• Journal of Broadcast Engineering
• /
• v.22 no.2
• /
• pp.162-172
• /
• 2017

In this paper, we propose facial expression recognition using CNN (Convolutional Neural Network), one of the deep learning technologies. To overcome the disadvantages of existing facial expression databases, various databases are used. In the proposed technique, we construct six facial expression data sets such as 'expressionless', 'happiness', 'sadness', 'angry', 'surprise', and 'disgust'. Pre-processing and data augmentation techniques are also applied to improve efficient learning and classification performance. In the existing CNN structure, the optimal CNN structure that best expresses the features of six facial expressions is found by adjusting the number of feature maps of the convolutional layer and the number of fully-connected layer nodes. Experimental results show that the proposed scheme achieves the highest classification performance of 96.88% while it takes the least time to pass through the CNN structure compared to other models.

• Journal of Practical Engineering Education
• /
• v.14 no.3
• /
• pp.491-498
• /
• 2022

In this paper, as a liberal arts course for non-major students, a case study of recurrent neural network SW practice, which is essential for designing a basic recurrent neural network subject curriculum, was developed. The developed SW practice case focused on understanding the operation principle of the recurrent neural network, and used a spreadsheet to check the entire visualized operation process. The developed recurrent neural network practice case consisted of creating supervised text completion training data, implementing the input layer, hidden layer, state layer (context node), and output layer in sequence, and testing the performance of the recurrent neural network on text data. The recurrent neural network practice case developed in this paper automatically completes words with various numbers of characters. Using the proposed recurrent neural network practice case, it is possible to create an artificial intelligence SW practice case that automatically completes by expanding the maximum number of characters constituting Korean or English words in various ways. Therefore, it can be said that the utilization of this case of basic practice of recurrent neural network is high.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.28B no.10
• /
• pp.783-789
• /
• 1991

A learing method for the recognition of printed Chinese character by using the input driven multi-layer perceptron model was proposed and the circuit representing the learning result was designed. In learning the extracted features from Chinese characters are used as inputs and the synapse's weight is integer value. So it is possible to implement the learning result with CMOS circuit.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1993.06a
• /
• pp.977.1-1000
• /
• 1993

This paper presents a fuzzy neural network, called the fuzzy counterpropagation network, that structures its inputs and generates its outputs in a manner based on counterpropagation networks. The fuzzy counterpropagation network is developed by incorporating the concept of fuzzy clustering into the hidden layer responses. Three learning algorithms are introduced for use with the proposed network. Simulations demonstrate that fuzzy counterpropagation networks with the proposed learning algorithms work well on approximating bipolar and continuous functions.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.23 no.3
• /
• pp.226-231
• /
• 2013

In The matching algorithm for automatic packaging of drugs is essential to determine whether the canister can exactly refill the suitable medicine. In this paper, we propose a hierarchical neural network with the upper and lower layers which can perform real-time processing and classification of many types of medicine bottles to prevent accidental medicine disaster. A few number of low-dimensional feature vector are extracted from the label images presenting medicine-bottle information. By using the extracted feature vectors, the lower layer of MLP(Multi-layer Perceptron) neural networks is learned. Then, the output of the learned middle layer of the MLP is used as the input to the upper layer of the MLP learning. The proposed hierarchical neural network shows good classification performance and real- time operation in the test of up to 30 degrees rotated to the left and right images of 100 different medicine bottles.

• Journal of Practical Engineering Education
• /
• v.16 no.1_spc
• /
• pp.59-64
• /
• 2024

As the aging population grows, the incidence of cancer is increasing. Skin cancer appears externally, but people often don't notice it or simply overlook it. As a result, if the early detection period is missed, the survival rate in the case of late stage cancer is only 7.5-11%. However, the disadvantage of diagnosing, serious skin cancer is that it requires a lot of time and money, such as a detailed examination and cell tests, rather than simple visual diagnosis. To overcome these challenges, we propose an Attention-based CNN model skin cancer classification system. If skin cancer can be detected early, it can be treated quickly, and the proposed system can greatly help the work of a specialist. To mitigate the problem of image data imbalance according to skin cancer type, this skin cancer classification model applies the Over Sampling, technique to data with a high distribution ratio, and adds a pre-learning model without an Attention layer. This model is then compared to the model without the Attention layer. We also plan to solve the data imbalance problem by strengthening data augmentation techniques for specific classes.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.10 no.5
• /
• pp.983-989
• /
• 2006

Current emigration and immigration control inspects passports by the naked eye, registers them by manual input, and compares them with items of database. In this paper, we propose the method to recognize information codes of passports. The proposed passport recognition method extracts character-rows of information codes by applying sobel operator, horizontal smearing, and contour tracking algorithm. The extracted letter-row regions is binarized. After a CDM mask is applied to them in order to recover the individual codes, the individual codes are extracted by applying vertical smearing. The recognizing of individual codes is performed by the RBF network whose hidden layer is applied by ART 2 algorithm and whose learning between the hidden layer and the output layer is applied by a generalized delta learning method. After a photo region is extracted from the reference of the starting point of the extracted character-rows of information codes, that region is verified by the information of luminance, edge, and hue. The verified photo region is certified by the classified features by the ART 2 algorithm. The comparing experiment with real passport images confirmed the good performance of the proposed method.

