Concentrate grade prediction method based on bimodal CNN secondary transfer learning

• Main Authors: YAN XIN, ZHU KUNHUA, LIAO YIPENG
• Format: Patent
• Language: Chinese; English
• Published: 28.07.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; DATA PROCESSING SYSTEMS OR METHODS, SPECIALLY ADAPTED FORADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORYOR FORECASTING PURPOSES; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE,COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTINGPURPOSES, NOT OTHERWISE PROVIDED FOR

The invention provides a concentrate grade prediction method based on bimodal CNN secondary transfer learning. The method comprises the following steps: firstly, constructing a foam bimodal image deep learning network model based on improved SE-DenseNet, and pre-training the model by means of an RGB-D big data set; secondly, small-scale data sets in different dosing states are constructed to retrain a final convolutional layer, a full connection layer and softmax of the migrated model; and finally, a self-adaptive depth kernel extreme learning machine is adopted to replace a full connection layer and softmax to carry out transfer learning again, and a concentrate grade prediction model under various dosing states is obtained. According to the method, the difference degree of adjacent grade image features can be effectively expanded under the condition of a small-scale training set, the error recognition rate is reduced, the overfitting problem of single transfer learning is effectively solved, and the method