Image segmentation of electrical imaging logging is core for characterizing complex reservoirs, and its precision determines the reliability of reservoir parameter extraction. In conventional methods, the image segmentation threshold is determined at the valley between peaks of the gray histogram. But in electrical imaging logging images, the differences between components are small and their distribution ranges are wide, resulting in a unimodal gray histogram without obvious valleys. To address this, a method is proposed to decompose the total spectrum into multiple sub-spectra based on the tailing characteristics of the unimodal distribution, thereby determining the segmentation threshold. Considering the asymmetry and spectral complexity of the gray histogram of electrical imaging logging, the Gaussian distribution is inadequate to characterize the features of each component sub-spectrum. Hence, spectral decomposition is realized based on the asymmetric generalized Gaussian distribution. In this distribution, bilateral skewness factors are introduced to characterize the asymmetry of the spectrum, and kurtosis factors are introduced to characterize spectral shapes from impulsive to flat-topped, thus greatly improving the ability to represent complex spectral shapes. Under the new parameter framework, the number of model parameters increases and the optimization difficulty is significantly enhanced. Therefore, the iterative update equations for all parameters are derived within the framework of the expectation-maximization algorithm, and the Newton iteration method is introduced to solve equations without analytical solutions. The proposed method is applied to the segmentation of electrical imaging logging images of gravelly sandstone of the W Formation in the M Sag, Junggar Basin. The results show that even for unimodal gray distributions, the gray spectrum is effectively decomposed and reliable thresholds are extracted using the proposed method. After segmentation, the boundaries of gravel and other components in the electrical imaging logging images are clear and complete. Lithology classification is performed based on the content of each component, and a coincidence rate of 85.6% with core descriptions is obtained, which is an improvement of 8.3% compared with the histogram method, indicating superior segmentation performance. A valuable reference is provided for electrical imaging log segmentation in complex reservoirs.