The Window Rejection Problem in H/V Analysis: Traditional Versus Neural Network-Based Algorithms

Abstract An essential component of the microtremor horizontal-to-vertical (H/V) spectral ratio analysis is the rejection of transient windows—short-time segments contaminated by local disturbances—that, if not properly excluded, can distort the final H/V curve and compromise interpretation. Traditional automated transient rejection algorithms, such as those based on short-term average versus long-term average ratios, rely on deterministic rules and often either remove an excessive number of windows or fail to exclude subtle but significant noise. Visually aided manual rejection remains among the best-performing approaches but is subjective and time-consuming. In this study, we propose a neural network-based alternative using a U-net architecture trained on a dataset of 1450 recordings that were manually cleaned. The network operates in the frequency domain, taking as input contour plots of H/V ratios over time and producing binary masks that classify each window as acceptable or rejectable. Its performance is compared with that of traditional approaches and evaluated in terms of the ratio of the percentage reduction in H/V standard deviation to the percentage of rejected windows. Results show that the U-net achieves nearly the accuracy of manual approaches, offering a promising tool for efficient and parameter-free window selection in H/V analysis.