Research On Detection Methods Of Copper Ore Content Based On Machine Learning And Near-Infrared Spectroscopy

China’s copper resources are characterized by many poor mines and few rich mines.However,China’s annual copper consumption accounts for about 49%of the world’s copper consumption and is the world’s largest copper consumer.Therefore,the needs of various industries can only be met through large amounts of imports.If an advanced technology can be found to accurately identify the grade of copper ore,China’s copper ore resources can be utilized to the maximum.At present,most of the existing copper ore grade identification methods are chemical inspection methods.Although this method is reliable,it takes a long time,the process is extremely complicated,and the waste liquid produced will pollute the environment.Therefore,the waste liquid needs to be treated again.This undoubtedly increases the cost of enterprise beneficiation.Therefore,it is necessary to find a suitable method to accurately and efficiently identify the grade of copper ore,so as to reduce the cost of enterprises and increase the utilization rate of copper ore.In view of the above problems,this thesis establishes the inversion model of copper ore content by combining machine learning algorithm with near infrared spectral data of ore.Because the Extreme Learning Machine(ELM)has advantages that traditional machine learning does not have,this thesis uses the ELM algorithm as a basis for in-depth research.Two Hidden Layer Extreme Learning Machine(TELM)is a neural network algorithm based on the ELM algorithm and expanded to a double hidden layer on the basis of a single hidden layer.However,the structure of the TELM algorithm is too fixed,that is,the two hidden layer neuron nodes must be the same,which leads to too few adjustable parameters of the algorithm,and it is easy to enlarge the numerical error when seeking the generalized inverse.Aiming at the problems of the TELM algorithm,this thesis proposes a two hidden layer Extreme learning machine with variable neuron nodes(VTELM).Unlike the TELM algorithm,the VTELM algorithm can freely adjust the number of neuron nodes in each hidden layer.Although this algorithm is flexible,it also causes the algorithm to be too sensitive to the number of neuron nodes.Therefore,the idea of the I-TELM algorithm is adopted to optimize the VTELM algorithm to the I-VTELM algorithm.The simulation results show that the I-VTELM algorithm has the best fitting effect,the Pearson correlation coefficient is 0.992,and the root mean square error is 0.0059.Then,using the I-VTELM algorithm and the Landsat-8 data of the mining area,the remote sensing inversion model of the mining area was established,and the copper content of the entire mining area was successfully reversed,which provided guidance for the company’s future mining and land reclamation.Through the modeling of the I-VTELM algorithm,the copper content in the ore can be accurately obtained,but when facing different mines,it needs to be re-modeled.This means that data need to be collected again and chemical experiment analysis is carried out,which will consume a lot of human and material resources.In response to the above problems,this thesis proposes the TSSA-TELM_AE algorithm to implement transfer learning.First,the adaptive T-distribution mutation operator is introduced into the sparrow search algorithm to improve the global search ability of the sparrow search algorithm.Then use the optimization algorithm to optimize the parameters in ELM_AE,obtain the optimal solution and achieve subspace alignment,and finally use the optimal solution as the input weight of the TELM algorithm to solve the output weight to complete the algorithm.The migration learning of the two mining areas was successfully completed by using the TSSA-TELM_AE algorithm.Although a certain accuracy was lost,the work efficiency was improved and the cost was reduced.