基于迁移学习的煤矸图像识别方法  

IMAGE RECOGNITION OF COAL-GANGUE BASED ON TRANSFER LEARNING

作　　者：张释如[1] 张达 Zhang Shiru;Zhang Da(College of Communication and Information Engineering,Xi’an University of Science and Technology,Xi’an 710054,Shaanxi,China)

机构地区：[1]西安科技大学通信与信息工程学院,陕西西安710054

出　　处：《计算机应用与软件》2024年第11期247-250,260,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(51774234);陕西省榆林市科技计划项目(CXY-2020-035)。

摘　　要：针对传统煤矸图像识别算法需提取并筛选图像灰度、纹理等特征,费时耗力,以及训练卷积神经网络需庞大数据集和高配置硬件设备等问题,提出基于迁移学习的煤矸图像识别方法。利用VGG16卷积基提取煤矸图像特征,并与机器学习算法结合,验证VGG16卷积基提取特征的有效性。分别通过特征提取和模型微调方式实现网络模型VGG16的迁移,并构建自定义密集连接分类器,形成两种识别模型。仿真结果显示,两种模型的准确率分别达到96.30%和98.15%。结果表明:提出的煤矸识别模型是有效的,可以快速准确识别煤和矸石图像。The traditional coal-gangue image recognition algorithms need to extract and filter specific image features,which is time-consuming and labor-intensive,and the reconstruction and training of convolutional neural networks require huge data sets and high-configuration hardware equipment.This paper proposes recognition methods for coal-gangue images based on the transfer learning.Combined with VGG16 convolution basis for extracting coal-gangue images features and machine learning algorithm,the effectiveness of VGG16 convolution basis feature extraction was verified.The migration of network model was realized through feature extraction and model fine-tuning.This paper constructed two customized dense connection classifiers,and obtained two classification models.The simulation results show that the accuracy rates are 96.30%and 98.15%respectively.The coal-gangue identification models obtained by the transfer learning are effective,and they are able to identify coal and gangue images quickly and accurately.

关 键 词：煤矸识别 机器学习 卷积神经网络 迁移学习 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]