-
用10 μm的荧光微球测试,当颗粒穿过散射体积时,硅光电倍增管将光信号转换为电压信号。提取一段100 ms的时间序列信号,如图2(a)所示。由此可以看出,偏振光散射和荧光同时出现,并且都是一串时域脉冲。颗粒经过散射体积时产生脉冲信号;没有颗粒经过散射体积时,此时背景信号远低于颗粒产生的脉冲信号;一个脉冲信号与一个颗粒对应。偏振光散射与荧光的强度之比的差异是由于荧光微球表面吸附的荧光物质分布不均匀造成的。在对脉冲进行筛选之前,先对信号低通滤波。滤波后的信号如图2(b)所示,背景噪声在30 mV附近波动。为了获得高信噪比的信号,选择50 mV作为阈值电压筛选脉冲。虚线是无效信号,实线是提取出的有效脉冲。
-
斯托克斯矢量S用于描述颗粒散射光的偏振态,可表示为S = [I, Q, U, V]T, I代表光强,Q、U和V都是偏振分量,其中Q表示光的水平线偏振分量与竖直线偏振分量的强度差,U表示光的45°线偏振分量与−45°线偏振分量的强度差,V表示光的右旋圆偏振分量与左旋圆偏振分量的强度差。该矢量能描述完全偏振光、部分偏振光和自然光。对光强归一化处理,可得到取值范围为−1~1、无量纲的偏振参数q、u、v如公式(1)所示:
$$ q\equiv \frac{Q}{I},u\equiv \frac{U}{I},v\equiv \frac{V}{I} $$ (1) 微藻自发辐射产生的荧光光谱与它内部色素的种类和含量有关。硅光电倍增管测得的荧光强度Flu是光电阴极光谱灵敏度$ V\left(\lambda \right) $和辐射能通量$ \varPsi \left(\lambda \right) $乘积对波长的积分,如公式(2)所示:
$$ Flu={\int }_{460}^{950}V\left(\lambda \right)\varPsi \left(\lambda \right){\rm{d}}\lambda $$ (2) Pearson相关系数常用于描述两个变量$ {\beta }_{i} $和$ {\beta }_{j} $的相关性,该方法使用协方差对它们之间的相关强度进行评估。其表达式如公式(3)所示:
$$ R\left({\beta }_{i},{\beta }_{j}\right)=\frac{{\rm{cov}}\left({\beta }_{i},{\beta }_{j}\right)}{\sqrt{{\rm{va}}r\left({\beta }_{i}\right)\times {\rm{var}}\left({\beta }_{j}\right)}} $$ (3) 式中:${\rm{cov}}({\beta }_{i},{\beta }_{j})$是两个变量的协方差;${\rm{var}}\left({\beta }_{i}\right)$与${\rm{var}}\left({\beta }_{j}\right)$为变量的方差;当系数$ R\left({\beta }_{i},{\beta }_{j}\right) $为0~0.2、0.2~0.4、0.4~0.6、0.6~0.8和0.8~1.0时,两个变量相关性分别为不相关、弱相关、中度相关、强相关和极强相关。
-
支持向量机 (Support Vector Machine, SVM) 作为一种监督式学习的方法被广泛应用在统计分类以及回归分析中。在数据集中,每个样本用特征向量$ {x}_{t} $表征它的属性,$ {y}_{t} $是样本的标签($ t=\mathrm{1,2},\cdots ,n $,$ n $为总样本数)。SVM分类的基本思想是在样本空间中找出最优分类超平面$ w\cdot {x}_{t}+b $将不同类别的样本分开[18],w和b分别是超平面法向量和位移项。将数据集按类别分成N个部分,每次将一个类别样本作为正例$ ({y}_{t}=1) $;其余N−1个类别样本作为反例$ ({y}_{t}=-1) $,训练一对其余(OvR)分类器。通过优化参数w、b,使得公式(4)的值最小,此时的分类效果最佳。
$$ \begin{split} &\underset{w,b}{\mathrm{m}\mathrm{i}\mathrm{n}}\;\; {||w||}^{2}+C\sum _{t=1}^{n}{\xi }_{t}\\ &{{\rm{s.t.}}\;\;{\xi }_{t}\geqslant 0,y}_{t}\left(w\cdot {x}_{t}+b\right)\geqslant 1-{\xi }_{t} \end{split} $$ (4) 式中:引入松弛因子$ {\xi }_{t} $,其目的是为了允许部分样本被错分;$ C $=1.001,是惩罚因子。文中基于Windows 10 (64位)操作系统, 配备第九代智能英特尔酷睿I5-9300 HF处理器和NVIDIA GeForce GTX1650独立显卡的电脑上,使用 matlab R2021a版本中分类学习器的线性SVM分类模型。
混淆矩阵常用于衡量分类器分类的准确程度。对于N分类任务,混淆矩阵为$ N\times N $的矩阵,其列代表预测类别,行代表真实归属类别。混淆矩阵对角线的值表示分类器对该类别预测正确的个数与实际个数的占比,越接近100%分类效果越好。除此之外,准确率、精确率和召回率也是衡量分类效果指标[19]。
