基于颜色分割与GA-SVM的花生表皮破损识别  

Peanut’s Cuticle Damage Recognition Based on Color Segmentation and GA-SVM

作　　者：申志超 赵志衡[1] 卢雷[1] 孙磊 罗思婕 胡琦渊 Shen Zhichao;Zhao Zhiheng;Lu Lei;Sun Lei;Luo Sijie;Hu Qiyuan(School of Electrical Engineering&Automation,Harbin Institute of Technology,Haerbin 150001;Shanghai Anzai Manufacturing Co.,Ltd.,Shanghai 201109)

机构地区：[1]哈尔滨工业大学电气工程及自动化学院,哈尔滨150001 [2]上海安西机械制造有限公司,上海201109

出　　处：《中国粮油学报》2021年第3期140-147,共8页Journal of the Chinese Cereals and Oils Association

摘　　要：传统农作物色选方法以设定颜色阈值为主,具有分类准确率较低、泛化能力较差等缺点,本研究提出基于颜色分割的预处理与遗传算法优化支持向量机参数的花生表皮破损识别算法。根据花生表皮的破损情况将花生分为完好花生及表皮破损花生2类,在不同光照条件下构建了含有多个品种的花生图像数据集。对花生图像提取方向梯度直方图特征,利用支持向量机对花生图像进行分类。为提高分类准确率,在RGB颜色空间基于支持向量机对彩色花生图像进行颜色分割预处理;同时采用软间隔非线性支持向量机模型,并基于遗传算法对模型参数进行寻优。综合优化后的算法在训练集上对花生图像分类时的准确率达到96.88%,在测试集上的准确率达到100%,测试时平均每张花生图像耗时5.6 ms。仿真测试结果表明本文构建的花生表皮破损识别算法对花生品种及光照变化等干扰有较好的鲁棒性,且算法不依赖于人的经验,泛化能力强,具有良好的应用前景。The traditional methods of crop color selection mainly set the color threshold,which featureslow classification accuracy,subjective judgment and poor generalization ability,and other disadvantages.In this paper,a peanut cuticle damage recognition algorithm based on color segmentation pre-processing and Genetic Algorithm to optimize Support Vector Machine parameters was proposed.According to the damage of peanut’s cuticle,peanuts were divided into two types:perfect peanuts and cuticle damaged peanuts.The image data sets of different varieties of peanuts were constructed under different light conditions.The Histogram of Oriented Gradient feature of peanut image was extracted,and the data set of peanut image was classified by Support Vector Machine.In this paper,in order to improve the classification accuracy,color peanut image was pre-processed based on Support Vector Machine in RGB color space,and a color peanut image was transformed into a black-and-white image.This method would be able to improve the classification accuracy more effectively than directly graying the color image.At the same time,this paper used the soft-margin and nonlinear Support Vector Machine model,based on Genetic Algorithm to optimize the model parameters.The accuracy of the algorithm was 96.88%in the training set,100%in the test set,and the average time of each peanut image was 5.6 ms.The simulation results showed that the algorithm has commendable robustness to the interference of peanut varieties and light changes,and the algorithm did not depend on human experience,had strong generalization ability,and had promisingapplication prospects.

关 键 词：颜色分割 遗传算法 方向梯度直方图特征 软间隔非线性支持向量机 破损识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]