基于无人机多光谱图像的水稻品种鉴定

Identification of rice varieties based on unmanned aerial vehicle(UAV)multispectral images

[目的]为解决传统人工水稻品种鉴定程序复杂的问题,建立一种快速鉴定水稻品种模型,利用多光谱成像技术对水稻叶色进行分析。[方法]分别利用线性拟合、多层感知器、径向基函数和随机森林4种机器学习算法模型,采用大疆精灵4多光谱版无人机航拍获得多光谱数据,对数据特征提取后得到的植被指数进行建模分析。通过输入不同组合的植被指数,筛选出准确率最高的变量组合。[结果]发现多层感知器模型对水稻品种的识别准确率可达94.87%,比线性拟合模型、径向基函数模型、随机森林模型的准确率分别高41.3%、8.55%、16.24%。将精度最高、拟合程度最好的多层感知器模型确定为最优模型。[结论]多层感知器分类模型对水稻的平均分类准确率最高,能够较好实现水稻品种鉴定的功能。

[Objectives]In order to solve the complex problem of traditional artificial rice variety identification procedure,a visual rice variety identification model was established by using multispectral imaging technology to analyze rice leaf color.[Methods]Four machine learning algorithms,linear fitting,multi-layer perceptron,radial basis function model and random forest model,were used to model and analyze the vegetation index obtained from the feature extraction of multispectral data,which were obtained from the aerial photography of Dajiang spirit 4 multispectral unmanned aerial vehicle(UAV).The variable combinations with the best accuracy would be applied to identify rice variety with different combinations of vegetation index.[Results]The study found that the accuracy of multi-layer perceptron model in identifying rice varieties reached to 94.87%.The accuracy was 41.3%,8.55%,and 16.24%better than that of linear fitting model,radial basis function model and random forest model,respectively.Therefore,the multilayer perceptron model with best accuracy and fitting degree was selected as the optimal model.[Conclusions]The rice variety identification model constructed in this study can accurately predict rice varieties by judging the leaf color of rice leaves,and provide reference value for the identification of rice varieties.