Abstract:The pulp grade is one of the key parameters in the flotation process, which plays a key role in guiding production, saving chemicals, controlling product quality and improving recovery. In order to predict the flotation concentrate grade online and solve the problem of the detection lag of the fluorescence analyzer, a deep learning-based online prediction model of concentrate grade is developed that does not require subjective extraction of features. The value and tailings grade value are input, and the output concentrate grade value is a regression problem. Comparing the differences in the prediction results when the backbone network is VGG-16, ResNet-50 and MobileNet-V2, the experimental results show that VGG-16 has the best prediction accuracy and robustness.The average prediction accuracy is 12.48%.