The gas diffusion layer (GDL) in fuel cell plays a crucial role in mass transfer, electron conduction and structural support. The distribution of binder and hydrophobic agent directly affects the mass transfer and conductivity of the fuel cell. In this study, a multi-threshold segmentation-based image processing method was used to distinguish between pores, fibers, and binder in the GDL. The relationship between binder distribution and fiber density as well as the thickness of the binder layer was investigated. Specifically, the study first utilized a joint distribution map based on grayscale values and non-local means values, with global Rényi entropy as the optimization objective. A bee algorithm was employed to search for the global optimal point, achieving threshold segmentation of micro-scale images. Subsequently, by comparing the results of element segmentation using one-dimensional and multi-dimensional segmentation methods, a three-dimensional convolutional kernel of different sizes was used to obtain the distance from each point in the structure to the edge, effectively separating adherent fibers. Finally, convolutional methods were used to analyze the relationship between the thickness of the binder layer and fiber density. It was found that there is a positive correlation between thickness and spatial density of fibers when the fiber layer is thin. These research findings provide theoretical basis for the content and distribution of binder in the basal layer process of GDL fabrication, and contribute to a more accurate description of the microstructure and properties of actual materials, enhancing the understanding of the preparation process and microstructure properties of GDL in fuel cells. The image processing method utilized in this study enables the computation of crucial parameters within the GDLs' multi-component segmentation algorithm, utilizing a limited sample size. Additionally, it offers valuable insights into the spatial distribution characteristics of fibers and the presence of thin matrix layers. These findings highlight the potential of this method for future applications in the realm of realistic three-dimensional structural reconstructions, thus holding promising prospects for further advancements in this field.