基于深度卷积特征和HOG特征融合的管道病害识别算法
首发时间:2018-06-06
摘要:提出一种利用预训练VGGNet提取的图像特征和HOG特征融合,采用多类SVM识别管道病害的方法。本文利用迁移学习和特征融合的策略,在小样本训练集上也能训练处具有较高识别率的分类模型。模型采用VGGNet的第一层全连接层输出值和图片的HOG特征融合作为SVM的训练特征来训练分类模型。试验表明,在小规模训练集下,本文采用的方法在病害识别中比单独采用HOG+SVM方法或预训练的VGGNet方法具有更高的准确率。
关键词: 管道病害 识别 特征融合 VGGNet 预训练 HOG SVM
For information in English, please click here
Pipeline Disease Recognition Algorithm Based on Deep Convolution Feature and HOG Feature Fusion
Abstract:This paper proposes a method that combines the image features extracted by pre-training VGGNet and the features extracted by HOG, and uses multiple classes of SVM to identify pipeline diseases. In this paper, the strategy of migration learning and feature fusion is used to train the classification model with high recognition rate at the small sample training set. The model uses the VGGNet's first full-connection layer output value and the picture's HOG feature to fuse together as the SVM training feature to train the classification model. Experiments show that under the small-scale training set, the method used in this paper has higher accuracy in disease identification than using the HOG+SVM method or the pre-trained VGGNet method alone.
Keywords: Pipeline disease Recognition Feature fusion VGGNet Pre-training HOG SVM
引用
No.****
同行评议
勘误表
基于深度卷积特征和HOG特征融合的管道病害识别算法
评论
全部评论0/1000