• Journal of Korea Multimedia Society
• /
• v.20 no.11
• /
• pp.1811-1819
• /
• 2017

As the induction motor is the core production equipment of the industry, it is necessary to construct a fault prediction and diagnosis system through continuous monitoring. Many researches have been conducted on motor fault diagnosis algorithm based on signal processing techniques using Fourier transform, neural networks, and fuzzy inference techniques. In this paper, we propose a fault diagnosis method of induction motor using LPC and DNN. To evaluate the performance of the proposed method, the fault diagnosis was carried out using the vibration data of the induction motor in steady state and simulated various fault conditions. Experimental results show that the learning time of our proposed method and the conventional spectrum+DNN method is 139 seconds and 974 seconds each executed on the experimental PC, and our method reduces execution time by 1/8 compared with conventional method. And the success rate of the proposed method is 98.08%, which is similar to 99.54% of the conventional method.

• International Journal of Internet, Broadcasting and Communication
• /
• v.11 no.4
• /
• pp.76-85
• /
• 2019

We introduce a vision-based object detection method for real-time video surveillance system in low-end edge computing environments. Recently, the accuracy of object detection has been improved due to the performance of approaches based on deep learning algorithm such as Region Convolutional Neural Network(R-CNN) which has two stage for inferencing. On the other hand, one stage detection algorithms such as single-shot detection (SSD) and you only look once (YOLO) have been developed at the expense of some accuracy and can be used for real-time systems. However, high-performance hardware such as General-Purpose computing on Graphics Processing Unit(GPGPU) is required to still achieve excellent object detection performance and speed. To address hardware requirement that is burdensome to low-end edge computing environments, We propose sub-frame analysis method for the object detection. In specific, We divide a whole image frame into smaller ones then inference them on Convolutional Neural Network (CNN) based image detection network, which is much faster than conventional network designed forfull frame image. We reduced its computationalrequirementsignificantly without losing throughput and object detection accuracy with the proposed method.

• Journal of Multimedia Information System
• /
• v.8 no.3
• /
• pp.147-158
• /
• 2021

Versatile Video Coding (VVC) is the latest video coding standard developed by Joint Video Exploration Team (JVET). In VVC, the quadtree plus multi-type tree (QT+MTT) structure of coding unit (CU) partition is adopted, and its computational complexity is considerably high due to the brute-force search for recursive rate-distortion (RD) optimization. In this paper, we aim to reduce the time complexity of inter-picture prediction mode since the inter prediction accounts for a large portion of the total encoding time. The problem can be defined as classifying the split mode of each CU. To classify the split mode effectively, a novel convolutional neural network (CNN) called multi-level tree (MLT-CNN) architecture is introduced. For boosting classification performance, we utilize additional information including inter-picture information while training the CNN. The overall algorithm including the MLT-CNN inference process is implemented on VVC Test Model (VTM) 11.0. The CUs of size 128×128 can be the inputs of the CNN. The sequences are encoded at the random access (RA) configuration with five QP values {22, 27, 32, 37, 42}. The experimental results show that the proposed algorithm can reduce the computational complexity by 11.53% on average, and 26.14% for the maximum with an average 1.01% of the increase in Bjøntegaard delta bit rate (BDBR). Especially, the proposed method shows higher performance on the sequences of the A and B classes, reducing 9.81%~26.14% of encoding time with 0.95%~3.28% of the BDBR increase.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.14 no.4
• /
• pp.775-780
• /
• 2019

Techniques for predicting the future can be categorized into statistics-based and deep-run-based techniques. Among them, statistic-based techniques are widely used because simple and highly accurate. However, working-level officials have difficulty using many analytical techniques correctly. In this study, we compared the accuracy of prediction by applying multinomial logistic regression, decision tree, random forest, support vector machine, and Bayesian inference to marketing related data. The same marketing data was used, and analysis was conducted by using R. The prediction results of various techniques reflecting the data characteristics of the marketing field will be a good reference for practitioners.

• ETRI Journal
• /
• v.44 no.6
• /
• pp.1034-1046
• /
• 2022

The learning-based multiview stereo (MVS) methods for three-dimensional (3D) reconstruction generally use 3D volumes for depth inference. The quality of the reconstructed depth maps and the corresponding point clouds is directly influenced by the spatial resolution of the 3D volume. Consequently, these methods produce point clouds with sparse local regions because of the lack of the memory required to encode a high volume of information. Here, we apply the atrous spatial pyramid pooling (ASPP) module in MVS methods to obtain dense feature maps with multiscale, long-range, contextual information using high receptive fields. For a given 3D volume with the same spatial resolution as that in the MVS methods, the dense feature maps from the ASPP module encoded with superior information can produce dense point clouds without a high memory footprint. Furthermore, we propose a 3D loss for training the MVS networks, which improves the predicted depth values by 24.44%. The ASPP module provides state-of-the-art qualitative results by constructing relatively dense point clouds, which improves the DTU MVS dataset benchmarks by 2.25% compared with those achieved in the previous MVS methods.