-
沉积物、微塑料、微藻是水体中常见的颗粒,也是水生态监测领域重点研究对象。因此把水中的颗粒分成这三类,并在实验室准备三类标准样品。二氧化硅是沉积物的主要成分[20],选择来自天津市倍思乐色谱技术开发中心的二氧化硅微球,以及500目的石英砂作为沉积物样品。聚苯乙烯(PS)、聚乙烯(PE)、聚对苯二甲酸乙二醇酯(PET)、聚丙烯(PP)、尼龙(PA)和聚氯乙烯(PVC)是常见的6种塑料材质,选用了以上6种材料代表微塑料。不同粒径的聚苯乙烯微球(0.5、1、2、3、4、5、6、7、8、9、10 μm)均来自苏州纳微科技有限公司;其余5种来自东莞华创塑胶科技有限公司,粒径均为500目。微藻的标准样品包括小球藻、斜生四链藻、菱形藻、三角褐指藻、铜绿微囊藻、蛋白核小球藻、球等鞭金藻等实验室培养藻悬浮液,它们均来源于中国科学院水生生物研究所淡水藻种库。微藻的培养环境为光照和黑暗12 h:12 h循环,光照度为2 000 lx。
SPC和水质多参数监测仪WQA (Xylem EXO1 Sonde, Xylem Water Solutions Ltd)被布放在珠江口黄茅海崖门水道现场测试(经纬度113°04′31.28″E,22°18′45.39″N),观测时间为2022年3月12日14:00至3月13日15:00,共25 h。图3(a)红色五角星处为布放点,图3(b)是用于收集水样的采水器和WQA,图3(c)显示了实验员在船上操作仪器。
-
考虑到真实的水体环境中悬浮颗粒种类的多样性和复杂性,在海试之前需要预先建立数据集并构建分类器。每个颗粒对应一个脉冲信号,对脉冲取平均值计算颗粒的特征向量 [I, q, u, v, Flu, Len2]T,前4个分量是偏振特征,Flu是荧光特征。Len是脉冲的持续时间,主要受颗粒粒径的影响,Len2定义为该颗粒“等效时间截面”。考虑到真实水体颗粒与实验室中使用的颗粒可能存在粒径差异,仅使用 [I, q, u, v, Flu]T作为颗粒的特征向量。
对于沉积物、微塑料、微藻样品做数据集,每一类都有3万个数据,使用60%的数据作为学习和调整分类器参数的训练集,20%的数据是用作验证集,以验证分类器训练过程中是否存在过拟合或欠拟合问题,另外20%的数据用于做出无偏性估计,SVM分类器的混淆矩阵如图4所示。
沉积物、微塑料和微藻预测的正确率分别为95.3%、93.3%和97.9%,表明该分类器对它们的分类效果好。这些颗粒的光学性质(如折射率)有区别,偏振对折射率敏感,所以能用偏振特征区分它们[12]。除了折射率的差异,有些微藻在蓝光(445 nm)的激发下还会自发荧光,特异性更强,所以微藻相对其余两者预测的正确率最高。
表1给出了不同样品的准确率、精确率和召回率。其中平均准确率高达97%,说明模型对沉积物、微塑料和微藻的分类效果很好,能够有效地识别出近乎所有正例,同时避免将反例错误地归类为正例。平均精确率和平均召回率均达到了95.5%,这说明模型具有很强的特征提取能力。
表 1 每个类别的准确率、精确率和召回率
Table 1. Accuracy, precision and recall of each category
Sediment Microplastic Microalgac Mean Accurancy 96.2% 96.1% 98.7% 97% Precision 93.3% 94.9% 98.3% 95.5% Recall 95.3% 93.3% 97.9% 95.5% -
通过采水器采集河道口上层(1~2 m)、中层(4~5 m)、下层(7~8 m)的水样。具体操作上,把采水器和WQA绑定在一起,通过船用绞车把它们投放到下层,打开第一批采水器;然后拉到中层处再打开另一批采水器,之后拉到船上;同时,用便携式采水器取上层的水样。每隔1 h,重复上述操作。WQA有深度、叶绿素a、藻红蛋白、浊度等探头,能在不同水深记录上述水质参数。采集的水样提供给SPC进行现场测量和分类作业。SPC采用45°线偏振光照射水样,每份水样测量5 min,给出颗粒数量($ {N}_{5} $)。之后,SVM分类器现场把颗粒分类,给出每种颗粒的数量,即沉积物的数量($ {N}_{5 s} $)、微塑料的数量($ {N}_{5 p} $)和微藻的数量($ {N}_{5 a} $)。
SPC对水中悬浮颗粒分类结果如图5所示。图5(a)显示了三类颗粒在不同水层中25个小时内的数量总和$ {N}_{25} $($ \equiv {\displaystyle\sum }_{i=1}^{25}{N}_{5}^{i} $, 其中$ {N}_{5}^{i} $是第i小时的颗粒数),横坐标代表不同水层,柱状图标注的百分比为同一水层不同类颗粒的比例。可以看出,$ {N}_{25} $随着水层深度增加而增加;沉积物是水体中主要的悬浮颗粒,微塑料次之,而微藻最少。同时,沉积物的数量和比例随深度的增加而增加,而微塑料和微藻的数量和比例随深度的增加而减少。
