一种基于多尺度特征复用残差网络的矿山图像重建算法  

A Mine Image Reconstruction Algorithm Based on Multi-scale Feature Multiplexing Residual Network

作　　者：马琳 苏明[1] 兰义湧[2] MA Lin;SU Ming;LAN Yiyong(College of Science and Information Technology,Beijing Open University,Beijing 100081,China;College of Science,Minzu University of China,Beijing 100081,China)

机构地区：[1]北京开放大学科学技术学院,北京100081 [2]中央民族大学理学院,北京100081

出　　处：《金属矿山》2023年第11期228-233,共6页Metal Mine

基　　金：北京市教委科技计划一般项目(编号:KM202251160001)。

摘　　要：针对矿山图像重建中细节损失导致重建质量低下的问题,提出了一种基于多尺度特征复用残差网络的矿山图像重建算法,旨在提高矿山场景下图像重建的精度和效率。首先,设计了一个多尺度特征提取模块,通过堆叠多个并行的卷积层和池化层,结合局部残差网络构建图像特征提取模块,通过不同尺度的多路组合网络,从输入图像中充分提取图像的多尺度细节特征。这些特征表示具有不同的语义信息和空间分辨率,能够捕捉到图像中的不同细节和纹理结构。然后,引入了特征复用模块,将不同尺度的特征进行融合和复用,以增强图像重建的准确性。通过多尺度的特征交互和信息传递,可以有效地利用全局和局部的上下文信息,提高图像的重建性能。通过在自建的矿山图像重建数据集上进行试验,结果表明:所提出的算法在重建精度和效率方面均得到了显著提升,与其他深度学习模型相比,该算法在重建图像的细节保留和结构准确性方面表现出更好的性能。此外,该算法具有较快的训练和推断速度,适用于实际应用场景。In order to improve the accuracy and efficiency of mine image reconstruction,a mining image reconstruction algorithm based on multi-scale feature multiplexing residual network is proposed to solve the problem of low reconstruction quality caused by detail loss in mine image reconstruction.Firstly,a multi-scale feature extraction module is designed.By stac-king multiple parallel convolution layers and pooling layers,the image feature extraction module is constructed in combination with the local residual network,and the multi-scale detail feature representation of the image is fully extracted from the input image through the multi-channel combination network of different scales.These features represent different semantic information and spatial resolution,which can capture different details and texture structures in the image.Then,a feature reuse module is introduced to fuse and reuse features of different scales to enhance the accuracy of image reconstruction.Through multi-scale feature interaction and information transfer,global and local context information can be effectively used to improve image recon-struction performance.Experiments on the self-built mine image reconstruction dataset show that the proposed algorithm has a-chieved significant improvement in reconstruction accuracy and efficiency.Compared with other deep learning models,the pro-posed algorithm has better performance in detail retention and structural accuracy of reconstructed images.In addition,the algo-rithm has a fast training and inference speed,which is suitable for practical application scenarios.

关 键 词：矿山图像重建 多尺度特征复用 残差网络 图像质量 

分 类 号：TD67[矿业工程—矿山机电]