Research On Mineral Microscopic Images Recognition Based On Machine Vision

Identification of mineral types is the basis of process mineralogy research.The detection of particle size and content is very important for the selection of beneficiation process and has important research significance.However,traditional method relies on manual identification and measurement,which is low efficiency,strong subjectivity,and difficult to unify.Aiming at those problems,based on the existing machine learning and image processing methods,the ore detection model is built to realize the automatic identification,segmentation and quantitative calculation of minerals.The main contents of the study are as follows:(1)Aiming at the low automation level of mineral recognition,four semantic segmentation networks for mineral recognition have been constructed.A total of 350mineral microscopic images were collected,and the dataset was expanded with data enhancement strategy.Four ore segmentation models is built,and the transfer learning method is used to improve convergence speed.The optimal super parameter of each network is selected through several experiments,and the performance of each network on the mineral microscopic image dataset is compared.The experimental results show that the m Io U of UNet is 87.1%,and the m Io U of Deep Labv3+is 87%.(2)Aiming at the problems of large network parameters and low precision,two improved methods for the structure are proposed.The optimization method of large kernel convolution is researched.The large kernel convolution is introduced into the UNet network to improve the expression ability of the model.A new LK-UNet model is designed.The experimental results show that the parameters of LK-UNet are only8.377×10~6,far less than the original UNet,and the m Io U increased by 1.85%.the Io U of chalcopyrite is 93.1%.An improved and efficient Deep Labv3+model is designed.A second Skip Connections is added to Deep Labv3+to obtain details,and an efficient multimodal convolution module is introduced into the ASPP module.The experimental results show that the m Io U of the improved Deep Labv3+is 1.35%higher than that of the original Deep Labv3+.Compare the two improved networks.The m Io U of LK-UNet is0.6%higher than the improved Deep Labv3+,and the image segmentation effect is also better.(3)Aiming at the complex process of traditional quantitative detection methods,a particle size distribution and content detection method based on digital image processing technology is proposed.The calculation flow of mineral particles is designed by morphological operation.The problem of mineral particle adhesion is solved by using open operation.The mineral content is calculated by the mineral pixel area,and the relationship between the mineral content distribution and the number of pictures is researched.The experimental results show that the method of quantitative statistics of minerals by image processing has good advantages and reliability compared with the manual statistical method.Figure 59;Table 15;Reference 62...