• 산업경영시스템학회지
• /
• 제46권1호
• /
• pp.55-67
• /
• 2023

Lightweight face recognition models, as one of the most popular and long-standing topics in the field of computer vision, has achieved vigorous development and has been widely used in many real-world applications due to fewer number of parameters, lower floating-point operations, and smaller model size. However, few surveys reviewed lightweight models and reimplemented these lightweight models by using the same calculating resource and training dataset. In this survey article, we present a comprehensive review about the recent research advances on the end-to-end efficient lightweight face recognition models and reimplement several of the most popular models. To start with, we introduce the overview of face recognition with lightweight models. Then, based on the construction of models, we categorize the lightweight models into: (1) artificially designing lightweight FR models, (2) pruned models to face recognition, (3) efficient automatic neural network architecture design based on neural architecture searching, (4) Knowledge distillation and (5) low-rank decomposition. As an example, we also introduce the SqueezeFaceNet and EfficientFaceNet by pruning SqueezeNet and EfficientNet. Additionally, we reimplement and present a detailed performance comparison of different lightweight models on the nine different test benchmarks. At last, the challenges and future works are provided. There are three main contributions in our survey: firstly, the categorized lightweight models can be conveniently identified so that we can explore new lightweight models for face recognition; secondly, the comprehensive performance comparisons are carried out so that ones can choose models when a state-of-the-art end-to-end face recognition system is deployed on mobile devices; thirdly, the challenges and future trends are stated to inspire our future works.

• Nuclear Engineering and Technology
• /
• 제55권8호
• /
• pp.2905-2918
• /
• 2023

The valve assembly used in nuclear power plants is important safety-related equipment. In the new standard, the physical attributes are measured using a valve diagnosis test, which is used in the expansion to other non-tested valves using a quantitative test-basis methodology. With a motor-operated actuator, the state of stem's lubrication is related to physical attributes such as the stem factor and the friction coefficient. This study analyzed the numerical transient of fluid and solid lubrication with a squeeze film effect due to the loading rate on the stem and the stem nut using the experimental data. The differential equation that governs the motion mechanism of the stem and stem nut is established and analyzed. The flow rate, the fluid and the solid contact forces are calculated with the friction coefficient. Finally, we found that a change in the friction coefficient results from a change of the shear force in the solid contact mode during the interchange process between the solid contact mode and the fluid contact mode. The qualitative understanding of the squeeze film effect is expanded quantitatively for forces, thread surface distance, velocity, and acceleration, with consideration of the metal solid contact and fluid contact.

• 한국주조공학회지
• /
• 제28권1호
• /
• pp.41-44
• /
• 2008

레오다이캐스팅 공정은 용탕상태의 원소재로부터 최종형상에 가장 가까운 정형제품(near-net shape)을 원하는 기계적 성질과 동시에 경제적으로 생산할 수 있는 제조방법으로 개발되어 왔다. 본 논문에서는 전자교반 시스템을 이용하여 결정립이 제어된 반용융 상태의 소재를 수직형 다이캐스팅기를 사용하여 리어파트(rear parts) 자동차 부품을 성형실험 하였다. 성형된 시제품들은 성형조건에 따른 조직 및 인장강도를 관찰하였으며, 시효경화 시간에 따른 T6 최적 열처리 조건을 조사하였다.

• 정보처리학회논문지:소프트웨어 및 데이터공학
• /
• 제11권1호
• /
• pp.29-34
• /
• 2022

모바일 디바이스 화면상의 클릭 가능한 객체를 인지하기 위한 데이터셋을 구축하고 새로운 네트워크 구조를 제안한다. 모바일 디바이스 화면에서 클릭 가능한 객체를 기준으로 다양한 해상도를 가진 디바이스에서 여러 애플리케이션을 대상으로 데이터를 수집하였다. 총 24,937개의 annotation data를 text, edit text, image, button, region, status bar, navigation bar의 7개 카테고리로 세분화하였다. 해당 데이터셋을 학습하기 위한 모델 구조는 Deconvolution Single Shot Detector를 베이스라인으로, backbone network는 기존 ResNet에 Squeeze-and-Excitation block을 추가한 Squeeze-and-Excitation networks를 사용하고, Single shot detector layers와 Deconvolution module을 Feature pyramid networks 형태로 쌓아 올려 header와 연결한다. 또한, 기존 input resolution의 1:1 비율에서 오는 특징의 손실을 최소화하기 위해 모바일 디바이스 화면과 유사한 1:2 비율로 변경하였다. 해당 모델을 구축한 데이터셋에 대하여 실험한 결과 베이스라인에 대비하여 mean average precision이 최대 101% 개선되었다.

• 대한건축학회논문집:구조계
• /
• 제36권3호
• /
• pp.113-120
• /
• 2020

The purpose of this study is to compare the models of Deep Learning-based Convolution Neural Network(CNN) for concrete crack detection. The comparison models are AlexNet, GoogLeNet, VGG16, VGG19, ResNet-18, ResNet-50, ResNet-101, and SqueezeNet which won ImageNet Large Scale Visual Recognition Challenge(ILSVRC). To train, validate and test these models, we constructed 3000 training data and 12000 validation data with 256×256 pixel resolution consisting of cracked and non-cracked images, and constructed 5 test data with 4160×3120 pixel resolution consisting of concrete images with crack. In order to increase the efficiency of the training, transfer learning was performed by taking the weight from the pre-trained network supported by MATLAB. From the trained network, the validation data is classified into crack image and non-crack image, yielding True Positive (TP), True Negative (TN), False Positive (FP), False Negative (FN), and 6 performance indicators, False Negative Rate (FNR), False Positive Rate (FPR), Error Rate, Recall, Precision, Accuracy were calculated. The test image was scanned twice with a sliding window of 256×256 pixel resolution to classify the cracks, resulting in a crack map. From the comparison of the performance indicators and the crack map, it was concluded that VGG16 and VGG19 were the most suitable for detecting concrete cracks.

