基于卷积神经网络干制哈密大枣纹理分级

Research on the Texture Classification of Dried Hami Jujube Based on Convolutional Neural Network

【目的】研究一种基于卷积神经网络干制哈密大枣纹理分级的方法。利用卷积神经网络解决干制哈密大枣的纹理分类问题。【方法】将大小统一的彩色图片输入网络,卷积核自动提取其纹理特征,进行分类。【结果】分类准确率达到了97.7%。【结论】与常用的灰度共生矩阵提取干制哈密大枣纹理特征(最大概率,相关性,对比度、能量、同质性和熵),再用BP神经网络和支持向量机(SVM)分类准确率相比的方法,避免了复杂纹理提取和图片预处理的过程,在测试时间相近的情况下识别率更高。

【Objective】 Texture detection and grading of dried Hami jujube is a difficult problem to realize the automatic classification of the dried date appearance quality, therefore, a method of texture classification of Hami jujube based on convolutional neural network was proposed.【Method】In this method, color images of uniform size were input into the network, and the convolutional kernel automatically extracted its texture features, and then classified them.【Result】Experimental results showed that the CNN could solve the problem of texture classification of Hami dried jujube, and the accuracy rate of the classification was up to 97.7%.【Conclusion】Compared with the commonly used gray level co-occurrence matrix(GLCM) to extract the texture characteristics(maximum probability, correlation, contrast, energy, homogeneity and entropy) of Hami dried jujube, and then compared with the accuracy classified by BP neural network(BP) and aupport vector machines(SVM), this method avoided complicated texture extraction and image preprocessing, and the recognition rate was higher when the test time was similar.