In view of the coexistence of image information and process data information in flotation process and small differences among features of different operation state， a novel operating performance assessment method based on multi?source heterogeneous information and deep learning was proposed for flotation process. Firstly， a residual network （ResNet） is established to extract deep features with more discrimination from original images of different performance grades. Secondly， a stacked sparse performance?relevant autoencoders （SSPAE） model is proposed， which introduces the state level label into the model training to overcome the problem that the traditional autoencoder ignores the state?related characteristics. Furthermore， an image and data feature fusion model based on attention mechanism （AM） is established， and then the fused features are used as the inputs of the SoftMax classifier to train the operating performance assessment model， realizing the reasonable and effective utilization of the multi?source heterogeneous information. Finally， the flotation process data is used for simulation verification. The simulation results show that the proposed method is superior to other comparative methods， verifying its superiority in evaluating the operating performance of flotation processes.
