• Journal of Information Processing Systems
• /
• v.17 no.5
• /
• pp.892-904
• /
• 2021

An image semantic segmentation model is proposed based on improved ENet network in order to achieve the low accuracy of image semantic segmentation in complex environment. Firstly, this paper performs pruning and convolution optimization operations on the ENet network. That is, the network structure is reasonably adjusted for better results in image segmentation by reducing the convolution operation in the decoder and proposing the bottleneck convolution structure. Squeeze-and-excitation (SE) module is then integrated into the optimized ENet network. Small-scale targets see improvement in segmentation accuracy via automatic learning of the importance of each feature channel. Finally, the experiment was verified on the public dataset. This method outperforms the existing comparison methods in mean pixel accuracy (MPA) and mean intersection over union (MIOU) values. And in a short running time, the accuracy of the segmentation and the efficiency of the operation are guaranteed.

• Journal of Energy Engineering
• /
• v.28 no.3
• /
• pp.18-25
• /
• 2019

In this study, in order to solve the problem of the program of calculating code input by experienced users in the power generation, the wide area energy network research group developed the local heating operation analysis program EntPLAN, which can be easily used by anyone, including scalability, with domestic technology. Therefore, the Commission intended to compare the heat sources, heat demand, and the results of operation of the combined heat plant (CHP) on the energy network through simulation with the EnetPLAN and the program A on the market. The results showed that the heat and power output on the energy network of the EnetPLAN and A programs were mostly similar in pattern in the simulation results of the heat supply and the operation method of the accumulator. This enabled the application of the simulation for the various operation modes of the cogeneration facilities existing on the energy network. It is expected that EntPLAN, which was developed with domestic technology, will be easily applied in the field in the future and will present efficient operation simulation results.

• Journal of the Korean Solar Energy Society
• /
• v.39 no.5
• /
• pp.65-78
• /
• 2019

In this study, we investigated the paraffin encapsulation using double-walled encapsulation technique. The first encapsulation used methyl methacrylic acid as the main component in acrylic polymer and the second encapsulation used melamine polymer. Particles size and distribution of the capsules were analyzed using scanning electron microscopy. In the first encapsulation, the stable capsules were obtained at 67% of phase change material ratio to methyl methacrylic acid monomer and the size of the capsule was from 0.2 to $0.3{\mu}m$. In the second encapsulation, the size of the capsules was almost the same with those capsules prepared in the first encapsulation. The particle size of single wall and double wall was about $0.3{\mu}m$. As a result of the encapsulation of paraffin using double-walled encapsulation technique, it was confirmed that the particle size was determined in the process of encapsulating using the acrylic polymer at the first wall material, and the physical and thermal stability of the capsules were imparted using melamine at the secondary wall material.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.6
• /
• pp.813-826
• /
• 2022

AI-based image processing technologies in various fields have been widely studied. However, the lighter the board, the more difficult it is to reduce the weight of image processing algorithm due to a lot of computation. In this paper, we propose a method using deep learning for object recognition algorithm in lightweight embedded boards. We can determine the area using a deep neural network architecture algorithm that processes semantic segmentation with a relatively small amount of computation. After masking the area, by using more accurate deep learning algorithm we could operate object detection with improved accuracy for efficient neural network (ENet) and You Only Look Once (YOLO) toward executing object recognition in real time for lightweighted embedded boards. This research is expected to be used for autonomous driving applications, which have to be much lighter and cheaper than the existing approaches used for object recognition.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.1175-1176
• /
• 2023

전기차의 자율주행을 위해선 차선 인식과 모터 제어가 필요하다. 카메라로 입력된 영상에 허프 변환을 적용하고, 변환된 이진 이미지에 Enet 및 DeepLabv3+ 구조를 활용한 LaneNet 모델을 적용하여 차선을 학습시키고, Fuzzy 제어 기법을 활용하여 모터의 조향이 원활이 되도록 하였다. 기존의 Rule base 기법에 비하여 차선 인식 정확도가 월등히 향상되었으며, 주행 결과 Real-Time 주행환경 판단에 대한 여지를 남겼다.