图5(b)~(d)显示$ {N}_{5 s} $、$ {N}_{5 p} $、$ {N}_{5 a} $随时间和水层的变化规律。$ {N}_{5 s} $在上层的值和波动幅度较中层和下层要小;它们都在12日17:00-18:00达到了最小值,在13日2:00-3:00出现最大值。中层和下层水样,$ {N}_{5 s} $在12日15:00和13日12:00处于高位,在13日06:00处于低位。上层$ {N}_{5 s} $在13日12:00也处于高位,13日00:00和09:00处于低位。总体而言,上层$ {N}_{5 s} $随时间变化与中层和下层有明显不同,这充分说明沉积物在水中的分层现象。
图 5 崖门水道水中悬浮颗粒的分类结果。(a) 颗粒在不同水层的数量和比例;(b)~(d)不同水层处$ {N}_{5 s} $、$ {N}_{5 p} $和$ {N}_{5 a} $随时间变化曲线
Figure 5. Classification results of suspended particles in water of Yamen Waterway by the prototype. (a) Number and proportion of particles in different water layers; (b)-(d) $ {N}_{5 s} $, $ {N}_{5 p} $ and $ {N}_{5 a} $ varied with time in different water layers
$ {N}_{5 p} $约在12日22:00-13日00:00、13日5:00-6:00出现极大值,在12日16:00-18:00、13日03:00和12:00值最低。上层$ {N}_{5 p} $几乎所有时刻比中层和下层的大,它在13日00:00出现最大值,并在13日03:00后出现大幅度波动且下降的趋势。中层和下层的$ {N}_{5 p} $整体趋势相同,它们处于低位的时间点都相同;但处于高位的时间点略微偏移。这说明漂浮在上层的微塑料数量最多,它们的数量随深度的增加而减少。
$ {N}_{5 a} $在12日14:00-18:00呈下降趋势,接着不断上升,13日2:00到达顶峰,最后缓慢下降。上层$ {N}_{5 a} $在13日09:00处于高位,在13日07:00和11:00处于低位。中层和下层$ {N}_{5 a} $的幅值和走向大致相同,它们处于高位的时刻有一两个小时的时间差。在13日02:00-03:00,中层和下层的$ {N}_{5 a} $大幅度下降,而上层的$ {N}_{5 a} $几乎不变。大部分时间段,微藻在上层的数量居多。
-
SPC与WQA几乎同时获得水中颗粒物和水质信息,将部分参数进行了关联分析。将在相同水层下所测数据绘制在图6,曲线的颜色与纵坐标轴颜色对应。SPC与WQA的测量值的Pearson相关系数如表2所示。下文结合图6和表2对结果分析。
图 6 SPC与WQA的数据对比。 (a)~(c)不同水层$ {N}_{5 a} $、$ Chla $和$ Pe $随时间的变化;(d)~(f) 不同水层$ {Len}_{5 s}^{2} $和$ T $随时间的变化
Figure 6. Comparison between SPC data and WQA data. (a)-(c) $ {N}_{5 a} $, $ Chla $ and $ Pe $ varied with time in different water layers; (d)-(f) $ {Len}_{5 s}^{2} $ and $ T $ varied with time in different water layers
表 2 SPC和WQA数据的Pearson相关系数
Table 2. Pearson correlation coefficient of SPC data and WQA data
Variables Surface Middle Bottom N5a and Chla 0.809 4 0.806 4 0.599 9 N5a and Pe 0.839 1 0.806 9 0.582 5 $ {Len}_{5 s}^{2} $ and T 0.044 1 0.946 0 0.964 7 在图6(a)~(c)分别显示了上层、中层和下层水样中微藻的数量$ {N}_{5 a} $、WQA测量的叶绿素a浓度($ Chla $)和藻红蛋白浓度($ Pe $)随时间变化的曲线,实线为均值而阴影部分为标准差。在上层和中层的水样中,$ {N}_{5 a} $随时间变化的趋势与$ Chla $、$ Pe $随时间变化的趋势相似,都是极强相关。在13日02:00至08:00时间内,三者的值都比较大;在12日16:00至18:00时段,三者值都比较小。从12日18:00到13日02:00时间,三者都在上升阶段,而在13日06:00之后,三者都下降。在某些细节上,比如在上层13日02:00时刻,三者都同时到达局部极大值。除此之外,$ {N}_{5 a} $也有与$ Chla\mathrm{和}Pe $不一致的地方。比如,在12日23:00,$ {N}_{5 a} $有个局部极大值,$ Chla\mathrm{和}Pe $在相应虽然也有变化,但不如$ {N}_{5 a} $明显。三者值在12日16:00 至 18:00 时间内不是同时到达较小值;在13日06:00后$ Chla\mathrm{和}Pe $持续下降,而$ {N}_{5 a} $是波动下降。总的来说,对于上层和中层水样,三者的相关性比较强,表2中的Pearson相关系数都大于0.8。
