摘要
针对柑橘黄龙病检测模型的准确度较低、可靠性较差等问题,提出了一种压缩自编码融合极限学习机(Contractive auto-encoder combined extreme learning machine,CAE-ELM)的柑橘黄龙病鉴别方法。此方法通过ELM代替CAE顶层的Softmax分类器和反向微调阶段,达到减少算法运行时间同时提高模型的稳定性及鉴别能力的目的。其中,CAE实现了样本深层特征提取,ELM可实现分类鉴别。为了评估CAE-ELM模型性能,以不同比例的柑橘叶片近红外光谱数据作为训练集进行实验,采用波形叠加极限学习机(Summation wavelet extreme learning machine, SWELM)、ELM、支持向量机(Support vector machine, SVM)、堆叠去噪自编码(Stacked denoising auto-encoder, SDAE)、反向传播模型(Back propagation, BP)、CAE作为对比方法。在柑橘黄龙病的鉴别实验中,无论训练集样本大小,CAE-ELM均能保持最高的分类准确度,尤其当训练集与测试集为1080/165时分类准确度达100.00。同时,CAE-ELM模型比SDAE、CAE和BP模型具有更快的训练速度,但慢于SVM、ELM和SWELM模型。结果表明,CAE-ELM模型可以准确鉴别柑橘黄龙病,且模型具有良好的鲁棒性和可扩展性。
To solve the problems of low accuracy and poor reliability of the citrus Huanglongbing identification model, a method of Citrus Huanglongbing identification based on Contractive auto-encoder combined extreme learning machine (CAE-ELM) was proposed. To reduce the running time and improve the stability and classification ability of the model, extreme learning machine was used to replace the Softmax of Contractive auto-encoder top layer and reverse fine-tuning stage. Among them, the Contractive auto-encoding realized the extraction of deep features of the samples, and the extreme learning machine realized the identification. To evaluate the performance of CAE-ELM model, experiments were carried out using near infrared spectroscopy data of citrus leaves with different proportions as training sets. Meanwhile, summation wavelet extreme learning machine (SWELM), ELM, support vector machine (SVM), stacked denoising auto-encoder (SDAE), back propagation (BP) and CAE were used as comparative methods. In the identification experiment of citrus Huanglongbing, CAE-ELM maintained the highest classification accuracy no matter how the training sets change. Especially, when the ratio of training sets and test sets was 1080/165, the classification accuracy reached 100.00. At the same time, CAE-ELM model trained faster than SDAE, CAE and BP, but slower than SVM, ELM and SWELM. The results showed that the CAE-ELM model could be used to accurately identify the citrus Huanglongbing, with good robustness and scalability.
作者
路皓翔
徐明昌
张卫东
杨辉华
刘振丙
LU Hao-Xiang;XU Ming-Chang;ZHANG Wei-Dong;YANG Hui-Hua;LIU Zhen-Bing(School of Electronic Engineering and Automation, Guilin University of Electronic Technology, Guilin 541004, China;School of Computer and Information Security, Guilin University of Electronic Technology, Guilin 541004, China;School of Automation, Beijing University of Posts and Telecommunications, Beijing 100876, China)
出处
《分析化学》
SCIE
EI
CAS
CSCD
北大核心
2019年第5期652-660,共9页
Chinese Journal of Analytical Chemistry
基金
国家自然科学基金项目(Nos.21365008
61562013)
广西自动检测技术与仪器重点实验室主任基金(No.YQ18108)资助~~
关键词
压缩自编码
极限学习机
近红外光谱
柑橘黄龙病
Contractive auto-encoder
Extreme learning machine
Near-infrared spectroscopy
Citrus Huanglongbing
