欢迎访问《图学学报》 分享到：
图学学报

图学学报 ›› 2021, Vol. 42 ›› Issue (6): 917-923.DOI: 10.11996/JG.j.2095-302X.2021060917

• 图像处理与计算机视觉 • 上一篇    下一篇

基于三通道分离特征融合与支持向量机的混凝土 图像分类研究

  

  1. 1. 中北大学理学院，山西 太原 030051； 2. 中北大学信息与通信工程学院，山西 太原 030051
  • 出版日期:2022-01-18 发布日期:2022-01-18
  • 基金资助:
    国家自然科学基金项目(61774137)；山西省自然科学基金项目(201801D121026) 

Research on concrete image classification based on three-channel separation feature fusion and support vector machine 

  1. 1. School of Science, North University of China, Taiyuan Shanxi 030051, China; 2. School of Information and Communication Engineering, North University of China, Taiyuan Shanxi 030051, China
  • Online:2022-01-18 Published:2022-01-18
  • Supported by:
    National Natural Science Foundation of China (61774137); Shanxi Provincial Natural Science Foundation of China (201801D121026)

PDF (PC)

64

摘要： 混凝土的不同配合比可决定材料的性能，对于多种配比和粒径大小混凝土图像的分类研究，有 利于工业废弃混凝土的高效回收利用。为了提升分类效果，提出了一种新的特征提取模块(ITFA-DLF)，该模块 在图像分离重构出的 R，G 和 B 3 个通道上，使用卷积神经网络(CNN)提取 3 通道图像的颜色特征，通过多块 局部二值模式(MB-LBP)提取 3 通道图像的纹理特征，将 2 种特征进行融合并输入到网格搜索算法(GS)优化的 支持向量机(SVM)中进行分类。采用混凝土图像进行实验，对比多种分类方法得出所提模型的效果最佳，9 类 图像识别率达到了 92%以上，在保证分类精度的同时缩短了分类时间，提高了混凝土图像的分类效率，验证了 所提方法的有效性。

关键词: 混凝土图像, 卷积神经网络, 多块局部二值模式, 特征融合, 支持向量机

Abstract: The different mix ratios of concrete determine the performance of the material. The research on the classification of concrete images with various mix ratios and particle sizes is conducive to the efficient recycling of industrial waste concrete. In order to improve the classification effect, a new feature extraction module—image texture feature aided deep learning feature (ITFA-DLF) was proposed. This module employed convolution on the R, G, and B channels reconstructed from image separation and reconstruction. Convolutional neural network (CNN) extracted the color features of the three-channel image, utilized the multi-block local binary pattern (MB-LBP) to extract the texture features of the three-channel image, and merged the two features and input them into the support vector machine (SVM) optimized by the grid search (GS) algorithm for classification. Experiments with concrete images were adopted to compare various classification methods. It is concluded that the model proposed can produce the best effect. The recognition rate of nine types of images has reached more than 92%, and the classification time was shortened while ensuring the classification accuracy, and the classification efficiency of the concrete image was improved, which verified the effectiveness of the proposed method. 

Key words:  , concrete image, convolutional neural network, multi-block local binary pattern, feature fusion, support vector machine 

中图分类号: