嵌入空洞卷积和批归一化模块的智能煤矸识别算法  

Intelligent coal and gangue identification algorithm embedded in dilated convolution and batch normalization module

作　　者：郭永存[1] 张勇[1,2] 李飞[3] 杨鹏 GUO Yongcun;ZHANG Yong;LI Fei;YANG Peng(State Key Laboratory of Mining Response and Disaster Prevention and Control in Deep Coal Mines,Anhui University of Science and Technology,Huainan 232001,China;School of Electrical and Information Engineering,Anhui University of Science and Technology,Huainan 232001,China;School of Mechanical Engineering,Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学深部煤矿采动响应与灾害防控国家重点实验室,安徽淮南232001 [2]安徽理工大学电气与信息工程学院,安徽淮南232001 [3]安徽理工大学机械工程学院,安徽淮南232001

出　　处：《矿业安全与环保》2022年第3期45-50,共6页Mining Safety & Environmental Protection

基　　金：安徽省科技重大专项项目(18030901049);国家自然科学基金项目(51904007)。

摘　　要：为解决现有机器视觉煤、矸石识别算法感受野小、特征提取能力低和训练收敛速度慢的问题,提出一种嵌入空洞卷积和批归一化模块的智能煤矸识别算法。该算法利用空洞卷积替换VGGNet16网络中尺寸为3×3的卷积核,增大卷积核感受野、提高网络的特征提取能力,同时在卷积层和激活层之间嵌入批归一化模块,在避免梯度消失的同时可加快模型训练收敛速度。利用搭建的实验装置采集煤和矸石图片,制作煤和矸石图像数据集,对模型进行训练,并基于浮点运算次数FLOPs和F1分数对模型的训练结果和预测效果进行评价。实验结果表明,改进后的煤矸识别算法FLOPs为71 632 538次,测试集F1分数为0.994 3,训练在第5个周期即收敛且准确率达到97%以上。通过与其他网络模型训练结果进行对比,说明所建模型具有较快的收敛速度且预测效果较好。The existing machine vision coal and gangue recognition algorithms have small sensing field, low feature extraction ability and slow training convergence speed. In order to solve these problems, an intelligent coal and gangue identification algorithm embedded in dilated convolution and batch normalization module was proposed. This algorithm used dilated convolution to replace the 3×3 convolution kernel in VGGNet16 network with dilated convolution to increase the receptive field and improve the feature extraction capability of the network. At the same time, a batch normalization module was embedded between the convolutional layer and the activation layer to avoid the disappearance of the gradient and accelerate the convergence rate of model training. The experimental equipment was used to collect coal and gangue images, make coal and gangue image data sets, train the model, and evaluate the training and prediction effect of the model based on FLOPs and F1 scores. The experimental results show that the FLOPs of the improved algorithm is 71 632 538 times, and the F1 score of the test set is 0.994 3. The training converges in the fifth cycle and the accuracy is above 97%. The comparison with the training results of other network models further shows that the proposed model has faster convergence speed and better prediction effect.

关 键 词：机器视觉 煤矸识别 数据增强 空洞卷积 批归一化 模型优化 

分 类 号：TD94[矿业工程—选矿]