• 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.

• Korean Journal of Remote Sensing
• /
• v.36 no.3
• /
• pp.449-460
• /
• 2020

Super-resolution is a technique used to reconstruct an image with low-resolution into that of high-resolution. Recently, deep-learning based super resolution has become the mainstream, and applications of these methods are widely used in the remote sensing field. In this paper, we propose a super-resolution method based on the deep back-projection network model to improve the satellite image resolution by the factor of four. In the process, we customized the loss function with the edge loss to result in a more detailed feature of the boundary of each object and to improve the stability of the model training using generative adversarial network based on Wasserstein distance loss. Also, we have applied the detail preserving image down-scaling method to enhance the naturalness of the training output. Finally, by including the modified-residual learning with a panchromatic feature in the final step of the training process. Our proposed method is able to reconstruct fine features and high frequency information. Comparing the results of our method with that of the others, we propose that the super-resolution method improves the sharpness and the clarity of WorldView-3 and KOMPSAT-2 images.

• Journal of Broadcast Engineering
• /
• v.23 no.3
• /
• pp.351-360
• /
• 2018

Human emotion recognition is a research topic that is receiving continuous attention in computer vision and artificial intelligence domains. This paper proposes a method for classifying human emotions through multiple neural networks based on multi-modal signals which consist of image, landmark, and audio in a wild environment. The proposed method has the following features. First, the learning performance of the image-based network is greatly improved by employing both multi-task learning and semi-supervised learning using the spatio-temporal characteristic of videos. Second, a model for converting 1-dimensional (1D) landmark information of face into two-dimensional (2D) images, is newly proposed, and a CNN-LSTM network based on the model is proposed for better emotion recognition. Third, based on an observation that audio signals are often very effective for specific emotions, we propose an audio deep learning mechanism robust to the specific emotions. Finally, so-called emotion adaptive fusion is applied to enable synergy of multiple networks. The proposed network improves emotion classification performance by appropriately integrating existing supervised learning and semi-supervised learning networks. In the fifth attempt on the given test set in the EmotiW2017 challenge, the proposed method achieved a classification accuracy of 57.12%.

• Journal of Internet Computing and Services
• /
• v.20 no.6
• /
• pp.55-63
• /
• 2019

Recently, the use of real-world image data has been increasing to express realistic virtual environments in various application fields such as education, manufacturing, and construction. In particular, with increasing interest in digital twins like smart cities, realistic 3D urban models are being built using real-world images, such as aerial images. However, the captured aerial image includes shadows from the sun, and the 3D city model including the shadows has a problem of distorting and expressing information to the user. Many studies have been conducted to remove the shadow, but it is recognized as a challenging problem that is still difficult to solve. In this paper, we construct a virtual environment dataset including the height map of buildings using 3D spatial information provided by VWorld, and We propose a new shadow detection method using height map and deep learning. According to the experimental results, We can observed that the shadow detection error rate is reduced when using the height map.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.11
• /
• pp.457-464
• /
• 2021

The advent of the big data era has enabled the rapid development of deep learning that learns rules by itself from data. In particular, the performance of CNN algorithms has reached the level of self-adjusting the source data itself. However, the existing image processing method only deals with the image data itself, and does not sufficiently consider the heterogeneous environment in which the image is generated. Images generated in a heterogeneous environment may have the same information, but their features may be expressed differently depending on the photographing environment. This means that not only the different environmental information of each image but also the same information are represented by different features, which may degrade the performance of the image analysis model. Therefore, in this paper, we propose a method to improve the performance of the image color constancy model based on Adversarial Learning that uses image data generated in a heterogeneous environment simultaneously. Specifically, the proposed methodology operates with the interaction of the 'Domain Discriminator' that predicts the environment in which the image was taken and the 'Illumination Estimator' that predicts the lighting value. As a result of conducting an experiment on 7,022 images taken in heterogeneous environments to evaluate the performance of the proposed methodology, the proposed methodology showed superior performance in terms of Angular Error compared to the existing methods.

• Korean Journal of Remote Sensing
• /
• v.38 no.6_2
• /
• pp.1743-1747
• /
• 2022

Sentinel-2 can be used as proxy data for the Korean Compact Advanced Satellite 500-4 (CAS500-4), also known as Agriculture and Forestry Satellite, in terms of spectral wavelengths and spatial resolution. This letter examined cloud detection for later use in the CAS500-4 based on deep learning technologies. DeepLabV3+, a traditional Convolutional Neural Network (CNN) model, and Shifted Windows (Swin) Transformer, a state-of-the-art (SOTA) Transformer model, were compared using 22,728 images provided by Radiant Earth Foundation (REF). Swin Transformer showed a better performance with a precision of 0.886 and a recall of 0.875, which is a balanced result, unbiased between over- and under-estimation. Deep learning-based cloud detection is expected to be a future operational module for CAS500-4 through optimization for the Korean Peninsula.

