Research On Coal Gangue Recognition Method Based On Convolutional Neural Network

Coal occupies a pivotal position in my country's energy consumption,and in the process of mining,it will contain a large amount of gangue.Gangue not only has a low burning value,but also produces a lot of harmful gases during the combustion process,so in the production process of coal Sorting out the gangue becomes an essential link.The current manual screening method is inefficient and the sorting effect is not stable.With the rapid development of machine vision technology,researchers began to use images to complete the distinction between coal and gangue.Traditional image-based coal and gangue recognition methods have the disadvantages of low recognition rate and difficulty in extracting features.In view of this,this paper introduces the convolutional neural network algorithm and applies it to the recognition of coal and gangue.Using convolutional neural networks to identify coal and gangue images,sample collection and data set production is a key step.In this paper,the image acquisition process has been reasonably designed.The black background is used to fit the actual production environment as much as possible.The distance between the camera and the acquisition background is fixed to reflect the actual size of coal and gangue as much as possible.Oversampling was performed to maintain the balance of the data,and the diversity of the samples was ensured by collecting images of coal and gangue produced from different locations.And the collected coal and gangue image samples used rotation,translation,and flipping methods to expand the data.In order to take into account the sample usage rate and the number of training times,this paper uses K-fold crossvalidation to divide the data set of coal and gangue images.The current convolutional neural network has obvious differences in algorithm design and structure.In order to obtain a network that is more suitable for coal and gangue identification,this article analyzes multiple classic convolutional neural networks,and specifically analyzes Alex Net,VGGNet,Goog Le Net,and Res Net,respectively compared the network models from the four dimensions of recognition accuracy,depth,parameter amount,and recognition speed,and finally determined to use the latter three to study the recognition of coal and gangue.And try to modify the above-mentioned classic convolutional neural structure,so that it can be better applied to the recognition of coal and gangue.Finally,a comparative test was carried out between the original network and the improved network.The results show that the F1 value of the improved network model has been improved,which proves the effectiveness of the improvement.In order to improve the model's recognition rate of coal and gangue,this paper introduces the method of migration learning.First,the model is fully trained on a largescale data set,and then the coal and gangue data set are used to fine-tune the parameters of the trained model.The experimental results show that the recognition effect of the model on coal and gangue has been significantly improved after using transfer learning.