Feature Extraction for Mining Industry Image Based on SAM: A Case Study From Froth Flotation

Froth image features serve as crucial reference indicators for optimizing productivity under complex flotation process conditions. However, due to the instability of the production environment, calculating the size distribution and froth velocity of froth images proves challenging. This article investigates a feature extraction method based on segment anything model (SAM), a visual foundation model, to address issues of low accuracy and time-intensive froth image feature extraction. Leveraging the instance segmentation performance of the SAM model, the low-rank adaptation fine-tuning method is employed to enhance SAM’s feature modeling capability for froth images. A mask-prompt generator is designed to guide SAM in froth segmentation. Subsequently, a froth velocity calculation method based on SAM’s instance segmentation results is proposed. Only a small amount of training data is required to achieve the desired level of model performance. Ultimately, this article implements a unified algorithmic framework for froth image feature extraction.