基于荧光光谱信息的绿色植物探测研究

Detecting Green Plants Based on Fluorescence Spectroscopy

针对农作物病、虫、草害化学防治时对靶变量施药以减少农药使用量、提高农药利用率的需求,本文研究了基于荧光光谱信息和主动光源方法在不同环境下探测绿色植物的方法。通过白色、蓝色和红色LED主动光源照射样本,采集了白天室内自然光照、白天太阳直射、白天无太阳直射和夜晚黑暗环境四种场景下的绿色植物和非绿色植物样本光谱。首先基于多波段光谱信息建立簇类独立软模式法(SIMCA)和线性判别分析(LDA)模型,验证利用主动光源照射下绿色植物荧光光谱探测绿色植物的可行性。试验结果表明,白色、蓝色和红色三种LED光源照射下SIMCA模型对预测集样本的识别率均达到92%以上,拒绝率均为100%;三种光源照射下LDA分类模型均能准确识别出预测集所有样本,检测效果优于SIMCA模型,且三种LED光源的效果无显著差异。为开发低成本绿色植物探测传感器,建立了绿色植物与非绿色植物样本分类目标函数,通过粒子群算法(PSO)优选单一连续光谱波段原始光谱并建立了绿色植物和非绿色植物样本的阈值分类模型。结果表明,白色、蓝色和红色LED光源照射下优选的原始光谱波段分别为731.1,730.76和731.1 nm,对应阈值分类模型分类预测集样本的F1-score分别为76.71%,80.52%和78.48%,蓝色LED光源的效果最好。该研究优选的主动光源类型和连续检测波段可为开发基于单波段的低成本绿色植物探测传感器提供理论依据。

Site-specific variable spraying is an effective approach to reducing pesticide use and improving the use efficiency for crop protection against disease,pests and weeds through chemical spraying,and target detection is a key procedure for site-specific variable spraying.Active illumination was adopted to detect green plant targets(crops and weeds),and the fluorescence spectral information of targets was analyzed.White,blue and red LEDs were utilized for illumination,and the spectra of green plants and others were collected in four circumstances,i.e.,day-indoor,day-under sunshine,day-shadow,and night-dark environment.Classification models were built based on multi-wavebands spectral features using soft independent modeling of class analogy(SIMCA)and linear discriminant analysis(LDA)methods.Results showed that with the illumination of the three types of LEDs,the recognition rates for the prediction dataset using SIMCA models were all above 92%,and corresponding rejection rates were all 100%.The LDA models could predict all samples with 100%accuracy,performing better than SIMCA models.And the difference in the effect of the three types of LEDs was indistinguishable.-The objective function for classifying green plants and others was proposed,and the particle swarm optimization(PSO)method was used to select the optimal single waveband.The optimal waveband for the three types of LEDs(white,blue and red)was 731.1,730.76 and 731.1 nm,respectively,and corresponding thresholding classification models were established.Results showed that the classification F1-scores for the three classification models were 76.71%,80.52%and 78.48%,respectively.Under complex circumstances,the blue LED provided the best illumination for greed plant detection.The selected blue LED light source and optimal waveband are valuable for developing low-cost green plant sensors.