3 (Chl-a>50 ㎍/L), where the number of data used for the model training were 27,192, 11,031, and 511, respectively. The macro averages of precision, recall, and F1-score for the three classes were 0.58, 0.58, and 0.58, respectively, while the weighted averages were 0.89, 0.90, and 0.89, for precision, recall, and F1-score, respectively. The model showed relatively poor performance for class 3 where the number of observations was much smaller compared to the other two classes. The imbalance of data distribution among the three classes was resolved by using the synthetic minority over-sampling technique (SMOTE) algorithm, where the number of data used for model training was evenly distributed as 26,868 for each class. The model performance was improved with the macro averages of precision, rcall, and F1-score of the three classes as 0.58, 0.70, and 0.59, respectively, while the weighted averages were 0.88, 0.84, and 0.86 after SMOTE application.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.16 no.5
• /
• pp.331-337
• /
• 2023

In this paper, we propose a deep learning model to detect lumbar 3 (L3) CT images to determine the occurrence and degree of sarcopenia. In addition, we would like to propose an optimization technique that uses oversampling ratio and class weight as design parameters to address the problem of performance degradation due to data imbalance between L3 level and non-L3 level portions of CT data. In order to train and test the model, a total of 150 whole-body CT images of 104 prostate cancer patients and 46 bladder cancer patients who visited Gangneung Asan Medical Center were used. The deep learning model used ResNet50, and the design parameters of the optimization technique were selected as six types of model hyperparameters, data augmentation ratio, and class weight. It was confirmed that the proposed optimization-based L3 level extraction model reduced the median L3 error by about 1.0 slices compared to the control model (a model that optimized only 5 types of hyperparameters). Through the results of this study, accurate L3 slice detection was possible, and additionally, we were able to present the possibility of effectively solving the data imbalance problem through oversampling through data augmentation and class weight adjustment.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.49 no.10
• /
• pp.883-892
• /
• 2021

This paper provides a process for developing deep learning-based aerial object detection models that can run in realtime on onboard. To improve object detection performance, we pre-process and augment the training data in the training stage. In addition, we perform transfer learning and apply a weighted cross-entropy method to reduce the variations of detection performance for each class. To improve the inference speed, we have generated inference acceleration engines with quantization. Then, we analyze the real-time performance and detection performance on custom aerial image dataset to verify generalization.

• Journal of Drive and Control
• /
• v.19 no.4
• /
• pp.104-109
• /
• 2022

For autonomous driving research using AI, datasets collected from road environments play an important role. In other countries, various datasets such as CityScapes, A2D2, and BDD have already been released, but datasets suitable for the domestic road environment still need to be provided. This paper analyzed and verified the dataset reflecting the Korean driving environment. In order to verify the training dataset, the class imbalance was confirmed by comparing the number of pixels and instances of the dataset. A similar A2D2 dataset was trained with the same deep learning model, ConvNeXt, to compare and verify the constructed dataset. IoU was compared for the same class between two datasets with ConvNeXt and mIoU was compared. In this paper, it was confirmed that the collected dataset reflecting the driving environment of Korea is suitable for learning.

• International Journal of Computer Science & Network Security
• /
• v.23 no.8
• /
• pp.190-198
• /
• 2023

To enhance customer satisfaction for higher profits, an e-commerce sector can establish a continuous relationship and acquire new customers. Utilize machine-learning models to analyse their customer's behavioural evidence to produce their competitive advantage to the e-commerce platform by helping to improve overall satisfaction. These models will forecast customers who will churn and churn causes. Forecasts are used to build unique business strategies and services offers. This work is intended to develop a machine-learning model that can accurately forecast retainable customers of the entire e-commerce customer data. Developing predictive models classifying different imbalanced data effectively is a major challenge in collected data and machine learning algorithms. Build a machine learning model for solving class imbalance and forecast customers. The satisfaction accuracy is used for this research as evaluation metrics. This paper aims to enable to evaluate the use of different machine learning models utilized to forecast satisfaction. For this research paper are selected three analytical methods come from various classifications of learning. Classifier Selection, the efficiency of various classifiers like Random Forest, Logistic Regression, SVM, and Gradient Boosting Algorithm. Models have been used for a dataset of 8000 records of e-commerce websites and apps. Results indicate the best accuracy in determining satisfaction class with both gradient-boosting algorithm classifications. The results showed maximum accuracy compared to other algorithms, including Gradient Boosting Algorithm, Support Vector Machine Algorithm, Random Forest Algorithm, and logistic regression Algorithm. The best model developed for this paper to forecast satisfaction customers and accuracy achieve 88 %.

