Research On Image Segmentation Of Rock Particles In Rockfill Dam Mining

The mining of rockfill dam is one of the important links in the construction of water conservancy projects.The block control of mining materials largely affects the quality and efficiency of the construction of the dam and the subsequent construction.The rapid and high-precision detection of the explosion pile is of great significance for optimizing the blasting parameters,blasting methods,improving the construction efficiency and the economic benefits in the construction process.The traditional methods for obtaining the particle size grading of the blasting pile is screening.Although the traditional on-site screening test has accurate results,it often takes more time and labor costs to meet the requirements of fast and real-time control.The image segmentation method produced in the 1960 s has become one of the main methods for particle size grading detection by virtue of its fast,accurate and low cost.However,the traditional segmentation methods have some problems,such as oversegmentation,under-segmentation,manual adjustment,etc.These problem limit the application in actual engineering.Especially in recent years,the water conservancy project is developing in the direction of intelligence and unmannedness.It is especially important to study a fast and high-precision rock particle image segmentation method.In response to the above questions,the research contents and results of the thesis are as follows:(1)Aiming at the problem of over-segmentation of traditional image segmentation methods for the super diameter rock with complex texture structure in the pile,this paper proposes a rock particle image segmentation method based on Mask R-CNN.In this paper,the candidate classifier structure of the network is improved,and the batch normalization layer is added in the network,which reduces the over-fitting phenomenon and accelerates the convergence speed of the network.By modifying the code of the visualization module,the single channel mask image segmentation is obtained.The experimental results show that the rock particle size segmentation based on Mask RCNN has higher accuracy for larger rocks and avoids over-segmentation caused by complex rock components.(2)Aiming at the shortage of rock particle image sample set for rockfill dam mining,a method based on GAN for rock particle image sample set expansion is proposed.In this paper,a DCGAN network adapted to the task of rock particle image is built.Random noise is input into the network,and real rock particle image samples are generated by the method of anti-learning.Through the traditional data expansion methods such as horizontal flipping and scale transformation,more than 1000 rock particle image data sets of rockfill dam mining materials are obtained.(3)Aiming at the problem of under-segmentation based on Mask R-CNN segmentation method for the fines in the explosion reactor,this paper proposes a method based on traversing pixels to fuse Mask R-CNN segmentation image and rock particle original image mask image and rock particle original image.The merged image is segmented using edge detection based watershed algorithm.The experimental results show that the rock particle image of the rockfill dam mining can be quickly and accurately segmented by using the segmentation method based on Mask R-CNN and watershed algorithm.(4)In this paper,the proposed method is compared with the other four traditional particle image segmentation methods,and analyze the results of their respective segments.The analysis results prove that it can achieve faster and more accurate segmentation of rock particle images,and can be better applied in practical engineering.