English

中国农机化学报

中国农机化学报

中国农机化学报 ›› 2023, Vol. 44 ›› Issue (12): 137-142.DOI: 10.13733/j.jcam.issn.2095-5553.2023.12.021

• 农业信息化工程 • 上一篇    下一篇

基于光谱信息和支持向量机的绿色植物检测方法研究

徐敏雅1，朱路生1，刘永华1，王苗林2，王慧1   

  1. 针对精准农业中靶变量施药时对植物靶标探测的实际需求，研究红色LED、蓝色LED和卤钨灯照射下基于光谱信息和支持向量机的绿色植物检测方法。分别在室外阳光直射、室外阴影和室内黑暗环境下采集三种光源照射下绿色植物样本和非绿色植物样本的反射光谱。研究常数1和标准差倒数1/SDev两种变量权重对支持向量机SVM模型精确度的影响，结果表明两种变量权重对线性核SVM模型的影响不大，但径向基函数RBF核模型在变量权重为1时效果较差，最低精确度只有5185%，变量权重为1/SDev时所有RBF核SVM模型精确度较为正常，最低精确度为9506%。之后建立变量权重为1/SDev的绿色植物检测SVM模型。结果表明，三种光源对于线性核SVM模型的性能影响较小，所有线性核SVM模型的F1-score均超过9900%，其中卤钨灯照射下建立的SVM模型精确度达到10000%，蓝色LED照射下建立的SVM模型F1-score最高，达到9979%；RBF核SVM模型中效果最好的为蓝色LED照射下建立的模型，训练集和测试集F1-score分别为9959%和9917%。本研究结果可为开发基于主动光源的绿色植物探测传感器提供理论依据。
  • 出版日期:2023-12-15 发布日期:2024-01-16
  • 基金资助:
    江苏省高校优秀科技创新团队项目(2020kj069)；江苏农林职业技术学院科技项目(2021kj62、2019kj003)

Research on green plant detection methods based on spectral information and support vector machine

Xu Minya1, Zhu Lusheng1, Liu Yonghua1, Wang Miaolin2, Wang Hui1   

  1. Aiming at the actual demand for plant target detection during target variable pesticide application in precision agriculture, a green plant detection method based on spectral information and support vector machines was studied under the illumination of red LED, blue LED, and halogen tungsten lamp. The reflectance spectra of green and non green plant samples irradiated by three light sources were collected under outdoor direct sunlight, outdoor shadow, and indoor dark environments. The effects of constant 1 and 1/SDev variable weights on the accuracy of support vector machine SVM models were studied. The results showed that the two variable weights had little impact on the linear kernel SVM model, but the RBF kernel model had a poor effect when the variable weight was 1, with a minimum accuracy of only 51.85%. When the variable weight was 1/SDev, all RBF kernel SVM models had a relatively normal accuracy, with a minimum accuracy of 95.06%. After that, a SVM model for green plant detection with variable weight of 1/SDev was established. The results showed that the three light sources had a small impact on the performance of linear kernel SVM models, and the F1-score of all linear kernel SVM models exceeded 99.00%. The accuracy of the SVM model established under halogen lamp irradiation reached 100.00%, while the F1-score of the SVM model established under blue LED irradiation reached the highest, by 99.79%. The RBF core SVM model with the best effect was the model established under blue LED illumination, and the training set and test set F1-score were 9959% and 99.17%, respectively. The results of this study can provide a theoretical basis for the development of green plant detection sensors based on active light sources.
  • Online:2023-12-15 Published:2024-01-16

PDF

摘要： 支持向量机；绿色植物；靶标探测；精准农业；光谱信息

Abstract: support vector machine; green plants; target detection; precision agriculture; spectral information

中图分类号: 

版权所有 © 2021《中国农机化学报 》编辑部

地址：南京市玄武区中山门外柳营100号 邮编： Tel: 025-84346270，84346296