16.Ash content category prediction method for foam image based on multi-scale network

Abstract:

To address the issue of delayed ash detection in the coal slime flotation process due to reliance on traditional laboratory analysis, Multi-scale Dual Attention coal slime flotation foam image ash content classification prediction network is proposed. The model initially constructs a dual-branch parallel feature extraction framework using ResNet101 and VGG16, with the innovative introduction of channel attention mechanism. A weighted feature fusion strategy is designed to optimize the integration of multi-scale features, and a sequentially connected DoubleAttention module is constructed to enhance the extraction of key features. Simultaneously, the incorporation of a LacunarityPooling module, which calculates local variance, effectively enhances the representation of texture features.Experiments demonstrate that the network model achieves a 94.55% accuracy in predicting ash content classification, improving by 31.83% compared to the traditional dual-VGG branch network. The inference speed of the model reaches 119.54 FPS, improved by 52.70% compared to BCNN. This method provides an effective technical solution for real-time monitoring of the flotation process, offering significant practical value in enhancing the production efficiency and economic benefits of coal washing plants.

Keywords: multi-scale feature fusion;deep learning;slime flotation;foam image analysis;ash classification prediction