Adaptive Multi-Scale Feature Fusion Based Residual U-net for Fracture Segmentation in Coal Rock Images
Abstract Accurate segmentation of fractures in coal rock CT images is important for safe production and the development of coalbed methane.However,to make segment coal rock fractures accurate,the challenges as the following:1)The coal rock CT images have the characteristics which are high background noise, sparse target, weak boundary information, uneven gray level, low contrast etc.; 2)There is no a public dataset of coal rock CT images;3)Limited coal rock CT images samples.In the paper,we proposed adaptive multi-scale feature fusion based residual U-uet(AMSFFRU-uet) for fracture segmentation in coal rock CT images to address the issues.In order to reduce the loss of tiny and weak fractures, dilated residual blocks (DResBlock) are embedded into the U-uet structure, which expand the receptive field and extract fracture information atdifferent scales.Furthermore, for reducing the loss of spatial information during the down-sampling process, feature maps of different sizes in the encoding branch are concatenated by adaptive multi-scale featurefusion module,which is as the input of the first up-sampling in the decoding branch.And we applieda set of comprehensive data augmentation operations to increase the diversity of training samples. Our network,U-net and ResU-net are tested on our dataset of coal rock CT images with 5 different textures.The experimental results show that compared with U-net and ResU-net, our proposed approach improve the average Dice coefficient by 5.1% and 2.9% and the average accuracy by 4.5% and 2%,respectively.Therefore,AMSFFRU-net can achieve better segmentation of coal rock fractures,and has stronger generalization ability and robustness.