Objective The advent of atomic force microscope (AFM) provides a powerful tool for the studies of life sciences. Particularly, AFM-based indentation assay has become an important method for the detection of cellular mechanics, yielding numerous novel insights into the physiological and pathological activities from the single-cell level and considerably complementing traditional biochemical ensemble-averaged assays. However, current AFM indentation technology is mainly dependent on manual operation with low efficiency, seriously restricting its practical applications in the field of life sciences. Here, a method based on the combination of deep learning image recognition and AFM is developed for precisely and efficiently detecting the mechanical properties of single isolated cells and clustered cells.Methods The YOLO deep learning algorithm was used to recognize the central region of the cell in the optical image, the dual UNet neural network with an embedded vision transformer (ViT) module was used to recognize the peripheral regions of cell, and the template matching algorithm was used to recognize the tip of spherical probe. Based on the automatic determination of the positional relationships between the microsphere tip and the different parts of cell, the AFM tip was accurately moved to the central and peripheral regions of the target cell for rapid measurements of cell mechanical properties. Two types of cells, including HEK 293 cell (human embryonic kidney cell) and HGC-27 cell (human undifferentiated gastric cancer cell), were used for the experiments. The Hertz model was applied to analyze the force curves obtained on cells to obtain cellular Young’s modulus.Results AFM probe can be precisely moved to the different parts (central areas and peripheral areas) of cells to perform mechanical measurements under the guidance of deep learning-based optical image automatic recognition. The experimental results show that the proposed method is not only suitable for single isolated cells, but also suitable for clustered cells.Conclusion The research demonstrates the great potential of deep learning image recognition to aid AFM in the precise and efficient detection of cellular mechanical properties mechanics, and combining deep learning-based image recognition with AFM will benefit the development of high-throughput AFM-based methodology to measure the mechanical properties of cells.