相对而言,下层水样的相关性较上层和中层差一些,为中度相关,从表2可知,$ {N}_{5 a} $与$ Chla\mathrm{和}Pe $的相关系数都小于0.6。从趋势上来看,三者与上层和中层数据类似,都有刚开始较低,中间变大,而后变小的过程。但相对而言,$ {N}_{5 a} $稍稍偏离这种趋势。具体而言,在下层$ Chla\mathrm{和}Pe $的差异较小,在13日03:00到达极大值,而此时,$ {N}_{5 a} $是极小值。在12日14:00至16:00时段,$ {N}_{5 a} $较大,$ Chla $和 $ Pe $值较小;类似地,在13日05:00时刻,$ {N}_{5 a} $有一个极大值,$ Chla \mathrm{和}Pe $值都较小。
笔者还观察了沉积物这类颗粒的特性。经过分析发现,沉积物的“等效时间截面”的均值($ {Len}_{5 s}^{2} $)与WQA给出的浊度值($ T $)的相关性值得关注。如图6(d)~(f)所示,实线为均值而阴影部分为标准差。从12日16:00至18:00时间段,$ T $出现了一个分布峰,而在13日03:00出现了最大值;在13日12:00出现了极大值。同时可以看出,随着深度增加,$ T $值整体增加。与之对应的$ {Len}_{5 s}^{2} $的值也随着深度增加。并且,在中层和下层水样中与$ T $类似,12日16:00至18:00时间段,出现了一个分布峰,而在13日03:00出现了最大值,在13日12:00出现了极大值。而在上层水中,虽然$ {Len}_{5 s}^{2} $在12日18:00出现了极大值,但其他时刻与$ T $并不相似。如表2所示,对于中层和下层的水样,$ {Len}_{5 s}^{2} $与$ T $的相关系数都大于0.9,这意味着两者的变化趋势非常一致;而上层的数据,相关系数小于0.1,说明几乎没有关联。从SPC提供的颗粒参数与WQA提供的水质参数之间的关联分析,可以看出,它们之间具有一定的相关性,这些具有必然性,因为它们反映的都是水体中颗粒的性质。
Field prototype for rapid classification of suspended particles in water based on polarized light scattering and fluorescence measurement
-
摘要: 水中悬浮颗粒是水体物质的重要成分,因此监测它们的种类和浓度对研究和保护水生态系统具有重要科学意义和实用价值。文中研制了一种水中悬浮颗粒分类仪(Suspended Particle Classifier, SPC),旨在现场检测野外采集的水样,快速得出水中悬浮颗粒的种类、数量和比例。SPC采用引流管将颗粒输送至散射体积内,通过同时探测单个颗粒的散射光偏振态和荧光信号,结合机器学习算法对颗粒分类。对沉积物、微塑料和微藻的标准样品做了数据集并训练分类器,SPC能以大于95%的预测正确率对它们进行分类。接着,将SPC和商业水质多参数监测仪(QWA)同时在崖门水道连续布放25个小时。SPC能快速测量现场采集的水样,获取不同水层的沉积物、微塑料和微藻的数量随时间的变化情况。SPC给出的微藻数量与QWA测得的叶绿素a浓度以及藻红蛋白浓度之间存在显著的相关性;此外,SPC给出的沉积物等效时间截面和QWA测得的浊度值也呈现出明显的相关性,由此可以证明SPC分类结果的可靠性。结果表明:SPC能够对水中的悬浮颗粒进行现场快速分类检测,有望成为探索水生态系统的关键技术。Abstract:
Objective Suspended particles in water include solid or liquid particles, such as sediment, microplastics, and microalgae. Accurate monitoring of their categories and concentration is of great scientific and practical significance for studying and protecting aquatic ecosystems. Various optical instruments have been developed to probe suspended particles in water, which can be divided into two categories based on the measurement methods. One category measures the overall characteristics of all particles