Research On Coal Gangue Image Recognition Based On Deep Learning

Coal gangue is similar to coal in color and shape,and has a low calorific value.It is easy to produce harmful gases and pollute the environment in the combustion process.Therefore,it is essential to select gangue from coal in coal mining industry.The traditional coal gangue picking method not only needs to consume a large amount of financial and material resources,but also the accuracy of coal gangue identification is low.At present,deep learning has achieved good results in many fields,and has been gradually applied to coal gangue image recognition.The deep learning algorithm for image recognition mainly relies on a large number of im age data sets.Through model training,the image features of coal gangue are automatically extracted and the images are classified.So the quality of the coal gangue image will directly affect the accuracy of coal gangue identification.However,the image quality of coal gangue is easily affected by illumination,dust,photographing equipment and other factors,and there is relative movement between photographing equipment and coal gangue,which will inevitably cause the problem of image blur.How to improve the quality of the coal gangue image data sets and how to increase the characteristic information of coal gangue image data set is the main research content in this paper.The main work of this paper is as follows:Study coal gangue image data,collected some groups of optical and thermal coal gangue image data sets,analyze the characteristics and disadvantages of thermal images and visible light images,and put forward a kind of fusion data sets,the visible image and thermal image of coal gangue are combined to increase the recognition features and useful image information of coal gangue image,so as to further improve the ability of coal gangue image data set to resist environmental interference in the recognition process.Aiming at the problem of image blurring in the process of coal gangue image collection,an image blurring method based on Pix2Pix model is proposed.Under the condition of unknown noise and unknown object motion speed,effective information of fuzzy degradation model is obtained through training and learning of a large number of data,finally the fuzzy image is restored.The experimental results show that using the Pix2Pix antagonistic generation model to restore the fuzzy image of coal gangue,although the image cannot be completely restored,the edge contour and basic texture structure of coal gangue can be effectively restore.In this paper,the classic AlexNet convolutional neural network is used to build a coal gangue image recognition model.In the different environment,the different image data of coal gangue are studied for coal gangue recognition.Comparing and analyzing the total training loss and recognition accuracy results after model training.The experimental results show that strength of light and temperature change have obvious influence on the accuracy of coal gangue identification.Finally in the general environment,using visible light images,thermal images,and fusion image training coal gangue identification model,the experimental results show that the fusion image is used to identify the coal gangue,can effectively improve the recognition accuracy of coal gangue.Figure[39]table[12]reference[75]...