摘要
依赖特征工程的传统图像识别技术对高度相似叶片图像识别困难,对此提出一种融合方向梯度直方图(HOG)与卷积神经网络(CNN)的图像识别方法。首先由HOG算子提取叶片图像的局部纹理特征,然后将特征向量导入卷积神经网络进行训练、测试和输出分类结果。通过组合对比试验结果表明,该方法能够有效提高数据的鲁棒性,提高叶片图像的平均正确识别率,比多层感知器(MLP)和支持向量机(SVM)分类器的准确率提高了12%左右,平均准确率达到85%。
The traditional image recognition technology relying on feature engineering is difficult to identify highly similar blade images. In this paper, an image recognition method based on histogram of oriented gradient(HOG) and convolutional neural network(CNN) is proposed. Firstly, the local texture features of the leaf image are extracted by the HOG operator, and then the feature vector is imported into the convolutional neural network to train, test and output the classification result. The results of combined comparison experiments show that the proposed method can effectively improve the robustness of data, improve the average correct recognition rate of blade images. Compared with MLP and SVM classifiers, the accuracy is improved by about12%, and the average accuracy rate is 85%.
作者
雷继呈
杨晓滨
罗道兴
上官毅祥
曾森灵
Lei Jicheng;Yang Xiaobin;Luo Daoxing;Shangguan Yixiang;Zeng Senling(Ningde Vocational and Technical College, Fu’an, Fujian 355000, China)
出处
《计算机时代》
2019年第9期53-56,共4页
Computer Era
基金
福建省教育厅中青年教师教育科研项目（JZ181071）
