Research On Some Problems Of Historical Seismogram Digitization

Historical seismograms contain important information about historical earthquakes and play a significant role in geoscience for providing critical data. They are recorded on mediums like paper, film and smoked paper. In early years of seismology, these historical seismograms are important materials for research, presenting firsthand data for scholars and technicians. Even in today's research, these valuable and irreplaceable seismograms remain their great value. However, with the research development in current digital age, the seismograms recorded on paper, film and smoked paper hardly can be applied to digital and quantitative analysis. Besides, they are of limited life span and uneasy to store for occupying large space, making it difficult to reorganize and maintain them. Therefore it is a reasonable option to digitize the historical seismograms. Manual digitization of historical seismograms would be a laborious task due to their huge quantity and complicated content. Therefore automatic digitization of massive historical seismograms tends to be an important issue. But due to the design flaws of recording equipment, those historical seismograms recorded in early years are not apt to their digitization for some reasons such as geometric distortions, intersections and disconnections in or between waveforms, which are the difficult issues in the digitization as well as the main concerns discussed in this paper.Through a summarization of existing methods of digitizing historical seismograms and a comprehensive analysis of characteristics of historical seismograms, some methods are put forward in this paper to deal with the three key problems, i.e. arc distortion correction, curve tracing and seismograms inpainting. The contributions of this paper can be summarized as follows.1) A method based on low-rank matrix recovery technique is proposed to correct misalignment distortion and arc distortion of seismograms. Misalignment distortion is caused by the misalignment of the seismogram during the scanning process, while the arc distortion is attributed to the finite length of the pen arm of seismometer. Given that a seismogram is inherently regular and symmetrical, this paper makes use of low-rank property to describe seismogram texture and proposes a distortion correction method based on low-rank matrix recovery technique. Thanks to the regular and symmetrical property of seismogram texture, this algorithm is able to estimate deformation parameters automatically and accurately and then to use them to correct the misalignment distortion and arc distortion. The proposed method can automatically correct the distorted seismograms, and works efficiently and robustly, making it applicable in practical use.2) An algorithm for automatic tracing seismograms based on hierarchical clustering is proposed to trace the seismic waveforms. To separate the intersections in waveforms remains a difficult problem in automatic tracing waveforms. Existing tracing waveform methods can hardly select the trend of waveforms at the intersection automatically due to the complexity and variety of seismic waveforms. Firstly, the proposed algorithm uses a differential geometric line extraction algorithm to extract the points of seismic curve and the scattered points are connected into smooth segments based on smoothness criterion. Then based on the hierarchical relationship between the entire seismic curve and its constituent parts, the algorithm employs the geometrical characteristics of each curve segment and adopts the hierarchical clustering method to group the curve segments that belong to the same seismic wave into one cluster and complete the automatic tracking of seismic wave curves. In contrast with traditional methods, the proposed algorithm can precisely determine the trend of seismic curves at the intersections automatically to trace the curves without human interaction.3) An inpainting algorithm using sparse representation with spectra characteristics as a prior is proposed when some portions of the waveform are lost. Design defects or malfunctions of recording equipment or deficiency of preservation medium can bring defections to the curves, hindering the scientific research. An inpainting algorithm based on sparse representation is presented to inpaint the lost information. In order to improve the precision of the method, the average value of spectra of every frame is incorporated as a prior into the inpainting algorithm. The proposed algorithm combines the sparsity of each frame and information between the frames to compensate for the lost information effectively. Experiments demonstrate that this algorithm works well even when the lost sample duration is very large.