• International Journal of Internet, Broadcasting and Communication
• /
• 제14권1호
• /
• pp.136-141
• /
• 2022

A very complex task in deep learning such as image classification must be solved with the help of neural networks and activation functions. The backpropagation algorithm advances backward from the output layer towards the input layer, the gradients often get smaller and smaller and approach zero which eventually leaves the weights of the initial or lower layers nearly unchanged, as a result, the gradient descent never converges to the optimum. We propose a two-factor non-saturating activation functions known as Bea-Mish for machine learning applications in deep neural networks. Our method uses two factors, beta (𝛽) and alpha (𝛼), to normalize the area below the boundary in the Mish activation function and we regard these elements as Bea. Bea-Mish provide a clear understanding of the behaviors and conditions governing this regularization term can lead to a more principled approach for constructing better performing activation functions. We evaluate Bea-Mish results against Mish and Swish activation functions in various models and data sets. Empirical results show that our approach (Bea-Mish) outperforms native Mish using SqueezeNet backbone with an average precision (AP50val) of 2.51% in CIFAR-10 and top-1accuracy in ResNet-50 on ImageNet-1k. shows an improvement of 1.20%.

• 대한방사선기술학회지:방사선기술과학
• /
• 제43권5호
• /
• pp.397-404
• /
• 2020

Coronavirus disease(COVID-19) is highly infectious disease that directly affects the lungs. To observe the clinical findings from these lungs, the Chest Radiography(CXR) can be used in a fast manner. However, the diagnostic performance via CXR needs to be improved, since the identifying these findings are highly time-consuming and prone to human error. Therefore, Artificial Intelligence(AI) based tool may be useful to aid the diagnosis of COVID-19 via CXR. In this study, we explored various Deep learning(DL) approach to classify COVID-19, other viral pneumonia and normal. For the original dataset and lung-segmented dataset, the pre-trained AlexNet, SqueezeNet, ResNet18, DenseNet201 were transfer-trained and validated for 3 class - COVID-19, viral pneumonia, normal. In the results, AlexNet showed the highest mean accuracy of 99.15±2.69% and fastest training time of 1.61±0.56 min among 4 pre-trained neural networks. In this study, we demonstrated the performance of 4 pre-trained neural networks in COVID-19 diagnosis with CXR images. Further, we plotted the class activation map(CAM) of each network and demonstrated that the lung-segmentation pre-processing improve the performance of COVID-19 classifier with CXR images by excluding background features.

• 소성∙가공
• /
• 제5권3호
• /
• pp.239-255
• /
• 1996

A new forming technology has been developed to fabricate near-net shape products using light metal. A semi-solid forming technology has some advantages compared with the conventional forming processes such as die casting squeeze casting and hot/cold forging. In this study the numerical analysis of semi-solid filling for a straight die shape and orifice die shape in gate pattern is studied on semi-solid materials(SSM) of solid fraction fs =30% in A356 aluminum alloy. The finite difference program of Navier-Stokes equation coupled with heat transfer and solidification has been developed to predict a filling pattern and the temperature distribution of SSM. The programdeveloped in this study gives die filling patterns of SSM and final solidifica-tion region.

• 한국정밀공학회지
• /
• 제20권2호
• /
• pp.58-65
• /
• 2003

The conventional forming process such as squeeze casting or die casting for fabricating metal matrix composites products have a disadvantage such as non homogenous distribution of reinforcement, weak bonding between matrix and reinforcement and cost increase in parts fabrication. Thixoforming process has been accepted as a new method for fabricating the net shaped metal matrix composites with lightweight and wear resistance. In this paper, the effect of volume fraction and reinforcement sizes on mechanical properties in cylinder liner part of metal matrix composites has been investigated with processes parameters such as pressure and velocity. Moreover, the methods to obtain the thixoforged composites cylinder liner with high quality has been proposed. To evaluate the composites cylinder linear fabricated at the conditions proposed in this study, mechanical properties of fabricated composites cylinder linear were compared with those of commercial composites cylinder linear.

• 한국주조공학회지
• /
• 제20권6호
• /
• pp.377-385
• /
• 2000

Al matrix composites as the most promising MMCs can be expected to be excellent engineering materials in the nearest future. So as to improve material properties of composite, many manufacturing processes have been developed. Among them, squeeze casting process which offers fine microstructure and near-net-shape is one of the most successful MMCs manufacturing processes. But, in case of with subsieve size particles (under 44 ${\mu}m$), it is very difficult to homogeneously distribute particles in matrix of Al matrix composite by various casting processes, including squeeze casting used so far. Duplex process which was developed in previous study was used to distribute the particle of subsieve size more homogeneously in matrix of Al matrix composite. Microstructures, wear and heat resistance characterization of Al-Si-Cu-Mg-(Ni)/SiCp manufactured by duplex process were examined to clarify the effect of manufacturing conditions, particle size of reinforcement and alloying elements. Al matrix composites reinforced with SiCp(10 ${\mu}m$) have the lowest wear amount among composites reinforced with 3 ${\mu}m$, 5 ${\mu}m$ and 10 ${\mu}m$ SiCp. The wear amount of Al matrix composites with 10 wt.% SiCp(3, 5, 10 ${\mu}m$) was decreased according to the increase of the sliding speed because abrasive wear takes place at high sliding speed of 4m/s and worn debris with block type occurs at low sliding speed of 1m/s. As for heat resistance, it is made clear that remarkable heat resistance property can be obtained by addition of Ni element in Al matrix composites.