基于多颜色和局部纹理的水果识别算法研究  

Research on Fruit Recognition Algorithm Based on Multi-Color and Local Texture

作　　者：陈雪鑫 卜庆凯[1] CHEN Xuexin;BU Qingkai(School of Electronic Information,Qingdao University,Qingdao 266071,China)

机构地区：[1]青岛大学电子信息学院

出　　处：《青岛大学学报（工程技术版）》2019年第3期52-58,共7页Journal of Qingdao University(Engineering & Technology Edition)

摘　　要：为提高水果种类识别的准确性,本文提出一种基于多颜色特征和纹理特征的水果识别算法。该研究选择不同种类的水果图像作为实验测试样本,使用最大类间方差法Otsu分割图像,得到水果图像的目标区域,分别对目标区域进行红、绿、蓝(RGB)颜色模型和色调、饱和度、明度(HSV)颜色模型的直方图分析,采用颜色矩算法和非均匀量化算法对RGB模型和HSV模型提取特征,利用局部二值模式(local binary patterns,LBP)对目标区域提取局部纹理特征,对颜色和纹理特征向量进行优化组合,结合基于梯度下降算法的BP神经网络对测试样本进行训练分类。针对输入层输入不同特征分别进行实验并比较,得到基于不同特征的水果识别率。研究结果表明,本算法分类识别率可达90%以上,高于单一特征算法识别率。该研究具有一定的实际应用价值。In order to improve the accuracy of recognition of fruit types,this paper proposes a fruit recognition algorithm based on multi-color feature and texture feature.Different kinds of fruit images are selected as experimental samples.By using the between-cluster variance method(Otsu)we get fruit image segmentation algorithm of target area,separate the target area and RGB(Red,Green,Blue)and HSV(Hue,Saturation,Value)histogram analysis of color model,combining with the characteristics of analysis,and use of color moment algorithm for RGB model to extract the characteristic;Non-uniform quantization algorithm is used to extract features from HSV model.Local Binary Patterns(LBP)is then used to extract Local texture features of the target region.The obtained color feature vectors and texture feature vectors are optimized and combined with BP neural network based on gradient descent algorithm to train and classify the test samples.At the same time,on the basis of the same test sample,the training and classification of a single feature vector are carried out.By comparing the simulation results with the algorithm results in this paper,it can be found that the classification recognition rate(the recognition rate is more than 90%)of the algorithm scheme in this study is more accurate,which has certain practical application value.

关 键 词：图像处理 局部二进制模式 颜色空间 神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]