FMI image lamina features and thickness extraction and recognition method basedon machine learning

  Due to the difficulty and high cost of core data, it is very important to use the logging data to continuously identify and divide the longitudinal upper stria structure types of a single well. The conventional logging resolution is mostly decimeter to meter level, while the high-resolution imaging logging resolution can reach 5 mm, and the imaging logging dynamic and static images obtained on the basis of the resistivity scale can clearly reflect the bedding changes of the formation, which is an important means for fine identification and characterization of the streak. However, it takes a lot of work to manually identify the type and thickness of the striae, and the thickness of the striae is difficult to observe intuitively with the naked eye.

  This study presents a machine learning and wavelet transform-based method for extracting and recognizing texture layer characteristics and thickness from imaging well logging images. Grayscale images are obtained from well logging slices, and grayscale curves are extracted at the fourth quartile. An average grayscale curve is constructed, and wavelet transform is applied to remove noise, yielding a transformed curve. Grayscale differences between pre- and post-transformation curves are calculated to form a difference curve. These grayscale values and differences serve as clustering features, classifying texture layers into four types. Texture layer thickness is then statistically analyzed based on these types. The method enables automated texture layer and thickness recognition, enhancing accuracy and efficiency in feature extraction.