摘要
为了提高可见-近红外(VNIR)高光谱花生图像分类的准确率和减少分类检测的运算时间,提出了基于连续投影算法(SPA)融合粒子群算法优化后向传播神经网络(PSO-BP)的分类检测模型。利用高光谱成像系统采集了7个花生品种样本的VNIR光谱数据,并进行了背景分割和光谱信息的提取,去除受噪声和杂散光影响大的波段后,运用Savitzky-Golay卷积平滑对400 nm~900 nm范围的波长进行预处理;采用SPA降维及均方根误差值选择了25个特征波长,同时利用PSO-BP神经网络的初始权重和阈值,构建PSO-BP模型作为分类器进行了实验,取得了测试集识别准确率为98.7%、kappa系数为0.98及遗漏误差为3的数据。结果表明,相较4个对比方法构建的分类模型,该模型的准确率分别提高了2.1%、8.6%、3.9%和4.3%。该方法在基于高光谱成像的花生品种分类技术中具有很好的应用前景,为花生品种的高精度、快速无损分类提供了新思路。
In order to improve the accuracy of visible-near infrared(VNIR)hyperspectral peanut image classification and to reduce the computing time of classification detection,a classification detection model based on successive projection algorithm(SPA)fused with particle swarm optimization back propagation(PSO-BP)neural network was proposed.A hyperspectral imaging system was used to acquire VNIR spectral data of seven peanut species samples and conducts background segmentation and extraction of spectral information.After removing the wavelengths that were highly affected by noise and stray light,the wavelengths in the range of 400 nm~900 nm were preprocessed by using Savitzky-Golay convolutional smoothing.The SPA was used to reduce the dimensionality,and 25 characteristic wavelengths were selected by virtue of the root mean square error values.The PSO was also used to optimize the initial weights and thresholds of the BP neural network,and the PSO-BP model was constructed as a classifier for the experiments,and a recognition accuracy of 98.7%,a kappa coefficient of 0.98,and a miss error of 3 for the test set were obtained.The results demonstrate that the accuracy of the model is improved by 2.1%,8.6%,3.9%,and 4.3%,respectively,compared with the classification models constructed by the four comparison methods.The proposed method has good application prospects in peanut variety classification based on hyperspectral imaging technology,and provides a new idea for high accuracy and fast nondestructive classification of peanut varieties.
作者
杨洋
徐熙平
薛航
张宁
张越
索科
YANG Yang;XU Xiping;XUE Hang;ZHANG Ning;ZHANG Yue;SUO Ke(Changchun University of Science and Technology,College of Optoelectronic Engineering,Changchun 130022,China;College of Electronic and Information Engineering,Beihua University,Jilin 132021,China)
出处
《激光技术》
CAS
CSCD
北大核心
2024年第4期556-564,共9页
Laser Technology
关键词
光谱学
图像分类
连续投影算法
粒子群算法
后向传播神经网络
花生
spectroscopy
image classification
successive projection algorithm
particle swarm optimization
back propagation neural network
peanut
