论文部分内容阅读
为了进行快速实时的杂草识别,研究了作物和杂草叶片的可见-近红外反射光谱特性。选择了两种常见的田间作物大豆(Glycine max)和玉米(Zea mays),以及铁苋菜(Acalypha australis L.)和田字草(Marsilea quadrifoliaL.)两种杂草作为研究对象,每种各30个样本,共120个样本。采用ASD Fieldspec便携式光谱仪进行光谱采集。在对400~1000 nm的光谱数据进行平滑和一阶求导预处理、。通过主成份分析,去除了一个奇异样本。最后用79个样本组成的建模集进行偏最小二乘法建模,对剩余的40个样本进行预测。预测模型结果的相关性达到0.986,识别率达到100%。说明研究中选用的作物和杂草叶片的可见-近红外反射光谱特性之间有较大的区别,可以用于进行杂草和作物的区分。
For fast and real-time weed recognition, the visible-near infrared reflectance spectra of crops and weed leaves were studied. Two common field crops, Glycine max and Zea mays, as well as Acalypha australis L. and Marsilea quadrifolia L.were selected as study objects, each of 30 Sample, a total of 120 samples. Spectral acquisition with ASD Fieldspec portable spectrometer. Smoothing and first-order derivative preprocessing of spectral data at 400-1000 nm is performed. By principal component analysis, a singular sample is removed. Finally, the model set composed of 79 samples was modeled by partial least squares, and the remaining 40 samples were predicted. The correlation of the predictive model results reached 0.986, and the recognition rate reached 100%. This shows that there is a big difference between the visible-near-infrared reflectance spectra of crops and weeds used in the study, which can be used to distinguish weed from crop.