• Journal of Biomedical Engineering Research
• /
• v.25 no.1
• /
• pp.11-19
• /
• 2004

Stage 1 sleep provides important information regarding interpretation of nocturnal polysomnography, particularly sleep onset. It is a short transition period from wakeful consciousness to sleep. Lack of prominent sleep events characterizing stage 1 sleep is a major obstacle in automatic sleep stage scoring. In this study, we attempted to utilize simultaneous EEC and EOG processing and analyses to detect stage 1 sleep automatically. Relative powers of the alpha waves and the theta waves were calculated from spectral estimation. Either the relative power of alpha waves less than 50％ or the relative power of theta waves more than 23％ was regarded as stage 1 sleep. SEM (slow eye movement) was defined as the duration of both eye movement ranging from 1.5 to 4 seconds and regarded also as stage 1 sleep. If one of these three criteria was met, the epoch was regarded as stage 1 sleep. Results f ere compared to the manual rating results done by two polysomnography experts. Total of 169 epochs was analyzed. Agreement rate for stage 1 sleep between automatic detection and manual scoring was 79.3％ and Cohen's Kappa was 0.586 (p＜0.01). A significant portion (32％) of automatically detected stage 1 sleep included SEM. Generally, digitally-scored sleep s1aging shows the accuracy up to 70％. Considering potential difficulties in stage 1 sleep scoring, the accuracy of 79.3％ in this study seems to be robust enough. Simultaneous analysis of EOG provides differential value to the present study from previous oneswhich mainly depended on EEG analysis. The issue of close relationship between SEM and stage 1 sleep raised by Kinnariet at. remains to be a valid one in this study.

• Korean Journal of Remote Sensing
• /
• v.39 no.5_3
• /
• pp.933-948
• /
• 2023

Atmospheric aerosols not only have adverse effects on human health but also exert direct and indirect impacts on the climate system. Consequently, it is imperative to comprehend the characteristics and spatiotemporal distribution of aerosols. Numerous research endeavors have been undertaken to monitor aerosols, predominantly through the retrieval of aerosol optical depth (AOD) via satellite-based observations. Nonetheless, this approach primarily relies on a look-up table-based inversion algorithm, characterized by computationally intensive operations and associated uncertainties. In this study, a novel high-resolution AOD direct retrieval algorithm, leveraging machine learning, was developed using top-of-atmosphere reflectance data derived from the Geostationary Ocean Color Imager-II (GOCI-II), in conjunction with their differences from the past 30-day minimum reflectance, and meteorological variables from numerical models. The Light Gradient Boosting Machine (LGBM) technique was harnessed, and the resultant estimates underwent rigorous validation encompassing random, temporal, and spatial N-fold cross-validation (CV) using ground-based observation data from Aerosol Robotic Network (AERONET) AOD. The three CV results consistently demonstrated robust performance, yielding R²=0.70-0.80, RMSE=0.08-0.09, and within the expected error (EE) of 75.2-85.1%. The Shapley Additive exPlanations(SHAP) analysis confirmed the substantial influence of reflectance-related variables on AOD estimation. A comprehensive examination of the spatiotemporal distribution of AOD in Seoul and Ulsan revealed that the developed LGBM model yielded results that are in close concordance with AERONET AOD over time, thereby confirming its suitability for AOD retrieval at high spatiotemporal resolution (i.e., hourly, 250 m). Furthermore, upon comparing data coverage, it was ascertained that the LGBM model enhanced data retrieval frequency by approximately 8.8% in comparison to the GOCI-II L2 AOD products, ameliorating issues associated with excessive masking over very illuminated surfaces that are often encountered in physics-based AOD retrieval processes.

• 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.