• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.46 no.1
• /
• 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.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.06a
• /
• pp.205-208
• /
• 2022

Multi-exposure high dynamic range (HDR) image reconstruction, the task of reconstructing an HDR image from multiple low dynamic range (LDR) images in a dynamic scene, often produces ghosting artifacts caused by camera motion and moving objects and also cannot deal with washed-out regions due to over or under-exposures. While there has been many deep-learning-based methods with motion estimation to alleviate these problems, they still have limitations for severely moving scenes. They also require large parameter counts, especially in the case of state-of-the-art methods that employ attention modules. To address these issues, we propose a frequency domain approach based on the idea that the transform domain coefficients inherently involve the global information from whole image pixels to cope with large motions. Specifically we adopt Residual Fast Fourier Transform (RFFT) blocks, which allows for global interactions of pixels. Moreover, we also employ Depthwise Overparametrized convolution (DO-conv) blocks, a convolution in which each input channel is convolved with its own 2D kernel, for faster convergence and performance gains. We call this LFFNet (Lightweight Frequency Fusion Network), and experiments on the benchmarks show reduced ghosting artifacts and improved performance up to 0.6dB tonemapped PSNR compared to recent state-of-the-art methods. Our architecture also requires fewer parameters and converges faster in training.

• Proceedings of the Korean Society of Crop Science Conference
• /
• 2022.10a
• /
• pp.88-88
• /
• 2022

Detection of stress responses in crops is important to diagnose crop growth and evaluate yield. Also, the multi-spectral sensor is effectively known to evaluate stress caused by nutrient and moisture in crops or biological agents such as weeds or diseases. Therefore, in this experiment, multispectral images were taken by an unmanned aerial vehicle(UAV) under field condition. The experiment was conducted in the long-term fertilizer field in the National Institute of Crop Science, and experiment area was divided into different status of NPK(Control, N-deficiency, P-deficiency, K-deficiency, Non-fertilizer). Total 11 vegetation indices were created with RGB and NIR reflectance values using python. Variations in nutrient content in plants affect the amount of light reflected or absorbed for each wavelength band. Therefore, the objective of this experiment was to evaluate vegetation indices derived from multispectral reflectance data as input into machine learning algorithm for the classification of nutritional deficiency in rice. RandomForest model was used as a representative ensemble model, and parameters were adjusted through hyperparameter tuning such as RandomSearchCV. As a result, training accuracy was 0.95 and test accuracy was 0.80, and IPCA, NDRE, and EVI were included in the top three indices for feature importance. Also, precision, recall, and f1-score, which are indicators for evaluating the performance of the classification model, showed a distribution of 0.7-0.9 for each class.

• Advances in nano research
• /
• v.15 no.6
• /
• pp.567-577
• /
• 2023

The killing of bacteria by mechanical forces on nanopatterned surfaces has been defined as a mechano-bactericidal effect. Inspired by nature, this method is a new-generation technology that does not cause toxic effects and antibiotic resistance. This study aimed to simulate the mechano-bactericidal effect of nanopatterned surfaces' geometric parameters and material properties against three implant-derived bacterial species. Here, in silico models were developed to explain the interactions between the bacterial cell and the nanopatterned surface. Numerical solutions were performed based on the finite element method. Elastic and creep deformation models of bacterial cells were created. Maximum deformation, maximum stress, maximum strain, as well as mortality of the cells were calculated. The results showed that increasing the peak sharpness and decreasing the width of the nanopatterns increased the maximum deformation, stress, and strain in the walls of the three bacterial cells. The increase in spacing between nanopatterns increased the maximum deformation, stress, and strain in E. coli and P. aeruginosa cell walls it decreased in S. aureus. The decrease in width with the increase in sharpness and spacing increased the mortality of E. coli and P. aeruginosa cells, the same values did not cause mortality in S. aureus cells. In addition, it was determined that using different materials for nanopatterns did not cause a significant change in stress, strain, and deformation. This study will accelerate and promote the production of more efficient mechano-bactericidal implant surfaces by modeling the geometric structures and material properties of nanopatterned surfaces together.

