Abstract:Time-series data is crucial for understanding and responding to natural disasters such as earthquakes. However, due to confidentiality, access restrictions, and other reasons, time-series data are often extracted from images containing time and frequency domain data. Traditional methods of directly digitizing time-series data from these images tend to have significant errors in both the time and the frequency domains. To address this issue, an image time-series data analysis method with the in clusion of time and frequency domain characteristics was proposed. First, the time and frequency do main data were extracted from the images using digital method, with the more accurate frequency do main data curve identified as the target frequency domain data. Continuous wavelet transform was then used to correct the time-series data directly extracted from the images, so as to improve the accuracy of the time-series data analysis. The proposed method was demonstrated and explained using the typi cal time-series data of ground motion as an example. A comparison was made with the traditional method of digitizing earthquake motion directly from images and frequency domain method based on response spectrum matching. Furthermore, the method was applied to the damage assessment of the 6.5-magnitude earthquake in Taitung County, Taiwan Province in 2022. The primary conclusions are as follows: the proposed image time-series data analysis method can take into account both the time and frequency domain information, allowing for more accurate extraction of time domain data. This re sults in smaller errors when used in subsequent engineering disaster prevention analysis, providing an important method for data identification from time-series images.
