• Journal of Digital Convergence
• /
• v.19 no.5
• /
• pp.307-313
• /
• 2021

In this paper, a lightweight network with fewer parameters compared to the existing object detection method is proposed. In the case of the currently used detection model, the network complexity has been greatly increased to improve accuracy. Therefore, the proposed network uses EfficientNet as a feature extraction network, and the subsequent layers are formed in a pyramid structure to utilize low-level detailed features and high-level semantic features. An attention process was applied between pyramid structures to suppress unnecessary noise for prediction. All computational processes of the network are replaced by depth-wise and point-wise convolutions to minimize the amount of computation. The proposed network was trained and evaluated using the PASCAL VOC dataset. The features fused through the experiment showed robust properties for various objects through a refinement process. Compared with the CNN-based detection model, detection accuracy is improved with a small amount of computation. It is considered necessary to adjust the anchor ratio according to the size of the object as a future study.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2016.05a
• /
• pp.727-730
• /
• 2016

TSC means classifying time series data based on pattern. Time series data is quite common data type and it has high potential in many fields, so data mining and machine learning have paid attention for long time. In traditional approach, distance and dictionary based methods are quite popular. but due to time scale and random noise problems, it has clear limitation. In this paper, we propose a novel approach to deal with these problems with CNN and data mutation. CNN is regarded as proven neural network model in image recognition, and could be applied to time series pattern recognition by extracting pattern. Data mutation is a way to generate mutated data with different methods to make CNN more robust and solid. The proposed method shows better performance than traditional approach.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• v.9 no.2
• /
• pp.469-472
• /
• 2005

This paper propose about effective speech feature parameter that have robust character in effect of noise in realizing speech recognition system. Established MFCC that is the basic parameter used to ASR(Automatic Speech Recognition) and DCTCs that use DCT in basic parameter. Also, proposed delta-Cepstrum and delta-delta-Cepstrum parameter that reconstruct Cepstrum to have information for variation of speech. And compared recognition performance in using HMM. For dimension reduction of each parameter LDA algorithm apply and compared recognition. Results are presented reduced dimension delta-delta-Cepstrum parameter in using LDA recognition performance that improve more than existent parameter in noise environment of various condition.

• Atmosphere
• /
• v.34 no.1
• /
• pp.55-67
• /
• 2024

This paper aims to provide a detailed introduction to the concept of the Ratio of Predictable Component (RPC) and the Signal-to-Noise Paradox. Then, we derive insights from them by exploring the paradoxical features by conducting a seasonal and regional analysis through ensemble expansion in KMA's climate prediction system (GloSea). We also provide an explanation of the ensemble generation method, with a specific focus on stochastic physics. Through this study, we can provide the predictability limits of our forecasting system, and find way to enhance it. On a global scale, RPC reaches a value of 1 when the ensemble is expanded to a maximum of 56 members, underlining the significance of ensemble expansion in the climate prediction system. The feature indicating RPC paradoxically exceeding 1 becomes particularly evident in the winter North Atlantic and the summer North Pacific. In the Siberian Continent, predictability is notably low, persisting even as the ensemble size increases. This region, characterized by a low RPC, is considered challenging for making reliable predictions, highlighting the need for further improvement in the model and initialization processes related to land processes. In contrast, the tropical ocean demonstrates robust predictability while maintaining an RPC of 1. Through this study, we have brought to attention the limitations of potential predictability within the climate prediction system, emphasizing the necessity of leveraging predictable signals with high RPC values. We also underscore the importance of continuous efforts aimed at improving models and initializations to overcome these limitations.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.43 no.4 s.310
• /
• pp.74-86
• /
• 2006

This paper proposes a robust and computationally efficient algorithm for automatic video object segmentation. For implementing the spatio-temporal segmentation, which aims for efficient combination of the motion segmentation and the color segmentation, an SOM-based hierarchical clustering method in which the segmentation process is regarded as clustering of feature vectors is employed. As results, problems of high computational complexity which required for obtaining exact segmentation results in conventional video object segmentation methods, and the performance degradation due to noise are significantly reduced. A measure of motion vector reliability which employs MRF-based MAP estimation scheme has been introduced to minimize the influence from the motion estimation error. In addition, a noise elimination scheme based on the motion reliability histogram and a clustering validity index for automatically identifying the number of objects in the scene have been applied. A cross projection method for effective object tracking and a dynamic memory to maintain temporal coherency have been introduced as well. A set of experiments has been conducted over several video sequences to evaluate the proposed algorithm, and the efficiency in terms of computational complexity, robustness from noise, and higher segmentation accuracy of the proposed algorithm have been proved.

