• The Journal of Korean Institute of Next Generation Computing
• /
• v.15 no.5
• /
• pp.64-74
• /
• 2019

Although a non-rigid registration has high demands in clinical practice, it has a high computational complexity and it is very difficult for ensuring the accuracy and robustness of registration. This study proposes a method of applying a non-rigid registration to 3D magnetic resonance images of brain in an unsupervised learning environment by using a deep-learning network. A feature vector between two images is produced through the network by receiving both images from two different patients as inputs and it transforms the target image to match the source image by creating a displacement vector field. The network is designed based on a U-Net shape so that feature vectors that consider all global and local differences between two images can be constructed when performing the registration. As a regularization term is added to a loss function, a transformation result similar to that of a real brain movement can be obtained after the application of trilinear interpolation. This method enables a non-rigid registration with a single-pass deformation by only receiving two arbitrary images as inputs through an unsupervised learning. Therefore, it can perform faster than other non-learning-based registration methods that require iterative optimization processes. Our experiment was performed with 3D magnetic resonance images of 50 human brains, and the measurement result of the dice similarity coefficient confirmed an approximately 16% similarity improvement by using our method after the registration. It also showed a similar performance compared with the non-learning-based method, with about 10,000 times speed increase. The proposed method can be used for non-rigid registration of various kinds of medical image data.

• Journal of Korean Society of Coastal and Ocean Engineers
• /
• v.33 no.3
• /
• pp.93-100
• /
• 2021

Since the IB (Immersed Boundary) method, which can perform coupling analysis with objects and fluids having an impermeable boundary of arbitrary shape on a fixed grid system, has been developed, the IB method in various CFD models is increasing. The representative IB methods are the directing-forcing method and the ghost cell method. The directing-forcing type method numerically satisfies the boundary condition from the fluid force calculated at the boundary surface of the structure, and the ghost-cell type method is a computational method that satisfies the boundary condition through interpolation by placing a virtual cell inside the obstacle. These IB methods have a disadvantage in that the computational algorithm is complex. In this study, the simplified immersed boundary (SIB) method enables the analysis of temporary structures on a fixed grid system and is easy to expand to three proposed dimensions. The SIB method proposed in this study is based on a one-field model for immiscible two-phase fluid that assumes that the density function of each phase moves with the center of local mass. In addition, the volume-weighted average method using the density function of the solid was applied to handle moving solid structures, and the CIP method was applied to the advection calculation to prevent numerical diffusion. To examine the analysis performance of the proposed SIB method, a numerical simulation was performed on an object falling to the free water surface. The numerical analysis result reproduced the object falling to the free water surface well.

• Journal of Korea Water Resources Association
• /
• v.56 no.spc1
• /
• pp.1027-1036
• /
• 2023

Coastal topographic information is crucial in coastal management, but point measurment based approeaches, which are labor intensive, are generally applied to land and underwater, separately. This study introduces an efficient method enabling land and undetwater surveys using an unmanned aerial vehicle (UAV). This method involves applying two different algorithms to measure the topography on land and water depth, respectively, using UAV imagery and merge them to reconstruct whole coastal digital elevation model. Acquisition of the landside terrain is achieved using the Structure-from-Motion Multi-View Stereo technique with spatial scan imagery. Independently, underwater bathymetry is retrieved by employing a depth inversion technique with a drone-acquired wave field video. After merging the two digital elevation models into a local coordinate, interpolation is performed for areas where terrain measurement is not feasible, ultimately obtaining a continuous nearshore terrain. We applied the proposed survey technique to Jangsa Beach, South Korea, and verified that detailed terrain characteristics, such as berm, can be measured. The proposed UAV-based survey method has significant efficiency in terms of time, cost, and safety compared to existing methods.

• Proceedings of the Korean Association of Geographic Inforamtion Studies Conference
• /
• 2003.04a
• /
• pp.363-368
• /
• 2003

We have successfully developed a more effective algorithm to extract the lineament in the area covered by wide alluvial deposits characterized by a relatively narrow range of brightness in the Landsat TM image, while the currently used algorithm is limited to the mountainous areas. In the new algorithm, flat areas mainly consisting of alluvial deposits were selected using the Local Enhancement from the Digital Elevation Model (DEM). The aspect values were obtained by 3${\times}$3 moving windowing of Zevenbergen & Thorno's Method, and then the slopes of the study area were determined using the aspect values. After the lineament factors in the alluvial deposits were revealed by comparing the threshold values, the first rank lineament under the alluvial deposits were extracted using the Hough transform In order to extract the final lineament, the lowest points under the alluvial deposits in a given topographic section perpendicular to the first rank lineament were determined through the spline interpolation, and then the final lineament were chosen through Hough transform using the lowest points. The algorithm developed in this study enables us to observe a clearer lineament in the areas covered by much larger alluvial deposits compared with the results extracted using the conventional existing algorithm. There exists, however, some differences between the first rank lineament, obtained using the aspect and the slope, and the final lineament. This study shows that the new algorithm more effectively extracts the lineament in the area covered with wide alluvlal deposits than in the areas of converging slope, areas with narrow alluvial deposits or valleys.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.18 no.2
• /
• pp.174-189
• /
• 2015

