Research On Online Detection Method Of Coal Quality Based On Deep Learning

In the construction of smart mines,the automatic identification of coal and gangue helps to promote the remote and unmanned/less human aspects of coal mining,sorting and transportation.Based on the background of online coal quality detection in intelligent coal mine construction,this paper studies the automatic identification method of coal and gangue based on deep learning algorithm and different neural network models.The main content of this article is as follows.Considering the small coal gangue image data set,an image recognition method of coal gangue based on artificial extraction features is proposed,that is,MPSO-BP neural network improved by intelligent optimization algorithm is used for classification recognition based on the combination of artificial extraction of gray scale and texture features.The MPSO-BP neural network algorithm adaptively adjusts the speed of the algorithm in optimization according to the multi-attractive and multi-repulsive forces,and randomly changes the updating strategy of the particles,so that the partial particles can carry out local fine search.The experimental results show that the image recognition accuracy of this method is 89%,although the recognition accuracy is not high,but the learning rate is fast,and it can effectively identify coal and gangue when the sample is small.Considering the large data sets,an image recognition method of coal gangue based on convolutional neural network is proposed,that is,transfer learning is introduced into the established Alex Net,VGG-16 and Inception-V3 convolutional neural network models,and different gradient descent algorithms are used to optimize network parameters,so as to classify and recognize coal gangue images.The theory based on transfer learning is to transfer the learned model parameters to the small sample coal gangue image data set in this paper through pre-training on the large data set.By comparing the three pre-trained models,the recognition accuracy is up to 96.72%.Therefore,in most cases,there are many image samples qualified for pre-training,and deep learning-based convolutional neural network coal-gangue image recognition method often achieves better results.