Flotation Foam Information Extraction Algorithm Based On Deep Learning

Foam flotation is the most widely used mineral separation method.The visual characteristics of flotation foam surface directly affect the adjustment of various operations and parameters in the flotation process.The traditional manual visual observation of flotation foam state cannot support flotation intelligence.The Machine Vision is used to process the foam image to extract the visual characteristics of the foam surface,and the relationship model is established according to the relationship between the foam characteristics and the flotation related parameters,which can realize the intelligent monitoring and control of flotation treatment.However,the flotation site environment is harsh and the flotation foam state is complex and changeable.The existing foam information extraction methods are difficult to accurately obtain the visual characteristics of the foam surface.This thesis studies the improvement of foam image quality and the accuracy of foam contour segmentation.Foam images often have noise caused by factors such as light and dust.In this thesis,a flotation froth image denoising network HDCR-WGAN based on hybrid dilation convolution residual generator is designed.Based on WGAN,the original generator is improved to a hybrid dilated convolution residual generator to enhance the feature expression ability of the model.The Markovian discriminator is used to guide the generator to obtain a denoised image that is closer to the real noise-free image.This thesis also constructs a multivariate loss function that combines the adversarial loss,pixel loss and multi-scale structural similarity loss of the gradient penalty term to obtain clearer edge details.The experimental results show that HDCR-WGAN can effectively denoise the noisy flotation froth image and retain the detailed information in the image.The existing algorithms have the problem of foam edge over-segmentation for dense foam segmentation.In this regard,this thesis designs a flotation froth image segmentation network RIG-UNet3 + based on multi-scale residual feature extraction and adaptive feature fusion.Based on the Inception module,a multi-scale residual feature extraction module is designed to fully extract the shallow features containing rich edge and texture information in the foam image.The adaptive feature fusion module is used to enhance the learning of important information and suppress the interference of irrelevant information.Aiming at the problem of uneven distribution of target area and background area in flotation froth image,Dice loss function and binary cross entropy loss function are combined to improve the accuracy of edge segmentation.The experimental results show that RIG-UNet3 + can accurately extract the contour edge of the foam in the dense flotation foam image,and then obtain the accurate number,size and shape information of the foam.There are 33 figures,11 tables and 87 references in this thesis.