Image Segmentation Algorithm Based on Improved U-Net for Mineral Froth Flotation Process

The froth size distribution in mineral flotation production is closely tied to production indicators. Therefore, accurate segmentation of froth images is of great significance for optimizing and controlling the flotation process. However, the high complexity of froth images poses a challenge to the implementation of precise segmentation using the existing segmentation methods. In this study, we propose a novel deep learning algorithm based on U-Net to address the issue of low accuracy in froth image segmentation. The algorithm utilizes the Convolutional Block Attention Module (CBAM) in the downsampling process to focus on edge features and the Atrous Spatial Pyramid Pooling (ASPP) Head module at the decoder end to restore multi-scale information. We create a froth dataset to evaluate the effectiveness of the proposed algorithm. Experimental results show that the proposed algorithm outperforms other existing image segmentation methods and is applicable to different conditions in the flotation process.