Detecting earthquakes using smartphones or IoT devices in real-time is an arduous and challenging task, not only because it is constrained with the hard real-time issue but also due to the similarity of earthquake signals and the non-earthquake signals (i.e., noise or other activities). Moreover, the variety of human activities also makes it more difficult when a smartphone is used as an earthquake detecting sensor. To that end, in this article, we leverage a machine learning technique with earthquake features rather than traditional seismic methods. First, we split the detection task into two categories including static environment and dynamic environment. Then, we experimentally evaluate different features and propose the most appropriate machine learning model and features for the static environment to tackle the issue of noisy components and detect earthquakes in real-time with less false alarm rates. The experimental result of the proposed model shows promising results not only on the given dataset but also on the unseen data pointing to the generalization characteristics of the model. Finally, we demonstrate that the proposed model can be also used in the dynamic environment if it is trained with different dataset.

The advancement of hardware and software technologies makes it possible to use smartphones or Internet of things for monitoring environments in realtime. In recent years, much effort has been made to develop a smartphone based earthquake early warning system, where low-cost acceleration sensors inside a smartphones are used for capturing earthquake signals. However, because a smartphone comes with a powerful CPU, spacious memory, and several sensors, it is waste of such resources to use it only for detecting earthquakes. Furthermore, because a smartphone is mostly in use during the daytime, the acquired data cannot be used for detecting earthquakes due to human activities. Therefore, in this article, we introduce a stand-alone device equipped with a low-cost acceleration sensor and least computing resources to detect earthquakes. To that end, we first select an appropriate acceleration sensor by assessing the performance and accuracy of four different sensors. Then, we design and develop an earthquake alert device. To detect earthquakes, we employ a simple machine learning technique which trains an earthquake detection model with daily motions, noise data recorded in buildings, and earthquakes recorded in the past. Furthermore, we evaluate the four acceleration sensors by recording two realistic earthquakes on a shake-table. In the experiments, the results show that the developed earthquake alert device can successfully detect earthquakes and send a warning message to nearby devices, thereby enabling proactive responses to earthquakes.

Recently, the Internet of Things (IoT) systems have been widely used for earthquake detection because of the easy construction of a dense seismic network, communication capabilities, and low cost of sensors. However, when utilizing MEMS sensors as seismic sensors, earthquake detection capabilities are often affected by various types of noises because such sensors are installed in heterogeneous environments. In earthquakes, P-waves first arrive, but their lengths are only a few seconds, and their amplitudes are also relatively smaller than to S-waves. As a result, it is difficult to accurately detect P-waves in IoT systems where environmental noises are always present. Furthermore, when using deep learning approaches for earthquake detection, inference time usually becomes a critical factor for real-time processing because of the complex architecture of a detection model. To that end, in this letter, we present a deep learning model that can detect P-waves in noisy environments. The model outputs the detection probability before the arrival of strong shakes. We tested our model on earthquakes recorded by the IoT-based seismic sensors deployed in South Korea. Our model can detect P-waves within 1.5–2.5 s after the first arrival of P-wave with the accuracy of 98.8%, making it applicable in real-time earthquake detection.

With the recent increase in the number of earthquakes in Korea, research efforts have been directed toward the real-time detection of earthquakes and the formulation of evacuation plans. Traditional seismometers can precisely record earthquakes but are incapable of processing them on-site to initiate an alert and response mechanism. By contrast, internet of things (IoT) devices equipped with accelerometers and CPUs can record and detect earthquake signals in real time and send out alert messages to nearby users. However, the signals recorded on IoT devices are noisy because of two main factors: the urban buildings and structures these devices are installed in and their cost-quality trade-off. Therefore, in this work, we provide an effective mechanism to deal with the problem of false alarms in IoT devices. We test our previously proposed artificial neural network (ANN) with different feature window sizes ranging from 2 seconds to 6 seconds and with various earthquake intensities. We find that setting the size of the feature window to a certain interval (i.e., 4-5 seconds) can improve model performance. Moreover, an evacuation route guidance platform that considers user location is proposed. The proposed platform provides and visualizes information to user devices in real time through the communication between server and user devices. In the event of a disaster, safe shelters are selected on the basis of the information entered from the server, and pedestrian paths are provided. As a result, the direct and secondary damages caused by earthquakes can be avoided.

The performance of a deep learning (DL) model depends on sufficient training datasets and its algorithmic structure. Even though seismological research using low-cost micro-electro-mechanical systems (MEMS) sensor received much attention recently, because of the lack of data recorded by such MEMS sensors whose data are usually polluted by different types of noise. Therefore, increasing seismic datasets is required by intelligently generating seismic data through data-augmentation techniques. However, it is difficult to characterize and measure the evolution process of seismic sequences, making the feature extraction and data generation of seismic sequences still a significant challenge. By combining the framework of Generative Adversarial Network (GAN) with long short-term memory (LSTM), attention mechanism and neural network (NN), a novel deep generation model (DGM) named EQGAN is developed to overcome the challenges, which can automatically capture the different time histories and dimension characteristics of seismic sequences, meanwhile stably generating high-quality seismic data. The reality of generated data is qualitatively clarified through the analysis of frequency domain and data autocorrelation distribution. Based on the High-throughput Screening (HTS) Theory, the quantitative evaluation index of statistical metrics is designed, and the generation performance of different machine learning models (standard GAN, LSTM, NN) is compared to prove the stability and effectiveness of EQGAN. The experimental results denote that the EQGAN has excellent stability and performance (up to 81%, much higher than that of other generation models), which provides a suitable data expansion approach for the field of seismological research.