• 제어로봇시스템학회:학술대회논문집
• /
• 2005.06a
• /
• pp.89-92
• /
• 2005

Using natural speech commands for controlling a human-robot is an interesting topic in the field of robotics. In this paper, our main focus is on the verification of a speaker who gives a command to decide whether he/she is an authorized person for commanding. Among possible dynamic features of natural speech, pitch period is one of the most important ones for characterizing speech signals and it differs usually from person to person. However, current techniques of pitch detection are still not to a desired level of accuracy and robustness. When the signal is noisy or there are multiple pitch streams, the performance of most techniques degrades. In this paper, we propose a two-level approach for pitch detection which in compare with standard pitch detection algorithms, not only increases accuracy, but also makes the performance more robust to noise. In the first level of the proposed approach we discriminate voiced from unvoiced signals based on a neural classifier that utilizes cepstrum sequences of speech as an input feature set. Voiced signals are then further processed in the second level using a modified standard AMDF-based pitch detection algorithm to determine their pitch periods precisely. The experimental results show that the accuracy of the proposed system is better than those of conventional pitch detection algorithms for speech signals in clean and noisy environments.

• The Transactions of the Korea Information Processing Society
• /
• v.6 no.8
• /
• pp.2262-2270
• /
• 1999

In this paper, we describe gesture recognition algorithm using computer vision sensor and HMM. The automatic hand region extraction has been proposed for initializing the tracking of the tele-operation gestures. For this, distance informations(disparity map) as results of stereo matching of initial left and right images are employed to isolate the hand region from a scene. PDOE(positive difference of edges) feature images adapted here have been found to be robust against noise and background brightness. The KNU/KAERI(K/K) gesture instruction set is defined for tele-operation in atomic electric power stations. The composite recognition model constructed by concatenating three gesture instruction models including pre-orders, basic orders, and post-orders has been proposed and identified by discrete HMM. Our experimental results showed that consecutive orders composed of more than two ones are correctly recognized at the rate of above 97%.

• KIPS Transactions on Software and Data Engineering
• /
• v.6 no.12
• /
• pp.573-580
• /
• 2017

Deep Neural Network (DNN) is a large layered neural network which is consisted of a number of layers of non-linear units. Deep Learning which represented as DNN has been applied very successfully in various applications. However, many issues in DNN have been identified through past researches. Among these issues, generalization is the most well-known problem. A Recent study, Dropout, successfully addressed this problem. Also, Dropout plays a role as noise, and so it helps to learn robust feature during learning in DNN such as Denoising AutoEncoder. However, because of a large computations required in Dropout, training takes a lot of time. Since Dropout keeps changing an inter-layer representation during the training session, the learning rates should be small, which makes training time longer. In this paper, using mutation operation, we reduce computation and improve generalization performance compared with Dropout. Also, we experimented proposed method to compare with Dropout method and showed that our method is superior to the Dropout one.

• Journal of the Institute of Convergence Signal Processing
• /
• v.4 no.1
• /
• pp.1-8
• /
• 2003

Various methods have been studying to maintain and apply the multimedia inform abruptly increasing over all social fields, in recent years. For retrieval of still images, we is implemented content-based image retrieval system in this paper that make possible to retrieve similar objects from image database after segmenting query object from background if user request query. Query image is processed median filtering to remove noise first and then object edge is detected it by canny edge detection. And query object is segmented from background by using convex hull. Similarity value can be obtained by means of histogram intersection with database image after securing color histogram from segmented image. Also segmented image is processed gray convert and wavelet transform to extract spacial gray distribution and texture feature. After that, Similarity value can be obtained by means of banded autocorrelogram and energy. Final similar image can be retrieved by adding upper similarity values that it make possible to not only robust in background but also better correct object retrieval by using object segmentation method.

• Smart Structures and Systems
• /
• v.30 no.6
• /
• pp.613-626
• /
• 2022

Guaranteeing the quality and integrity of structural health monitoring (SHM) data is very important for an effective assessment of structural condition. However, sensory system may malfunction due to sensor fault or harsh operational environment, resulting in multiple types of data anomaly existing in the measured data. Efficiently and automatically identifying anomalies from the vast amounts of measured data is significant for assessing the structural conditions and early warning for structural failure in SHM. The major challenges of current automated data anomaly detection methods are the imbalance of dataset categories. In terms of the feature of actual anomalous data, this paper proposes a data anomaly detection method based on data-level and deep learning technique for SHM of civil engineering structures. The proposed method consists of a data balancing phase to prepare a comprehensive training dataset based on data-level technique, and an anomaly detection phase based on a sophisticatedly designed network. The advanced densely connected convolutional network (DenseNet) and Transformer encoder are embedded in the specific network to facilitate extraction of both detail and global features of response data, and to establish the mapping between the highest level of abstractive features and data anomaly class. Numerical studies on a steel frame model are conducted to evaluate the performance and noise immunity of using the proposed network for data anomaly detection. The applicability of the proposed method for data anomaly classification is validated with the measured data of a practical supertall structure. The proposed method presents a remarkable performance on data anomaly detection, which reaches a 95.7% overall accuracy with practical engineering structural monitoring data, which demonstrates the effectiveness of data balancing and the robust classification capability of the proposed network.