• Steel and Composite Structures
• /
• v.47 no.1
• /
• pp.91-102
• /
• 2023

To study the evaluation standard and control limit of mortar filling layer void length, in this paper, the train sub-model was developed by MATLAB and the track-bridge sub-model considering the mortar filling layer void was established by ANSYS. The two sub-models were assembled into a train-track-bridge coupling dynamic model through the wheel-rail contact relationship, and the validity was corroborated by the coupling dynamic model with the literature model. Considering the randomness of fastening stiffness, mortar elastic modulus, length of mortar filling layer void, and pier settlement, the test points were designed by the Box-Behnken method based on Design-Expert software. The coupled dynamic model was calculated, and the support vector regression (SVR) nonlinear mapping model of the wheel-rail system was established. The learning, prediction, and verification were carried out. Finally, the reliable probability of the amplification coefficient distribution of the response index of the train and structure in different ranges was obtained based on the SVR nonlinear mapping model and Latin hypercube sampling method. The limit of the length of the mortar filling layer void was, thus, obtained. The results show that the SVR nonlinear mapping model developed in this paper has a high fitting accuracy of 0.993, and the computational efficiency is significantly improved by 99.86%. It can be used to calculate the dynamic response of the wheel-rail system. The length of the mortar filling layer void significantly affects the wheel-rail vertical force, wheel weight load reduction ratio, rail vertical displacement, and track plate vertical displacement. The dynamic response of the track structure has a more significant effect on the limit value of the length of the mortar filling layer void than the dynamic response of the vehicle, and the rail vertical displacement is the most obvious. At 250 km/h - 350 km/h train running speed, the limit values of grade I, II, and III of the lengths of the mortar filling layer void are 3.932 m, 4.337 m, and 4.766 m, respectively. The results can provide some reference for the long-term service performance reliability of the ballastless track-bridge system of HRS.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.16 no.5
• /
• pp.325-330
• /
• 2023

This paper proposes to apply machine learning to the process of predicting the slab thickness based on the structural analysis results or experience and subjectivity of engineers in the design of bridge data construction to enable digital-based decision-making. This study aims to build a reliable design environment by utilizing machine learning techniques to provide guide values to engineers in addition to structural analysis for slab thickness selection. Based on girder bridges, which account for the largest proportion of bridge data, a prediction model process for predicting slab thickness among superstructures was defined. Various machine learning models (Linear Regress, Decision Tree, Random Forest, and Muliti-layer Perceptron) were competed for each process to produce the prediction value for each process, and the optimal model was derived. Through this study, the applicability of machine learning techniques was confirmed in areas where slab thickness was predicted only through existing structural analysis, and an accuracy of 95.4% was also obtained. models can be utilized in a more reliable construction environment if the accuracy of the prediction model is improved by expanding the process

• 제어로봇시스템학회:학술대회논문집
• /
• 2003.10a
• /
• pp.142-145
• /
• 2003

Many control applications using Neural Network need a priori information about the objective system. But it is impossible to get exact information about the objective system in real world. To solve this problem, several control methods were proposed. Reinforcement learning control using neural network is one of them. Basically reinforcement learning control doesn't need a priori information of objective system. This method uses reinforcement signal from interaction of objective system and environment and observable states of objective system as input data. But many methods take too much time to apply to real-world. So we focus on faster learning to apply reinforcement learning control to real-world. Two data types are used for reinforcement learning. One is reinforcement signal data. It has only two fixed scalar values that are assigned for each success and fail state. The other is observable state data. There are infinitive states in real-world system. So the number of observable state data is also infinitive. This requires too much learning time for applying to real-world. So we try to reduce the number of observable states by classification of states with Self-Organizing Map. We also use neural dynamic programming for controller design. An inverted pendulum on the cart system is simulated. Failure signal is used for reinforcement signal. The failure signal occurs when the pendulum angle or cart position deviate from the defined control range. The control objective is to maintain the balanced pole and centered cart. And four states that is, position and velocity of cart, angle and angular velocity of pole are used for state signal. Learning controller is composed of serial connection of Self-Organizing Map and two Multi-layer Feed-Forward Neural Networks.