• The Journal of the Acoustical Society of Korea
• /
• v.25 no.7
• /
• pp.312-318
• /
• 2006

The conventional lip-synch system has a two-step process, speech segmentation and recognition. However, the difficulty of speech segmentation procedure and the inaccuracy of training data set due to the segmentation lead to a significant Performance degradation in the system. To cope with that, the connected vowel recognition method using Head-Body-Tail (HBT) model is proposed. The HBT model which is appropriate for handling relatively small sized vocabulary tasks reflects co-articulation effect efficiently. Moreover the 7 vowels are merged into 3 classes having similar lip shape while the system is optimized by employing a class dependent SCHMM structure. Additionally in both end sides of each word which has large variations, 8 components Gaussian mixture model is directly used to improve the ability of representation. Though the proposed method reveals similar performance with respect to the CHMM based on the HBT structure. the number of parameters is reduced by 33.92%. This reduction makes it a computationally efficient method enabling real time operation.

• Anatomy and Cell Biology
• /
• v.56 no.1
• /
• pp.86-93
• /
• 2023

Age at death estimation has always been a crucial yet challenging part of identification process in forensic field. The use of human skeletons have long been explored using the principle of macro and micro-architecture change in correlation with increasing age. The clavicle is recommended as the best candidate for accurate age estimation because of its accessibility, time to maturation and minimal effect from weight. Our study applies pre-trained convolutional neural network in order to achieve the most accurate and cost effective age estimation model using clavicular bone. The total of 988 clavicles of Thai population with known age and sex were radiographed using Kodak 9000 Extra-oral Imaging System. The radiographs then went through preprocessing protocol which include region of interest selection and quality assessment. Additional samples were generated using generative adversarial network. The total clavicular images used in this study were 3,999 which were then separated into training and test set, and the test set were subsequently categorized into 7 age groups. GoogLeNet was modified at two layers and fine tuned the parameters. The highest validation accuracy was 89.02% but the test set achieved only 30% accuracy. Our results show that the use of medial clavicular radiographs has a potential in the field of age at death estimation, thus, further study is recommended.

• Current Optics and Photonics
• /
• v.8 no.3
• /
• pp.230-238
• /
• 2024

The refractive index is a key material-design parameter, especially for high-refractive-index glasses, which are used for precision optics and devices. Increased demand for high-precision optical lenses produced by the glass-mold-press (GMP) process has spurred extensive studies of proper glass materials. B₂O₃, SiO₂, and multiple heavy-metal oxides such as Ta₂O₅, Nb₂O₅, La₂O₃, and Gd₂O₃ mostly compose the high-refractive-index glasses for GMP. However, due to many oxides including up to 10 components, it is hard to predict the refractivity solely from the composition of the glass. In this study, the refractive index of optical glasses based on the B₂O₃-La₂O₃-Ta₂O₅-SiO₂ system is predicted using machine learning (ML) and compared to experimental data. A dataset comprising up to 271 glasses with 10 components is collected and used for training. Various ML algorithms (linear-regression, Bayesian-ridge-regression, nearest-neighbor, and random-forest models) are employed to train the data. Along with composition, the polarizability and density of the glasses are also considered independent parameters to predict the refractive index. After obtaining the best-fitting model by R² value, the trained model is examined alongside the experimentally obtained refractive indices of B₂O₃-La₂O₃-Ta₂O₅-SiO₂ quaternary glasses.

• Journal of Korean Association for Spatial Structures
• /
• v.24 no.3
• /
• pp.43-51
• /
• 2024

