• Journal of Information Processing Systems
• /
• 제17권2호
• /
• pp.369-384
• /
• 2021

In a personalized product recommendation system, when the amount of log data is large or sparse, the accuracy of model recommendation will be greatly affected. To solve this problem, a personalized product recommendation method using deep factorization machine (DeepFM) to analyze user behavior is proposed. Firstly, the K-means clustering algorithm is used to cluster the original log data from the perspective of similarity to reduce the data dimension. Then, through the DeepFM parameter sharing strategy, the relationship between low- and high-order feature combinations is learned from log data, and the click rate prediction model is constructed. Finally, based on the predicted click-through rate, products are recommended to users in sequence and fed back. The area under the curve (AUC) and Logloss of the proposed method are 0.8834 and 0.0253, respectively, on the Criteo dataset, and 0.7836 and 0.0348 on the KDD2012 Cup dataset, respectively. Compared with other newer recommendation methods, the proposed method can achieve better recommendation effect.

• Journal of Electrical Engineering and Technology
• /
• 제12권2호
• /
• pp.683-688
• /
• 2017

The unscented Kalman filter features a direct transforming process involving unscented transformation for removing the linearization process error that may occur in the extended Kalman filter. This paper proposes a reduced-order unscented Kalman filter for the sensorless control of a permanent magnet synchronous motor. The proposed method can reduce the computational load without degrading the accuracy compared to the conventional Kalman filters. Moreover, the proposed method can directly estimate the electrical rotor position and speed without a back-electromotive force. The proposed Kalman filter for the sensorless control of a permanent magnet synchronous motor is verified through the simulation and experimentation. The performance of the proposed method is evaluated over a wide range of operations, such as forward and reverse rotations in low and high speeds including the detuning parameters.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제5권11호
• /
• pp.2191-2203
• /
• 2011

In this paper, we propose a fast and robust algorithm for fighting behavior detection based on Motion Vectors (MV), in order to solve the problem of low speed and weak robustness in traditional fighting behavior detection. Firstly, we analyze the characteristics of fighting scenes and activities, and then use motion estimation algorithm based on block-matching to calculate MV of motion regions. Secondly, we extract features from magnitudes and directions of MV, and normalize these features by using Joint Gaussian Membership Function, and then fuse these features by using weighted arithmetic average method. Finally, we present the conception of Average Maximum Violence Index (AMVI) to judge the fighting behavior in surveillance scenes. Experiments show that the new algorithm achieves high speed and strong robustness for fighting behavior detection in surveillance scenes.

• 한국CDE학회논문집
• /
• 제10권1호
• /
• pp.27-39
• /
• 2005

Implicit surfaces are geometric shapes which are defined by implicit functions and exist in three-dimensional space. Recently, implicit surfaces have received much attention in solid modeling applications because they are easy to represent the location of points and to use boolean operations. However, it is difficult to chart points on implicit surfaces for rendering. As efficient rendering method of implicit surfaces, the original Adaptively Sampled Distance Fields (ADFs) $method^{[1]}$ is to use sampled distance fields which subdivide the three dimensional space of implicit surfaces into many cells with high sampling rates in regions where the distance field contains fine detail and low sampling rates where the field varies smoothly. In this paper, in order to maintain the sharp features efficiently with small number of cells, an extended ADFs method is proposed, applying the Dual/Primal mesh optimization $method^{[2]}$ to the original ADFs method. The Dual/Primal mesh optimization method maintains sharp features, moving the vertices to tangent plane of implicit surfaces and reconstructing the vertices by applying a curvature-weighted factor. The proposed extended ADFs method is applied to several examples of implicit surfaces to evaluate the efficiency of the rendering performance.

• 대한전자공학회논문지SP
• /
• 제44권4호통권316호
• /
• pp.93-100
• /
• 2007

웨이브렛 변환에 근거하여 디지털영상으로부터 문자를 처리하는 새로운 접근법을 제시한다. 대상은 각필(刻筆)문자 영상이다. 각필문자에는 형성된 결상에 유사성이 존속하며 배경부분과 함께 서로 다른 준위의 다해상도 특성들로 분해된다는 점을 착안하였다. 우선 Daubechies 웨이브렛을 적용하여 영상을 부대역들로 분해한다. 저주파 부대역은 분할처리와 FCM근거 퍼지 군집분리 및 면적기반 영역처리기법을 적용하여 문자특성을 추출한다. 고주파 부대역들에는 이동창을 설정하고, 이동창의 국부 에너지를 추정하여 고주파 특성들을 활성화한다. 이들 특성들은 조합되어 역웨이브렛 과정을 통해 본래 영상 상태로 복원되고 배경부분이 배제된 문자를 추출한다. 실험 결과는 제안된 기법의 효과를 보이고 있다.

• 대한전자공학회논문지SP
• /
• 제46권2호
• /
• pp.99-107
• /
• 2009

본 논문에서는 자동차 번호판 숫자의 특징을 추출하기 위해 강화된 독립성분분석(independent component analysis)의 혼합모델을 제안한다 독립성분분석은 고차 통계적 특성만을 이용하기 때문에 고차 통계적 특성과 숫자 종류별 상관관계에 대한 특성을 고려하지 못한다. 이러한 독립성분분석의 한계를 극복하기 위해, 본 논문에서는 주성분분석(principle component analysis)과 선형판별분석(linear discriminant analysis)을 조합한 혼합 모델 형태의 독립성분분석을 제안한다. 실험 결과, 제안된 혼합 모델은 독립성분분석이나 다른 혼합 모델들보다 특징 추출과 인식에서 우수한 성능을 보임을 확인하였다.