• Journal of the Korea Society of Computer and Information
• /
• v.22 no.2
• /
• pp.9-19
• /
• 2017

Useful continuous sensing applications are increasingly emerging as a new class of mobile applications. Meanwhile, open, multi-use sensor devices are newly adopted beyond smartphones, and provide huge opportunities to expand potential application categories. In this upcoming environment, uncoordinated use of sensor devices would cause severe imbalance in power consumption of devices, and thus result in early shutdown of some sensing applications depending on power-hungry devices. In this paper, we propose EnergyCordy, a novel inter-device energy use coordination system; with a system-wide holistic view, it coordinates the energy use of concurrent sensing applications over multiple sensor devices. As its key approach, we propose a relaxed sensor association; it decouples the energy use of an application from specific sensor devices leveraging multiple context inference alternatives, allowing flexible energy coordination at runtime. We demonstrated the effectiveness of EnergyCordy by developing multiple example applications over custom-designed wearable senor devices. We show that EnergyCordy effectively coordinates the power usage of concurrent sensing applications over multiple devices and prevent undesired early shutdown of applications.

• Journal of international Conference on Electrical Machines and Systems
• /
• v.3 no.4
• /
• pp.383-388
• /
• 2014

In This paper, the output characteristics and vibrations were compared and analyzed according to operation method in 100W class. The voltage source is applied only two phase in BLDC drive system therefore commutation torque ripple and imbalance of RMF occurred. Due to this efficiency was significantly degraded because mechanical loss is increased, besides the vibration and noise were greatly generated. The vibration and output characteristics were compared and analyzed according to three phase and BLDC drive system.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.434-439
• /
• 2019

딥러닝 기반 분류 모델에 있어 데이터의 클래스 불균형 문제는 소수 클래스의 분류 성능을 크게 저하시킨다. 본 논문에서는 앞서 언급한 클래스 불균형 문제를 보완하기 위한 방안으로 적대적 학습 기법을 제안한다. 적대적 학습 기법의 성능 향상 여부를 확인하기 위해 총 4종의 딥러닝 기반 분류 모델을 정의하였으며, 해당 모델 간 분류 성능을 비교하였다. 실험 결과, 대화 데이터셋을 이용한 모델 학습 시 적대적 학습 기법을 적용할 경우 다수 클래스의 분류 성능은 유지하면서 동시에 소수 클래스의 분류 성능을 크게 향상시킬 수 있음을 확인하였다.

• Journal of the Korea Convergence Society
• /
• v.10 no.5
• /
• pp.35-42
• /
• 2019

Recently, as the Internet and various wearable devices have appeared, Internet technology has contributed to obtaining more convenient information and doing business. However, as the internet is used in various parts, the attack surface points that are exposed to attacks are increasing, Attempts to invade networks aimed at taking unfair advantage, such as cyber terrorism, are also increasing. In this paper, we propose a feature selection method to improve the classification performance of the class to classify the abnormal behavior in the network traffic. The UNSW-NB15 dataset has a rare class imbalance problem with relatively few instances compared to other classes, and an undersampling method is used to eliminate it. We use the SVM, k-NN, and decision tree algorithms and extract a subset of combinations with superior detection accuracy and RMSE through training and verification. The subset has recall values of more than 98% through the wrapper based experiments and the DT_PSO showed the best performance.