• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.10
• /
• pp.3668-3684
• /
• 2021

Video action recognition is widely used in video surveillance, behavior detection, human-computer interaction, medically assisted diagnosis and motion analysis. However, video action recognition can be disturbed by many factors, such as background, illumination and so on. Two-stream convolutional neural network uses the video spatial and temporal models to train separately, and performs fusion at the output end. The multi segment Two-Stream convolutional neural network model trains temporal and spatial information from the video to extract their feature and fuse them, then determine the category of video action. Google Xception model and the transfer learning is adopted in this paper, and the Xception model which trained on ImageNet is used as the initial weight. It greatly overcomes the problem of model underfitting caused by insufficient video behavior dataset, and it can effectively reduce the influence of various factors in the video. This way also greatly improves the accuracy and reduces the training time. What's more, to make up for the shortage of dataset, the kinetics400 dataset was used for pre-training, which greatly improved the accuracy of the model. In this applied research, through continuous efforts, the expected goal is basically achieved, and according to the study and research, the design of the original dual-flow model is improved.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.11 no.3
• /
• pp.215-222
• /
• 2001

Neural network based models of semiconductor manufacturing processes have been shown to offer advantages in both accuracy and generalization over traditional methods. However, model development is often complicated by the fact that back-propagation neural networks contain several adjustable parameters whose optimal values unknown during training. These include learning rate, momentum, training tolerance, and the number of hidden layer neurOnS. This paper presents an investigation of the use of genetic algorithms (GAs) to determine the optimal neural network parameters for the modeling of plasma-enhanced chemical vapor deposition (PECVD) of silicon dioxide films. To find an optimal parameter set for the neural network PECVD models, a performance index was defined and used in the GA objective function. This index was designed to account for network prediction error as well as training error, with a higher emphasis on reducing prediction error. The results of the genetic search were compared with the results of a similar search using the simplex algorithm.

• Journal of Korean Society of Forest Science
• /
• v.84 no.1
• /
• pp.63-70
• /
• 1995

This paper discusses applications of neural network to stand stocking control problems. The scope of this research was to develop a neural network model for finding optimal stand management regimes and examining the performance of the model for field application. Performance was analyzed in consideration of the number of training examples and structural aspects of neural network. Research on network performance was based on extensive optimization studies for pure longleaf pine(Pinus palustris) stands. For experimental purposes. an existing nonlinear even-aged stand optimization model with a whole-stand growth and yield simulator was used to generate data samples required for the performance analysis.

• Journal of Welding and Joining
• /
• v.31 no.5
• /
• pp.35-40
• /
• 2013

In this paper, we propose the Tangible Virtual Reality Representation Method to using haptic device and feature to morphology of created bead from Flux Cored Arc Welding. The virtual reality was started to rising for reduce to consumable materials and welding training risk. And, we will expected maximize virtual reality from virtual welding training. In this paper proposed method is get the database to changing the input factor such as work angle, travelling angle, speed, CTWD. And, it is visualization to bead from extract to optimal morphological feature information to using the Neural Network algorithm. The database was building without error to extract data from automatic robot welder. Also, the Neural Network algorithm was set a dataset of the highest accuracy from verification process in many times. The bead was created in virtual reality from extract to morphological feature information. We were implementation to final shape of bead and overlapped in process by time to using bead generation algorithm and calibration algorithm for generate to same bead shape to real database in process of generating bead. The best advantage of virtual welding training, it can be get the many data to training evaluation. In this paper, we were representation bead to similar shape from generated bead to Flux Cored Arc Welding. Therefore, we were reduce the gap to virtual welding training and real welding training. In addition, we were confirmed be able to maximize the performance of education from more effective evaluation system.

• Proceedings of the KSME Conference
• /
• 2004.04a
• /
• pp.1183-1188
• /
• 2004

The inverse heat conduction problem (IHCP) is a problem of estimating boundary condition from temperature measurement at one or more interior points. Neural networks are general information processing systems inspired by the connectionist theory of human brain. By properly training the network by the learning rule, the neural network method can handle many non-linear or other complex problems. In this work, neural network is applied to complicated inverse heat conduction problems. Efficiency of the procedure is enhanced by incorporating the radial basis functions (RBF). The RBF is trained faster than other neural network and can find smooth solution. In order to demonstrate the effectiveness of the current scheme, a typical one-dimensional IHCP is considered. At one surface, the temperature as well as the heat flux is known. The unknown temperature of interest is estimated on the other side of the slab. The results from the proposed method based on RBF neural network are compared with the conventional method.

