Research On Image Recognition Method Of Coal Gangue Based On Machine Learning

Coal mining is accompanied by the mixing of gangue,at present,the selection of coal-gangue is mainly based on the combination of artificial and machine selection method,the level of intelligence needs to be improved.By analyzing the differences between the density attributes,machinery and optical characteristics,summarizing the current method of selecting and comparing its merits and faults,on the basis of understanding the research status of the analysis and classification algorithm of the coal-gangue image,the identification rate of the image selection method which is convenient for intelligent control is not high,and the single classification algorithm of SVM,BP,ELM and its AdaBoost integrated strategy is applied to the research and recognition idea of the coal-gangue image.Based on the understanding of the selection and identification method,the Kalman-FOPID controlled ring LED light source is designed,combined with the development board Hi3518 to collect the sample image of the coal-gangue,and the difference between the images of different grayscale samples is compared,as well as the effect of the filtering methods of different convolution cores,select the weighted averaging method and the Gaussian filter of 3*3 convolution cores to pre-process the coal-gangue image,so as to extract different texture characteristics such as Tamura,and combine with the color characteristics of the coal-gangue extracted from the original image,to obtain a data set of the coal-gangue sample with a dimension of 396*93.In turn,each single classification algorithm is applied to the coal-gangue data set,the best parameters of the SVM model obtained by grey wolf and other optimization algorithms,and the recognition effect of different nodes to determine the best structure of each neural network,and the performance comparison of the models under the optimal parameter structure is compared.The recognition effect of SVM and KELM is similar and not less than 99.24%,which is better than BP and ELM,in which the SVM training time is slightly longer than ELM but shorter than BP,and the shortest model training time is KELM,which is a simple parameter determination process,at 0.0088s.AdaBoost integration under the best structural parameters of the single classification algorithm,and applied to the coal-gangue data set,according to the different model recognition effect of the base classifier number,select the suitable integration model,compared with the respective base classifier,SVM and KELM model integration recognition effect is limited,The training time of the ELM and BP integrated models increased,but the recognition effect improved significantly,among which the ELM integration was more obvious,and the recognition effect of AdaBoost-ELM was 1,but the training time was longer than the KELM,compared with the best performing KELM in the single classification algorithmBased on the research of the coal-gangue identification method,the image acquisition system is designed,the STM32F4 control AR0130 is used to collect the coal-gangue image,transmitted via Ethernet to the upper computer,and the design interface of C#is used to control the lower machine,receive the coal-gangue image data and call different algorithms for the identification of the coal-gangue.Figur[42]table[12]reference[131]...