• Korean Chemical Engineering Research
• /
• v.48 no.1
• /
• pp.93-97
• /
• 2010

In this study, we have measured the thermo-physical properties of pure organic phase change materials (PCMs) and their mixtures which have the melting points from 0 to $15^{\circ}C$ by using a modified T-history method. These organic PCMs can be used as coolant materials for packaging and shipping of vaccines. Through measuring the thermophysical properties of pure paraffins, we were able to know that we could improve the reliability of measurement if we considered the melting point of each material and subsequently decided an optimum coolant temperature for each system. The modified T-history method showed a potential usefulness for reliably measuring thermo-physical properties of organic mixtures with avoiding possible inaccuracy of measurement due to using a small amount of sample at DSC measurement.

• Korean Chemical Engineering Research
• /
• v.49 no.2
• /
• pp.206-210
• /
• 2011

We fabricated rod-like poly(lactic acid)(PLA) specimens through applying various methods of equal-channel angular extrusion(ECAE) process and investigated the change of thermal and mechanical properties of specimens before and after each ECAE process. Combining three re-injection routes(A, BC, and C) and three pass counts(1, 2 and 4) allowed us to fabricate 7 different PLA specimens. Thermal properties of each specimen were measured by both differential scanning calorimeter and thermo-gravimetric analyzer. Shear strains of each specimen with respect to applied loads were measured by indentation hardness tester. Field emmision scanning electron microscopy was used to observe internal microstructure of cross-section of each specimen. The observed microstructures qualitatively supported the explanation of hardness test results. Among 7 specimens, PLA-P2A showed the biggest shear strain probably due to its dense microstructure.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.2
• /
• pp.33-42
• /
• 2022

In this paper, we proposed a deep learning model based on multi-scale dilated convolution feature fusion for the segmentation of black ice region in road image to send black ice warning to drivers in real time. In the proposed multi-scale dilated convolution feature fusion network, different dilated ratio convolutions are connected in parallel in the encoder blocks, and different dilated ratios are used in different resolution feature maps, and multi-layer feature information are fused together. The multi-scale dilated convolution feature fusion improves the performance by diversifying and expending the receptive field of the network and by preserving detailed space information and enhancing the effectiveness of diated convolutions. The performance of the proposed network model was gradually improved with the increase of the number of dilated convolution branch. The mIoU value of the proposed method is 96.46%, which was higher than the existing networks such as U-Net, FCN, PSPNet, ENet, LinkNet. The parameter was 1,858K, which was 6 times smaller than the existing LinkNet model. From the experimental results of Jetson Nano, the FPS of the proposed method was 3.63, which can realize segmentation of black ice field in real time.

• ETRI Journal
• /
• v.45 no.5
• /
• pp.822-835
• /
• 2023

Recent semantic segmentation frameworks usually combine low-level and high-level context information to achieve improved performance. In addition, postlevel context information is also considered. In this study, we present a Context ReFinement Network (CRFNet) and its training method to improve the semantic predictions of segmentation models of the encoder-decoder structure. Our study is based on postprocessing, which directly considers the relationship between spatially neighboring pixels of a label map, such as Markov and conditional random fields. CRFNet comprises two modules: a refiner and a combiner that, respectively, refine the context information from the output features of the conventional semantic segmentation network model and combine the refined features with the intermediate features from the decoding process of the segmentation model to produce the final output. To train CRFNet to refine the semantic predictions more accurately, we proposed a sequential training scheme. Using various backbone networks (ENet, ERFNet, and HyperSeg), we extensively evaluated our model on three large-scale, real-world datasets to demonstrate the effectiveness of our approach.

• Journal of Korean Society of Environmental Engineers
• /
• v.36 no.3
• /
• pp.172-177
• /
• 2014

Low pressure air flotation (LAF) pilot plant for sludge thickening was installed in Chung Nam N.S. municipal waste water treatment plant to verify its application possibility. Effects of operating conditions such as coagulant dosages and microbubble water ratio on thickening of the mixed sludge were examined. Microbubbles which were generated in the chamber of $1.5kgf/cm^2$ by high speed collision method with foaming agent were used to float sludge. Solid loading of $30kg/m^2/hr$, solid contents in thickened sludge of 60,300 mg/L and SS removal efficiency of 99% were obtained through long period operating LAF in conditions of mixed sludge concentration of 14,400 mg/L, coagulant dosage of 27.6 mg/L, foaming agent addition of 4.0 mg/L and microbubble water injection ratio of 9.7%.