基于ResNet的石油焦与冶金焦图像分类  

Image classification of petroleum coke and metallurgical coke based on ResNet

作　　者：王洪栋 储杰 高思念 陈晨[1] 曹英华 孙金萍[1] WANG Hongdong;CHU Jie;GAO Sinian;CHEN Chen;CAO Yinghua;SUN Jinping(College of Information Engineering,Xuzhou University of Technology,Xuzhou 221018,China;School of Information and Control Engineering,China University of Mining and Technology,Xuzhou 221116,China;Jiangsu Shineng Industrial Technology Co.,Ltd,Xuzhou 221000,China)

机构地区：[1]徐州工程学院信息工程学院,江苏徐州221018 [2]中国矿业大学信息与控制工程学院,江苏徐州221116 [3]江苏仕能工业技术有限公司,江苏徐州221000

出　　处：《江苏理工学院学报》2024年第4期79-84,共6页Journal of Jiangsu University of Technology

基　　金：江苏省高等学校基础科学(自然科学)研究重大项目“复杂场景下基于相关滤波的单目标跟踪技术研究”(22KJA520012);徐州市科技计划项目“智能交通复杂场景下基于相关滤波的交通目标跟踪关键技术研究”(KC22305)。

摘　　要：文章针对石油焦和冶金焦显微图像分类准确率低和小样本的问题,提出基于ResNet的石油焦与冶金焦图像分类模型,通过使用ImageNet大规模数据集上预先训练过的模型,获取更好的特征表示,实验对比了ResNet不同层的分类性能,且对比了训练模型前后的分类效果,确定了ResNet-50在处理该分类任务时的优势。将ResNet-50与其他深度学习模型进行对比分析,研究结果表明:ResNet-50结合预训练模型,能够准确区分2种焦炭类型,且鲁棒性较好。Aiming at the problems of low accuracy and small sample size of petroleum coke and metallurgical coke microscopic image classification,this paper proposes an image classification model of petroleum coke and metallurgical coke based on ResNet.By using the pre-trained model on the large-scale ImageNet dataset,a better feature representation is obtained.The experiment compared the classification performance of different layers of ResNet,and compared the classification effect before and after the training model,and determined the advantages of ResNet-50 in handling this classification task.ResNet-50 is compared and analyzed with other deep learning models,and the results of the study show that ResNet-50,in combination with a pre-trained model,is able to accurately differentiate between the two coke types with good robustness.

关 键 词：ResNet 石油焦 冶金焦 图像分类 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]