The invention discloses a method for extracting a port wharf image by utilizing multi-spectral interpretation. The method comprises the steps of: firstly, assigning a grayscale value to extract a fuzzy coastline; secondly, performing smoothing processing and noise removal on a remote sensing image in a target region, and extracting edge information; thirdly, establishing a multi-spectral databaseof a port wharf target; extracting a port wharf according to projected eigenvectors, subjecting a regularized kernel function to MAF transformation again, projecting a multivariate observed value to an original eigenvector, regarding a remote sensing image region corresponding to the original eigenvector less than a transformation variance as the port wharf to be extracted, and carrying out verification; and identifying the port wharf quickly and precisely through the comparison of characteristic spectra based on grayscale value assignment and MAF transformation, and further carrying out verification based on a spatial correlation relationship, thereby improving the accuracy rate of identification.