• Journal of Korea Society of Digital Industry and Information Management
• /
• v.19 no.1
• /
• pp.91-107
• /
• 2023

With the recent expansion of the commerce ecosystem, a large number of user evaluations have been produced. Accordingly, attempts to create business insights using user evaluation data have been actively made. However, since user evaluation can change after the user experiences the product, it is difficult to say that the analysis based only on reviews immediately after purchase fully reflects the user's evaluation of the product. Moreover, studies conducted so far on user evaluation have overlooked the fact that the length of time a user has used a product can affect the user's product evaluation. Therefore, in this study, we build a model that predicts the direction of change in the user's rating after use from the user's rating and reviews immediately after purchase. In particular, the proposed model reflects the product's period of use in predicting the change direction of the star rating. However, since the posterior information on the duration of product use cannot be used as input in the inference process, we propose a structure that utilizes information about the product's period of use using an auxiliary classifier. As a result of an experiment using 599,889 user evaluation data collected from the shopping platform 'N' company, we confirmed that the proposed model performed better than the existing model in terms of accuracy.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.8
• /
• pp.2199-2213
• /
• 2024

To enhance the quality of defect detection for Printed Circuit Board Assembly (PCBA) during electronic product manufacturing, this study primarily focuses on optimizing the YOLOv7-based method for PCBA defect detection. In this method, the Mish, a smoother function, replaces the Leaky ReLU activation function of YOLOv7, effectively expanding the network's information processing capabilities. Concurrently, a Squeeze-and-Excitation attention mechanism (SEAM) has been integrated into the head of the model, significantly augmenting the precision of small target defect detection. Additionally, considering angular loss, compared to the CIoU loss function in YOLOv7, the SIoU loss function in the paper enhances robustness and training speed and optimizes inference accuracy. In terms of data preprocessing, this study has devised a brightness adjustment data enhancement technique based on split-filtering to enrich the dataset while minimizing the impact of noise and lighting on images. The experimental results under identical training conditions demonstrate that our model exhibits a 9.9% increase in mAP value and an FPS increase to 164 compared to the YOLOv7. These indicate that the method proposed has a superior performance in PCBA defect detection and has a specific application value.

• Journal of Intelligence and Information Systems
• /
• v.29 no.1
• /
• pp.249-263
• /
• 2023

Collaborative Filtering, a representative recommendation system methodology, consists of two approaches: neighbor methods and latent factor models. Among these, the latent factor model using matrix factorization decomposes the user-item interaction matrix into two lower-dimensional rectangular matrices, predicting the item's rating through the product of these matrices. Due to the factor vectors inferred from rating patterns capturing user and item characteristics, this method is superior in scalability, accuracy, and flexibility compared to neighbor-based methods. However, it has a fundamental drawback: the need to reflect the diversity of preferences of different individuals for items with no ratings. This limitation leads to repetitive and inaccurate recommendations. The Adaptive Deep Latent Factor Model (ADLFM) was developed to address this issue. This model adaptively learns the preferences for each item by using the item description, which provides a detailed summary and explanation of the item. ADLFM takes in item description as input, calculates latent vectors of the user and item, and presents a method that can reflect personal diversity using an attention score. However, due to the requirement of a dataset that includes item descriptions, the domain that can apply ADLFM is limited, resulting in generalization limitations. This study proposes a Generalized Adaptive Deep Latent Factor Recommendation Model, G-ADLFRM, to improve the limitations of ADLFM. Firstly, we use item ID, commonly used in recommendation systems, as input instead of the item description. Additionally, we apply improved deep learning model structures such as Self-Attention, Multi-head Attention, and Multi-Conv1D. We conducted experiments on various datasets with input and model structure changes. The results showed that when only the input was changed, MAE increased slightly compared to ADLFM due to accompanying information loss, resulting in decreased recommendation performance. However, the average learning speed per epoch significantly improved as the amount of information to be processed decreased. When both the input and the model structure were changed, the best-performing Multi-Conv1d structure showed similar performance to ADLFM, sufficiently counteracting the information loss caused by the input change. We conclude that G-ADLFRM is a new, lightweight, and generalizable model that maintains the performance of the existing ADLFM while enabling fast learning and inference.

• KIPS Transactions on Software and Data Engineering
• /
• v.9 no.5
• /
• pp.169-176
• /
• 2020

Recently, deep neural networks (DNNs) have brought revolutions to many mobile and embedded devices by providing human-level machine intelligence for various applications. However, high inference accuracy of such DNNs comes at high computational costs, and, hence, there have been significant efforts to reduce computational overheads of DNNs either by compressing off-the-shelf models or by designing a new small footprint DNN architecture tailored to resource constrained devices. One notable recent paradigm in designing small footprint DNN models is sharing parameters in several layers. However, in previous approaches, the parameter-sharing techniques have been applied to large deep networks, such as ResNet, that are known to have high redundancy. In this paper, we propose a parameter-sharing method for already parameter-efficient small networks such as ShuffleNetV2. In our approach, small templates are combined with small layer-specific parameters to generate weights. Our experiment results on ImageNet and CIFAR100 datasets show that our approach can reduce the size of parameters by 15%-35% of ShuffleNetV2 while achieving smaller drops in accuracies compared to previous parameter-sharing and pruning approaches. We further show that the proposed approach is efficient in terms of latency and energy consumption on modern embedded devices.

• Journal of Korean Tunnelling and Underground Space Association
• /
• v.20 no.6
• /
• pp.1161-1175
• /
• 2018

An unexpected event could be easily followed by a large secondary accident due to the limitation in sight of drivers in road tunnels. Therefore, a series of automated incident detection systems have been under operation, which, however, appear in very low detection rates due to very low image qualities on CCTVs in tunnels. In order to overcome that limit, deep learning based tunnel incident detection system was developed, which already showed high detection rates in November of 2017. However, since the object detection process could deal with only still images, moving direction and speed of moving vehicles could not be identified. Furthermore it was hard to detect stopping and reverse the status of moving vehicles. Therefore, apart from the object detection, an object tracking method has been introduced and combined with the detection algorithm to track the moving vehicles. Also, stopping-reverse discrimination algorithm was proposed, thereby implementing into the combined incident detection processes. Each performance on detection of stopping, reverse driving and fire incident state were evaluated with showing 100% detection rate. But the detection for 'person' object appears relatively low success rate to 78.5%. Nevertheless, it is believed that the enlarged richness of image big-data could dramatically enhance the detection capacity of the automatic incident detection system.