• Korean Journal of Remote Sensing
• /
• v.38 no.5_2
• /
• pp.793-810
• /
• 2022

Synthetic Aperture Radar (SAR) is considered to be suitable for near real-time inundation monitoring. The distinctly different intensity between water and land makes it adequate for waterbody detection, but the intrinsic speckle noise and variable intensity of SAR images decrease the accuracy of waterbody detection. In this study, we suggest two modules, named 'morphology module' and 'edge-enhanced module', which are the combinations of pooling layers and convolutional layers, improving the accuracy of waterbody detection. The morphology module is composed of min-pooling layers and max-pooling layers, which shows the effect of morphological transformation. The edge-enhanced module is composed of convolution layers, which has the fixed weights of the traditional edge detection algorithm. After comparing the accuracy of various versions of each module for U-Net, we found that the optimal combination is the case that the morphology module of min-pooling and successive layers of min-pooling and max-pooling, and the edge-enhanced module of Scharr filter were the inputs of conv9. This morphologic and edge-enhanced U-Net improved the F1-score by 9.81% than the original U-Net. Qualitative inspection showed that our model has capability of detecting small-sized waterbody and detailed edge of water, which are the distinct advancement of the model presented in this research, compared to the original U-Net.

• Journal of Korea Society of Industrial Information Systems
• /
• v.28 no.5
• /
• pp.1-14
• /
• 2023

Soybeans are one of the world's top five staple crops and a major source of plant-based protein. Due to their susceptibility to climate change, which can significantly impact grain production, the National Agricultural Science Institute is conducting research on crop phenotypes through growth analysis of various soybean varieties. While the process of capturing growth progression photos of soybeans is automated, the verification, recording, and analysis of growth stages are currently done manually. In this paper, we designed and trained a YOLOv5s model to detect soybean leaf objects from image data of soybean plants and a Convolution Neural Network (CNN) model to judgement the unfolding status of the detected soybean leaves. We combined these two models and implemented an algorithm that distinguishes layers based on the coordinates of detected soybean leaves. As a result, we developed a program that takes time-series data of soybeans as input and performs growth analysis. The program can accurately determine the growth stages of soybeans up to the second or third compound leaves.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.1
• /
• pp.21-30
• /
• 2024

In this paper, we propose a method to build a sample dataset of the features of eight sensor-only facilities built as infrastructure for autonomous cooperative driving. The feature extracted from point cloud data acquired by LiDAR and build them into the sample dataset for recognizing the facilities. In order to build the dataset, eight sensor-only facilities with high-brightness reflector sheets and a sensor acquisition system were developed. To extract the features of facilities located within a certain measurement distance from the acquired point cloud data, a cylindrical projection method was applied to the extracted points after applying DBSCAN method for points and then a modified OTSU method for reflected intensity. Coordinates of 3D points, projected coordinates of 2D, and reflection intensity were set as the features of the facility, and the dataset was built along with labels. In order to check the effectiveness of the facility dataset built based on LiDAR data, a common CNN model was selected and tested after training, showing an accuracy of about 90% or more, confirming the possibility of facility recognition. Through continuous experiments, we will improve the feature extraction algorithm for building the proposed dataset and improve its performance, and develop a dedicated model for recognizing sensor-only facilities for autonomous cooperative driving.

• Journal of Korean Tunnelling and Underground Space Association
• /
• v.26 no.3
• /
• pp.281-301
• /
• 2024

The power cable tunnels which are part of the underground transmission line project, are constructed using the shield TBM method. The disc cutter among the shield TBM components plays an important role in breaking rock mass. Efficient tunnel construction is possible only when appropriate replacement occurs as the wear limit is reached or damage such as uneven wear occurs. A study was conducted to determine the wear conditions of disc cutter using a deep learning algorithm based on real-time measurement data of wear and rotation speed. Based on the results of full-scaled tunnelling tests, it was confirmed that measurement data was obtained differently depending on the wear conditions of disc cutter. Using real-time measurement data, an algorithm was developed to determine disc cutter wear characteristics based on a convolutional neural network model. Distributional patterns of data can be learned through CNN filters, and the performance of the model that can classify uniform wear and uneven wear through these pattern features.