Abstract: In order to address the issues of strong subjectivity in human observation and significant delay in chemical analysis
during the picromerite flotation process,this study proposes an online monitoring method for flotation froth grade based
on deep learning models. A dataset containing 766 000 froth images with corresponding SO2-
4 grade values was constructed,and
image enhancement and feature optimization were comprehensively applied using Gaussian blur,morphological processing,and
BRISK feature extraction. The prediction performance of three models including AlexNet,VGG16,and ResNet50 was systematically
compared. The results indicated a strong correlation between froth morphology and SO2-
4 grade. Among the models,Res-
Net50,leveraging its residual learning mechanism,effectively improved feature representation capabilities,achieving an overall
classification accuracy of 94. 16%,significantly outperforming the other models. This research provides a reliable technical approach
for real-time perception and intelligent control of picromerite flotation grade.

Key words: picromerite,flotation froth,deep learning,grade prediction