The ash content data of tailings is of great significance for guiding the coal slime flotation process. In response to the existing issues of lag and low accuracy in current tailings ash content detection methods, an improved Vision Mamba (Vim) method for rapid ash content detection of coal slime flotation tailings is proposed. Without using attention mechanisms, Vim achieves data-driven global visual context modeling by integrating bidirectional State Space Model (SSM) and leverages positional embedding for location-sensitive visual perception. Furthermore, while maintaining high modeling efficiency, it only has sub-quadratic time complexity and linear memory complexity. This paper presents optimizations and improvements to Vim by adding a global feature extraction branch to the Vim Encoder. This enhancement increases the model's ability to perceive features at different scales, accelerates model convergence, and improves inference speed. Compared to other models, the improved Vim model demonstrates superior performance in the task of tailings ash content detection.