Ash Content Detection in Coal Slime Flotation Tailings Based on MobileViT

In order to effectively guide the production of coal slime flotation, it is necessary to detect the ash content of the tailings in a timely and accurate manner. Addressing the issues of low accuracy, high parameter count, and lack of model lightweightness in existing ash content detection algorithms, a tailings ash content detection method based on the improved lightweight network MobileViT is proposed. The MobileViT integrates the characteristics of both CNN and Vision Transformer. It employs CNN to provide spatial inductive bias, accelerating the convergence and inference speed of the network. Simultaneously, it incorporates the self-attention mechanism and global receptive field of Transformer. This approach achieves lightweight design while balancing the requirements for performance and accuracy. In this paper, optimizations were conducted on MobileViT. Utilizing MobileViT as the backbone network, a residual structure was introduced between the input and fusion blocks within the MobileViT module, optimizing the network model at a deeper level. Compared to other models, the improved MobileViT demonstrates superiority in both accuracy and lightweightness in ash content detection tasks.