摘要
针对现有煤矸分选方法存在模型复杂、实时性差、特征易丢失等问题,构建了一种轻量化煤矸分选网络GC-ResNet18。GC-ResNet18利用幽灵卷积(ghost convolution, GC)线性生成ghost映射的特性,剔除煤和矸石相似性特征的冗余信息。借助Softpool的下采样激活映射,保留、凸显煤和矸石的特征信息并去除冗余参数,防止过拟合现象。引入GC自注意力机制,融合SENet的轻量化和NLNet长距离信息全局捕获的优势,使网络记忆、放大煤矸图像间的细微差异特征,提升煤矸图像的识别准确率。实验结果表明,GC降低了46.6%的参数量,GC自注意力机制在CIFAR10、CIFAR100上分别提升1.44%、2.32%的准确率,而Softpool池化在上述两个数据集中分别提升了0.22%、0.17%。通过对比实验,全面改进后的GC-ResNet18网络在训练效率和分类精度上优于VGG19-S-GDCNN、SBP-VGG-16等模型,在CIFAR10和CIFAR100数据集中的分类精度与同规模的网络相比均达到最优的94.07%和74.95%,并最终在自建煤矸数据集上达到了97.2%的分类准确率。
Aiming at the problems of complex model, poor real-time performance and easy loss of characteristics in the existing coal gangue separation methods, a lightweight coal gangue separation network GC-resnet18 is constructed.GC-resnet18 uses ghost convolution(GC) to linearly generate ghost map to eliminate redundant information of similarity characteristics of coal and gangue.With the help of Softpool′s down sampling activation mapping, the characteristic information of coal and gangue is retained and highlighted, and redundant parameters are removed to prevent over fitting.GC self-attention mechanism is introduced to integrate the advantages of lightweight of SEnet and global capture of long-distance information of NLNet, so as to make the network remember and enlarge the subtle difference characteristics between coal gangue images, and improve the accuracy of coal gangue image recognition accuracy.The experimental results show that GC reduces the amount of parameters by 46.6%,and the accuracy of GC self-attention mechanism is improved by 1.44% and 2.32% on CIFAR10 and CIFAR100 respectively, while Softpool pooling is improved by 0.22% and 0.17% on the above two data sets respectively.Through comparative experiments, the comprehensively improved GC-ResNet18 network is better than VGG19-S-GDCNN,SBP-VGG-16 and other models in training efficiency and classification accuracy.The classification accuracy in CIFAR10 and CIFAR100 data sets is the best 94.07% and 74.95% compared with the network of the same scale, and finally reaches 97.2% classification accuracy in self-built coal gangue data sets.
作者
王天奇
贾晓芬
杜圣杰
郭永存
黄友锐
赵佰亭
WANG Tianqi;JIA Xiaofen;DU Shengjie;GUO Yongcun;HUANG Yourui;ZHAO Baiting(School of Electrical and Information Engineering,Anhui University of Science and Technology,Huainan,Anhui 232001,China;State Key Laboratory of Mining Response and Disaster Prevention and Control in Deep Coal Mines,Anhui University of Science and Technology,Huainan,Anhui 232001,China)
出处
《光电子．激光》
CAS
CSCD
北大核心
2023年第1期19-25,共7页
Journal of Optoelectronics·Laser
基金
国家自然科学基金面上项目(52174141)
安徽省自然科学基金面上项目(2108085ME158)
安徽省重点研究与开放计划(202104a07020005)
安徽高校协同创新项目(GXXT-2020-54)资助项目。
关键词
煤矸分选
神经网络
幽灵卷积(GC)
高效池化层
自注意机制
coal gangue sorting
neural network
ghost convolution(GC)
high efficient pooling layer
self-attention mechanism
