基于颜色分割和PSO-RELM算法的花生种子筛选研究
Research on peanut seeds selection based on color segmentation and PSO-RELM algorithm
杨丽 1薛亚许 1李鹏飞 1彭信杰1
作者信息
- 1. 平顶山学院电气与机械工程学院,河南平顶山,467000
- 折叠
摘要
针对花生种子人工筛选存在工作量大、效率低等的问题,提出一种基于颜色分割和改进ELM的花生种子筛选算法.根据花生图像的聚类特性,采用限定RGB和HSV颜色空间中颜色范围的方法对花生图像进行颜色分割,获取花生种子图像目标区域.采用颜色、形状、改进HU矩特征对花生图像进行描述,结合改进HU矩平移、旋转和缩放不变性,对提取到的花生图像特征进行数量扩充,获得花生图像数据集.采用黄金分割法,确定隐含层神经元个数.引入正则化参数,提高ELM算法隐含层神经元与输出层之间连接权值矩阵的稳定性;采用PSO算法,获取最优输入权值和隐含层神经元阈值,构建PSO-RELM算法模型,并与BP、ELM、RELM算法进行比较.试验结果表明,PSO-RELM算法不仅对完好花生有很高的识别准确率(100%),还对破损花生也有很高的识别准确率(96.71%),平均测试时间为0.006 8 s,均方根误差为0.052 0,决定系数达0.987 4,能够满足花生种子筛选的实时性要求.
Abstract
Aiming at the problems of large workload and low efficiency in the process of peanut seeds manual selection,a peanut seed selection algorithm based on color segmentation and improved ELM was put forward.Based on clustering characteristics of peanut image,peanut image was segmented by limiting the color range in RGB and HSV color space,to obtain the peanut seeds of image target area.The feature of peanut image was described by using color,shape,improvement of HU moment,based on improved HU moment translation,rotation and scaling invariance,to expand the image characteristics of quantity and get the peanut image data sets.Golden section method was used to select number of hidden layer neurons quickly.Introducing the regularization parameter,the weights matrix stability was improved of the ELM algorithm between neurons in hidden layer and output layer connection.Using PSO algorithm to obtain the optimal input weights and threshold of the hidden layer neurons,on the basis,the PSO-RELM algorithm model was set up,and comparing with BP,ELM,RELM algorithm.Experimental results showed that PSO-RELM had a high recognition accuracy,not only for the intact peanut(100%),but also for the damage peanut(96.71%),average test time was 0.006 8 s,root mean square error was 0.052 0,determination coefficient was 0.987 4,which can meet the real-time requirements for peanut seeds.
关键词
花生种子筛选/颜色分割/极限学习机/正则化参数/粒子群算法Key words
peanut seeds selection/color segmentation/extreme learning machine/regularization parameter/particle swarm optimization引用本文复制引用
基金项目
河南省重点研发与推广专项(科技攻关)(232102211022)
平顶山学院青年基金项目(PXY—QNJJ-2019007)
出版年
2024
国家科技期刊平台
NSTL
万方数据