基于HOG+LBP特征的油茶果壳籽自动识别方法

Automated Identification of Camellia Shells and Seeds using a Combination of HOG and LBP Features

摘要:

该文利用自行搭建的油茶果壳籽分选装置采集壳籽图像，提出一种HOG和LBP特征融合的方法应用于壳籽的自动识别，实现了油茶果壳籽快速分选。将采集图像进行预处理，分割出壳籽前景图像，提取壳籽的HOG形状特征和LBP纹理特征信息，为了提升检测速率，分别采用主成分分析法（PCA）对HOG特征、LBP特征以及HOG+LBP特征进行降维，利用BP神经网络（BP）、朴素贝叶斯（NBM）、K近邻（KNN）和支持向量机（SVM）等4种机器学习算法训练壳籽分类模型，并通过测试集对比各分类模型性能。实验结果表明，HOG+LBP融合特征的支持向量机分类模型识别效果最好，其训练准确率为100%，平均测试准确率为96.00%，并且特征融合后的识别准确率均高于单一特征，说明该方法对于油茶果壳籽的自动识别有效。

Abstract:

A self-built camellia fruit shell and seed sorting device has achieved rapid sorting of camellia fruit shells and seeds using a HOG-LBP feature fusion method developed for the automated identification of camellia fruit shells and seeds. The device collected shell and seed images, segmented the foreground images of shells and seeds by pre-processing the collected images, and extracted the HOG shape features and LBP texture features of shells and seeds. To improve the detection rate, the dimensionality of HOG features, LBP features, and HOG+LBP features were reduced using principal component analysis (PCA), and the shell and seed classifier was trained using four machine learning algorithms, namely, BP neural network, naive Bayesian model, K-nearest neighbor, and support vector machine. The performance of the trained classifiers were compared with the results from the test set. The experimental results showed that the HOG+LBP features based classifier trained by support vector machine performed best in identification of shells and seeds, with a training accuracy of 100% and an average test accuracy of 96.00%. Moreover, the identification accuracies after feature fusion were all higher than those based on single features, indicating that the proposed method is effective for the automated identification of camellia fruit shells and seeds.