The objective of this study was to reconstruct spatial information using the results of the investigation and evaluation of the health of the living organisms, habitat, and water quality at the investigation points for the aquatic ecosystem health of the Nakdong River basin, to support the rational decision making of the aquatic ecosystem preservation and restoration policies of the Nakdong River basin using spatial analysis techniques, and to present efficient management methods. To analyze the aquatic ecosystem health of the Nakdong River basin, punctiform data were constructed based on the position information of each point with the aquatic ecosystem health investigation and evaluation results of 250 investigation sections. To apply the spatial analysis technique, the data need to be reconstructed into areal data. For this purpose, spatial influence and trends were analyzed using the Kriging interpolation(ArcGIS 10.1, Geostatistical Analysis), and were reconstructed into areal data. To analyze the spatial distribution characteristics of the Nakdong River basin health based on these analytical results, hotspot(Getis-Ord Gi, $G^*_i$), LISA(Local Indicator of Spatial Association), and standard deviational ellipse analyses were used. The hotspot analysis results showed that the hotspot basins of the biotic indices(TDI, BMI, FAI) were the Andong Dam upstream, Wangpicheon, and the Imha Dam basin, and that the health grades of their biotic indices were good. The coldspot basins were Nakdong River Namhae, the Nakdong River mouth, and the Suyeong River basin. The LISA analysis results showed that the exceptional areas were Gahwacheon, the Hapcheon Dam, and the Yeong River upstream basin. These areas had high bio-health indices, but their surrounding basins were low and required management for aquatic ecosystem health. The hotspot basins of the physicochemical factor(BOD) were the Nakdong River downstream basin, Suyeong River, Hoeya River, and the Nakdong River Namhae basin, whereas the coldspot basins were the upstream basins of the Nakdong River tributaries, including Andong Dam, Imha Dam, and Yeong River. The hotspots of the habitat and riverside environment factor(HRI) were different from the hotspots and coldspots of each factor in the LISA analysis results. In general, the habitat and riverside environment of the Nakdong River mainstream and tributaries, including the Nakdong river upstream, Andong Dam, Imha Dam, and the Hapcheon Dam basin, had good health. The coldspot basins of the habitat and riverside environment also showed low health indices of the biotic indices and physicochemical factors, thus requiring management of the habitat and riverside environment. As a result of the time-series analysis with a standard deviation ellipsoid, the areas with good aquatic ecosystem health of the organisms, habitat, and riverside environment showed a tendency to move northward, and the BOD results showed different directions and concentrations by the year of investigation. These aquatic ecosystem health analysis results can provide not only the health management information for each investigation spot but also information for managing the aquatic ecosystem in the catchment unit for the working research staff as well as for the water environment researchers in the future, based on spatial information.

• Journal of Intelligence and Information Systems
• /
• v.25 no.1
• /
• pp.163-177
• /
• 2019

As smartphones are getting widely used, human activity recognition (HAR) tasks for recognizing personal activities of smartphone users with multimodal data have been actively studied recently. The research area is expanding from the recognition of the simple body movement of an individual user to the recognition of low-level behavior and high-level behavior. However, HAR tasks for recognizing interaction behavior with other people, such as whether the user is accompanying or communicating with someone else, have gotten less attention so far. And previous research for recognizing interaction behavior has usually depended on audio, Bluetooth, and Wi-Fi sensors, which are vulnerable to privacy issues and require much time to collect enough data. Whereas physical sensors including accelerometer, magnetic field and gyroscope sensors are less vulnerable to privacy issues and can collect a large amount of data within a short time. In this paper, a method for detecting accompanying status based on deep learning model by only using multimodal physical sensor data, such as an accelerometer, magnetic field and gyroscope, was proposed. The accompanying status was defined as a redefinition of a part of the user interaction behavior, including whether the user is accompanying with an acquaintance at a close distance and the user is actively communicating with the acquaintance. A framework based on convolutional neural networks (CNN) and long short-term memory (LSTM) recurrent networks for classifying accompanying and conversation was proposed. First, a data preprocessing method which consists of time synchronization of multimodal data from different physical sensors, data normalization and sequence data generation was introduced. We applied the nearest interpolation to synchronize the time of collected data from different sensors. Normalization was performed for each x, y, z axis value of the sensor data, and the sequence data was generated according to the sliding window method. Then, the sequence data became the input for CNN, where feature maps representing local dependencies of the original sequence are extracted. The CNN consisted of 3 convolutional layers and did not have a pooling layer to maintain the temporal information of the sequence data. Next, LSTM recurrent networks received the feature maps, learned long-term dependencies from them and extracted features. The LSTM recurrent networks consisted of two layers, each with 128 cells. Finally, the extracted features were used for classification by softmax classifier. The loss function of the model was cross entropy function and the weights of the model were randomly initialized on a normal distribution with an average of 0 and a standard deviation of 0.1. The model was trained using adaptive moment estimation (ADAM) optimization algorithm and the mini batch size was set to 128. We applied dropout to input values of the LSTM recurrent networks to prevent overfitting. The initial learning rate was set to 0.001, and it decreased exponentially by 0.99 at the end of each epoch training. An Android smartphone application was developed and released to collect data. We collected smartphone data for a total of 18 subjects. Using the data, the model classified accompanying and conversation by 98.74% and 98.83% accuracy each. Both the F1 score and accuracy of the model were higher than the F1 score and accuracy of the majority vote classifier, support vector machine, and deep recurrent neural network. In the future research, we will focus on more rigorous multimodal sensor data synchronization methods that minimize the time stamp differences. In addition, we will further study transfer learning method that enables transfer of trained models tailored to the training data to the evaluation data that follows a different distribution. It is expected that a model capable of exhibiting robust recognition performance against changes in data that is not considered in the model learning stage will be obtained.