• 한국멀티미디어학회논문지
• /
• 제24권2호
• /
• pp.178-185
• /
• 2021

Three-dimensional high-resolution magnetic resonance imaging (MRI) provides fine-level anatomical information for disease diagnosis. However, there is a limitation in obtaining high resolution due to the long scan time for wide spatial coverage. Therefore, in order to obtain a clear high-resolution(HR) image in a wide spatial coverage, a super-resolution technology that converts a low-resolution(LR) MRI image into a high-resolution is required. In this paper, we propose a super-resolution technique through filter learning based on information on the surrounding gradient information in 3D space from 3D MRI images. In the learning step, the gradient features of each voxel are computed through eigen-decomposition from 3D patch. Based on these features, we get the learned filters that minimize the difference of intensity between pairs of LR and HR images for similar features. In test step, the gradient feature of the patch is obtained for each voxel, and the filter is applied by selecting a filter corresponding to the feature closest to it. As a result of learning 100 T1 brain MRI images of HCP which is publicly opened, we showed that the performance improved by up to about 11% compared to the traditional interpolation method.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제14권11호
• /
• pp.4426-4442
• /
• 2020

Since scene segmentation is becoming a hot topic in the field of autonomous driving and medical image analysis, researchers are actively trying new methods to improve segmentation accuracy. At present, the main issues in image semantic segmentation are intra-class inconsistency and inter-class indistinction. From our analysis, the lack of global information as well as macroscopic discrimination on the object are the two main reasons. In this paper, we propose a Densely connected residual Attention network (DA-Res2Net) which consists of a dense residual network and channel attention guidance module to deal with these problems and improve the accuracy of image segmentation. Specifically, in order to make the extracted features equipped with stronger multi-scale characteristics, a densely connected residual network is proposed as a feature extractor. Furthermore, to improve the representativeness of each channel feature, we design a Channel-Attention-Guide module to make the model focusing on the high-level semantic features and low-level location features simultaneously. Experimental results show that the method achieves significant performance on various datasets. Compared to other state-of-the-art methods, the proposed method reaches the mean IOU accuracy of 83.2% on PASCAL VOC 2012 and 79.7% on Cityscapes dataset, respectively.

• 디지털산업정보학회논문지
• /
• 제18권4호
• /
• pp.47-54
• /
• 2022

In this paper, MEMS embedded system design implemented the sensor events via analyzing the characteristics that dynamically happened to an abnormal status in power IoT environments in order to guarantee a maintainable operation. We used three kinds of tools in this paper, at first Bluetooth Low Energy (BLE) technology which is a suitable protocol that provides a low data rate, low power consumption, and low-cost sensor applications. Secondly LSM6DSOX, a system-in-module containing a 3-axis digital accelerometer and gyroscope with low-power features for optimal motion. Thirdly BM1422AGMV Digital Magnetometer IC, a 3-axis magnetic sensor with an I2C interface and a magnetic measurable range of ±120 uT, which incorporates magneto-impedance elements to detect the magnetic field when the current flowed in the power devices. The proposed MEMS system was developed based on an nRF5340 System on Chip (SoC), previously compared to the standalone embedded system without bluetooth technology via mobile App. And also, MEMS embedded system with BLE 5.0 technology broadcasted the MEMS system status to Android mobile server. The experiment results enhanced the performance of MEMS system design by combination of sensors, BLE technology and mobile application.

• 한국정밀공학회:학술대회논문집
• /
• 한국정밀공학회 2005년도 춘계학술대회 논문집
• /
• pp.1206-1209
• /
• 2005

In recent, injection molding process for features in sub-micron scale is under active development as patterning nano-scale features, which can provide the master or stamp for molding, and becomes available around the world. Injection molding has been one of the most efficient processes for mass production of the plastic product, and this process is already applied to nano-technology products successfully such as optical storage media like DVD or BD which is a large area plastic thin substrate with nano-scale features on its surface. Bio chip for like DNA sequencing may be another application of this plastic substrate. The DNA can be sequenced using order of 100 nm pore structure when making the DNA flow through the pore structure. Agarose gel and silicon based chip have been used to sequence the DNA, but injection molded plastic chip may have benefit in terms of cost. This plastic DNA sequencing chip has plenty of pillars in order of 100 nm in diameter on the substrate. When the usual features in case of DVD or BD have very low aspect ratio, even less than 0.5, but the DNA chip will have relatively high aspect ratio of about 2. It is not easy to injection mold the large area thin substrate with sub-micron features on its surface due to the characteristics of the molding process and it becomes much more difficult when the aspect ratio of the features becomes high. We investigated the effect of the molding parameters for injection molding with high aspect ratio nano-scale features and injection molded some plastic DNA sequencing chips. We also fabricated PR masters and Ni stamps of the DNA chip to be used for molding