Cell confluency, as an important parameter for evaluating cell growth status and determining the timing of cell processing, reflects the degree of coverage of cells on the culture surface. However, traditional manual estimation of confluency has strong subjectivity, and stem cell images under the microscope are often affected by image blur and uneven illumination, which influence the segmentation results. To address these challenges, this paper proposes a dual-threshold Otsu cell image segmentation method combining adaptive preprocessing and PSO optimization. Initially, the adaptive weighted multi-scale Retinex detail enhancement technique is used to preprocess the original cell images for illumination balance and cell feature enhancement. Subsequently, an improved PSO algorithm is used to accelerate the dual-threshold Otsu method to segment the preprocessed images, followed by post-processing to optimize the segmentation, and finally the cell confluency is calculated. The proposed method is validated using images of umbilical cord stem cells during the proliferation phase. Results show that the calculated confluency values align closely with expert evaluations. In addition, the method improves computational efficiency, reduces manual subjectivity and error, and provides a practical solution for automated stem cell culture analysis.