Study Of Selection Algorithm Of Waste Rock From Coal Bulk Based On Machine Vision

Coal is the main energy in our country,plays a vital role in the development of national economy.In the process of coal mining,a lot of coal gangue is often produced.The main component of coal gangue is rock whose density is large,ash is high,and the calorific value is small.If the coal gangue is mixed in the coal used for industrial production,it will seriously affect the quality of the coal combustion.Therefore,coal and coal gangue must be separated before industrial production.In recent years,thanks to the popularity of high-performance computers and imaging equipment,the method of computer vision based coal gangue sorting has become feasible.Differences in coal gangue,the gloss of coal is darker,and the texture between them is also quite different.Therefore,gray and its space distributing of coal and coal gangue are also different.It can be seen that the analysis of the gray and texture of the coal and coal gangue will help to identify them.Based on the above analysis,this paper proposes 3 features extracted from gray level co-occurrence matrix(GLCM)and then classified by support vector machine(SVM).Furthermore,an image object detection algorithm is presented,so better results is gained.In the process of the calculation of GLCM,GPU is used to accelerate the process,and brings forward following viewpoints:(1)In this paper,we discuss the change of gray level can be a major hindrance to the running time of GPU.The main reason is that there are read-write collisions happens when multiple threads read and write the same address,resulting in serialization of these thread operations.In view of the above problems,a algorithm is put forward,in this paper,multiple GLCM copies are placed in the shared memory of each active thread block,and threads in the thread block use different rules to write the voting results into different replicas.(2)It takes half the total time to transmit data from the host to the device.it produces great limitations on the efficiency and real-time performance of the program.To solve this problem,this paper proposes an image blocking processing strategy based on CUDA Streaming.The purpose of this strategy is to allow data transmission and program computation to be carried out at the same time as possible,so as to achieve load balancing between data transmission and kernel execution.The convolution neural network is a data driven machine learning method,which is different from the traditional algorithm.Unlike the above algorithm,convolutional neural network is a data driven machine learning method,which greatly promotes the development of computer vision technology.It can independently study the texture features of coal and coal gangue,without the need for manual selection,by training the nonlinear model,the original data will be transformed into a higher level and more abstract expression.For classification tasks,high-level expression can enhance the ability to distinguish input data and weaken the unrelated factors.Therefore,this paper combines the classification task of coal and coal gangue and the convolution neural network,and puts forward the coal gangue sorting algorithm based on the convolution neural network.The main contributions are as follows:(1)According to the current research and application status,this paper first applies convolution neural network to the classification of coal and coal gangue.(2)Because the data set for training is small,it may increase the difficulty of the training of the convolution neural network,and it is very easy to appear over fitting and so on.In this paper,the strategy of using migration learning is proposed,and the convolution neural network is initialized by using the weight of convolution neural network learned from other fields.The effectiveness of the strategy is proved by the experiment.(3)In order to facilitate the future algorithm to be more stable and efficient to be transplanted on the hardware platform.In this paper,the weight pruning operation of the network after training is carried out.Experiments show that the strategy makes the storage space of the model reduced effectively,and the speed of operation is accelerated.