To address the problems of low segmentation efficiency, long computation time and low accuracy in multi-threshold image segmentation, a multi-threshold image segmentation algorithm based on an improved mayfly algorithm was proposed. Firstly, the Sobol sequence in the initialization stage was introduced to enhance the traversal and diversity of the population. Secondly, an adaptive nonlinear inertia weight was proposed to balance the global and local merit-seeking ability, improved the convergence efficiency of the algorithm, and facilitated the approximation of the population to the optimal solution. Finally, the exponential entropy was used as the objective function to calculate the fitness, and the multi-threshold combination of image segmentation was determined by the improved mayfly algorithm. Finally, the multi-threshold combination of image segmentation was optimized by improving the mayfly algorithm to determine the optimal segmentation threshold. In order to verify the effectiveness of this improved algorithm, Berkeley images were selected for segmentation verification and compared with other intelligent algorithms. The experimental results show that the improved algorithm outperforms the comparison algorithm in segmentation accuracy, computation time, structure similarity index measure (SSIM) and peak signal-to-noise ratio, (PSNR) and can quickly and effectively solve the multi-threshold segmentation problem of complex multi-target images, which has strong engineering practicality.