1D Local Matching Filter for High-Resolution Seismic Imaging

The patent introduces an adaptive filter as an intermediary component that processes the relationship between low-frequency and high-frequency seismic data. This filter is trained using machine learning to learn the transformation patterns, serving as a mediator that converts low-frequency data to high-frequency data more reliably than traditional spectral extrapolation methods. The adaptive filter reduces errors by learning from training data while maintaining adaptability to different input conditions.

The patent changes the parameters of the filtering process by using adaptive filters with machine learning training. Instead of fixed spectral extrapolation parameters, the system learns optimal filtering parameters from training data, allowing dynamic adjustment of the transformation process. This enables more accurate reconstruction of high-frequency components by adapting to the specific characteristics of the input low-frequency data.

The patent applies partial action by using adaptive filtering to generate high-frequency seismic data from low-frequency data, rather than performing complete full waveform inversion. This partial approach retrieves the essential high-frequency information needed for improved imaging without committing to the full computational expense of iterative FWI. The adaptive filter provides a computationally efficient alternative that captures the most critical frequency transformation needs.

The patent performs preliminary action by training adaptive filters in advance using machine learning on representative seismic data. This pre-training creates a ready-to-use transformation model that can be applied quickly to actual seismic data without requiring iterative computational processes during the actual imaging task. The preliminary training phase separates the computationally intensive learning from the actual application, reducing real-time computational costs.

The patent creates a copy of high-frequency seismic data by transforming low-frequency data through adaptive filtering. Instead of directly acquiring high-frequency data that would be affected by attenuation and noise, the system generates a synthetic copy of what the high-frequency data would look like based on the more robust low-frequency measurements. This copying approach bypasses the harmful effects that plague direct high-frequency acquisition.

The adaptive filter serves as an intermediary that translates information from low-frequency data to high-frequency data domain. This intermediary process allows the system to overcome the direct path problems of high-frequency acquisition (attenuation, noise) by using low-frequency data as a stable foundation and computationally transforming it to the desired high-frequency representation.

The patent changes the frequency parameter of the seismic data through adaptive filtering. The system takes low-frequency data (good for depth penetration) and transforms it to high-frequency data (good for resolution) by learning the appropriate frequency transformation from training data. This parameter transformation allows the system to simultaneously achieve both the depth coverage of low-frequency data and the resolution of high-frequency data.

The patent introduces dynamics into the frequency transformation process by using adaptive filters that can adjust their parameters based on the input data characteristics. Rather than a fixed frequency transformation, the adaptive filter dynamically adjusts its filtering behavior to optimize the conversion from low-frequency to high-frequency data, allowing the system to adapt to different geological conditions and data qualities.