• Tribology and Lubricants
• /
• v.38 no.4
• /
• pp.152-161
• /
• 2022

Chemical mechanical planarization (CMP) is a technology that planarizes the surfaces of semiconductor devices using chemical reaction and mechanical material removal, and it is an essential process in manufacturing highly integrated semiconductors. In the CMP process, a conditioning process using a diamond conditioner is applied to remove by-products generated during processing and ensure the surface roughness of the CMP pad. In previous studies, prediction of pad wear by CMP conditioning has depended on numerical analysis studies based on mathematical simulation. In this study, using an artificial neural network, the ratio of conditioner coverage to the distance between centers in the conditioning system is input, and the average conditioning density, standard deviation, nonuniformity (NU), and conditioning density distribution are trained as targets. The result of training seems to predict the target data well, although the average conditioning density, standard deviation, and NU in the contact area of wafer and pad and all areas of the pad have some errors. In addition, in the case of NU, the prediction calculated from the training results of the average conditioning density and standard deviation can reduce the error of training compared with the results predicted through training. The results of training on the conditioning density profile generally follow the target data well, confirming that the shape of the conditioning density profile can be predicted.

• The Korean Journal of Applied Statistics
• /
• v.26 no.5
• /
• pp.847-856
• /
• 2013

We propose a modified learning process for generalized neural network using a learning algorithm by Liu et al. (2001). We consider the effect of initial weights, training results and learning errors using a modified learning process. We employ an incremental training procedure where training patterns are learned systematically. Our algorithm starts with a single training pattern and a single hidden layer neuron. During the course of neural network training, we try to escape from the local minimum by using a weight scaling technique. We allow the network to grow by adding a hidden layer neuron only after several consecutive failed attempts to escape from a local minimum. Our optimization procedure tends to make the network reach the error tolerance with no or little training after the addition of a hidden layer neuron. Simulation results with suitable initial weights indicate that the present constructive algorithm can obtain neural networks very close to minimal structures and that convergence to a solution in neural network training can be guaranteed. We tested these algorithms extensively with small training sets.

• Journal of Institute of Control, Robotics and Systems
• /
• v.7 no.11
• /
• pp.918-926
• /
• 2001

In this paper, intelligent control method using neural network as a nonlinear controller is presented. Simulation studies for three link rotary robot are performed. Neural network controller is implemented on DSP board in PC to make real time computing possible. On-line training algorithms for neural network control are proposed. As a test-bed, a large x-y table was build and interface with PC has been implemented. Experiments such as inverted pendulum control and large x-y table position control are performed. The results for different PD controller gains with neural network show excellent position tracking for circular trajectory compared with those for PD controller only. Neural control scheme also works better for controlling inverted pendulum on x-y table.

• 제어로봇시스템학회:학술대회논문집
• /
• 2005.06a
• /
• pp.275-280
• /
• 2005

The principal context of this research is the approach to an artificial neural network algorithm which solves multivariable nonlinear equation systems by estimating the state of line power flow. First a dynamical neural network with feedback is used to find the minimum value of the objective function at each iteration of the state estimator algorithm. In second step a two-layer neural network structures is derived to implement all of the different matrix-vector products that arise in neural network state estimator analysis. For hardware requirements, as they relate to the total number of internal connections, the architecture developed here preserves in its structure the pronounced sparsity of power networks for which state the estimator analysis is to be carried out. A principal feature of the architecture is that the computing time overheads in solution are independent of the dimensions or structure of the equation system. It is here where the ultrahigh-speed of massively parallel computing in neural networks can offer major practical benefit.

• The Journal of Korea Robotics Society
• /
• v.17 no.2
• /
• pp.133-141
• /
• 2022

This paper introduces model compression algorithms which make a deep neural network smaller and faster for embedded systems. The model compression algorithms can be largely categorized into pruning, quantization and knowledge distillation. In this study, gradual pruning, quantization aware training, and knowledge distillation which learns the activation boundary in the hidden layer of the teacher neural network are integrated. As a large deep neural network is compressed and accelerated by these algorithms, embedded computing boards can run the deep neural network much faster with less memory usage while preserving the reasonable accuracy. To evaluate the performance of the compressed neural networks, we evaluate the size, latency and accuracy of the deep neural network, DenseNet201, for image classification with CIFAR-10 dataset on the NVIDIA Jetson Xavier.