<![CDATA[Earthquake detection is a critical component of disaster management, as early identification of seismic events can help mitigate potential damage and support timely response efforts. This study evaluates the application of deep learning for binary earthquake detection using a lightweight Convolutional Neural Network (CNN) based on the MobileNetV2 architecture. The experiments were conducted using seismic waveform data from the Stanford Earthquake Dataset (STEAD), which were transformed into time–frequency representations through the Short-Time Fourier Transform (STFT). Spectrogram images derived from the seismic signals were used as input to the CNN models. Transfer learning was applied to MobileNetV2 to adapt the pretrained architecture to the earthquake detection task. The proposed approach achieved an accuracy of 99%, precision of 100%, recall of 97.96%, and an F1-score of 98.97% on the test dataset. In terms of model complexity, MobileNetV2 has 7,176,600 total parameters and 1,639,538 trainable parameters, indicating a favorable balance between performance and computational efficiency. For comparative evaluation, MobileNetV2 was benchmarked against several commonly used CNN architectures, including CNN Vanilla, MobileNetV1, VGG16, and ResNet, under the same experimental conditions. The results indicate that MobileNetV2 provides competitive detection performance while maintaining a significantly smaller model size. Although real-time deployment on mobile devices was not implemented in this study, the findings suggest that lightweight CNN architectures, such as MobileNetV2, hold promise for future earthquake detection systems operating in resource-constrained environments.]]>
Copyrights © 2026