Dynamic responses of nuclear power plant structure subjected to earthquake loads should be carefully investigated for safety. Because nuclear power plant structure are usually constructed by material of reinforced concrete, the aging deterioration of R.C. have no small effect on structural behavior of nuclear power plant structure. Therefore, aging deterioration of R.C. nuclear power plant structure should be considered for exact prediction of seismic responses of the structure. In this study, a machine learning model for seismic response prediction of nuclear power plant structure was developed by considering aging deterioration. The OPR-1000 was selected as an example structure for numerical simulation. The OPR-1000 was originally designated as the Korean Standard Nuclear Power Plant (KSNP), and was re-designated as the OPR-1000 in 2005 for foreign sales. 500 artificial ground motions were generated based on site characteristics of Korea. Elastic modulus, damping ratio, poisson's ratio and density were selected to consider material property variation due to aging deterioration. Six machine learning algorithms such as, Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Artificial Neural Networks (ANN), eXtreme Gradient Boosting (XGBoost), were used t o construct seispic response prediction model. 13 intensity measures and 4 material properties were used input parameters of the training database. Performance evaluation was performed using metrics like root mean square error, mean square error, mean absolute error, and coefficient of determination. The optimization of hyperparameters was achieved through k-fold cross-validation and grid search techniques. The analysis results show that neural networks present good prediction performance considering aging deterioration.

• Computers and Concrete
• /
• v.34 no.2
• /
• pp.247-265
• /
• 2024

Measuring the fracture toughness of concrete in laboratory settings is challenging due to various factors, such as complex sample preparation procedures, the requirement for precise instruments, potential sample failure, and the brittleness of the samples. Therefore, there is an urgent need to develop innovative and more effective tools to overcome these limitations. Supervised learning methods offer promising solutions. This study introduces seven machine learning algorithms for predicting concrete's effective fracture toughness (K-eff). The models were trained using 560 datasets obtained from the central straight notched Brazilian disc (CSNBD) test. The concrete samples used in the experiments contained micro silica and powdered stone, which are commonly used additives in the construction industry. The study considered six input parameters that affect concrete's K-eff, including concrete type, sample diameter, sample thickness, crack length, force, and angle of initial crack. All the algorithms demonstrated high accuracy on both the training and testing datasets, with R² values ranging from 0.9456 to 0.9999 and root mean squared error (RMSE) values ranging from 0.000004 to 0.009287. After evaluating their performance, the gated recurrent unit (GRU) algorithm showed the highest predictive accuracy. The ranking of the applied models, from highest to lowest performance in predicting the K-eff of concrete, was as follows: GRU, LSTM, RNN, SFL, ELM, LSSVM, and GEP. In conclusion, it is recommended to use supervised learning models, specifically GRU, for precise estimation of concrete's K-eff. This approach allows engineers to save significant time and costs associated with the CSNBD test. This research contributes to the field by introducing a reliable tool for accurately predicting the K-eff of concrete, enabling efficient decision-making in various engineering applications.

• Journal of The Korean Society of Integrative Medicine
• /
• v.12 no.3
• /
• pp.83-91
• /
• 2024

Purpose : This case study aimed to investigate the effects of backward walking exercises with a front-loaded bag on craniovertebral angle (CVA), craniorotational angle (CRA), and gait variables in subjects with forward head posture (FHP). Methods : Two individuals in their twenties with FHP performed backward walking exercises on a treadmill while carrying a front-loaded bag with a load equivalent to 20 % of their body weight, for 30 minutes per day, three times a week, over two weeks. CVA and CRA were measured before and after the intervention using side view photographs taken from 1.5 meters away. CVA was calculated by marking C7, the tragus of the ear, and the outer canthus of the eye, and CRA was determined using the same landmarks. Image J software was used for angle analysis, with measurements taken three times and averaged. Gait variables such as step length and cadence were recorded using a step analysis treadmill and analyzed with the software included with the equipment, with measurements taken at baseline and after the two-week intervention. Results : Both participants demonstrated notable improvements in the CVA, indicating enhanced head alignment relative to the cervical spine. There was also a marked decrease in the CRA, suggesting a reduction in rotational misalignment. Although differences were observed in gait variables, such as step length and cadence, these changes were not consistent across measurements. The results suggest that backward walking exercises with a load carried in front can positively influence postural adjustments by aligning the cervical spine in individuals with FHP. Conclusion : The findings of this case study indicate that backward walking exercises with a front-loaded bag can effectively improve cervical spine alignment in individuals with FHP. Differences were observed in gait variables, such as step length and cadence, but these changes were not consistent across measurements. Future studies should explore these effects more comprehensively and consider optimizing the exercise protocol for better therapeutic outcomes.