in a body of water, while the other measures individual particles. Water Quality Analyzer (QWA) provide estimates of particle concentration and size distribution, chlorophyll-a concentration, and other water quality parameters. However, QWA are limited in their ability to accurately identify the categories of suspended particles in water. Underwater flow cytometry enables the characterization of various categories of particles by breaking up a water sample into individual particles that are then to be measured. However, this technique is expensive and requires complex sample pretreatment, which limits its application. Therefore, it is needed to develop a prototype for field detection of water samples collected in the wild, with the goal of quickly determining the categories, numbers, and proportions of suspended particles in water. Methods Suspended Particle Classifier (SPC) has been developed in this paper and its diagram is depicted (Fig.1). The SPC employs a 445 nm laser as the excitation source to induce chlorophyll fluorescence, and the polarization state of the laser is modulated by a polarization state generator. The SPC obtains individual particle polarized light scattering and fluorescence signals, which are combined with a Support Vector Machine (SVM) to classify particles based on their optical properties. To ensure its suitability for field use, the SPC is equipped with a drainage tube for the transportation of water samples and an industrial computer for instrument control and data analysis. Standard samples of sediments, microplastics, and microalgae are collected. Then, datasets are created to train the SVM classifier. Subsequently, SPC was deployed alongside QWA in the Yamen Waterway for 25 hours to evaluate its performance (Fig.3). The accuracy of the SPC classification was verified using data obtained from the QWA. Results and Discussions The SPC's classification accuracy for standard samples of sediment, microplastics, and microalgae was found to be 95.3%, 93.3%, and 97.9% (Fig.4), respectively, indicating that the classifier has good performance in classifying these particles. The average accuracy and recall rate were found to be 95.5% (Tab.1), indicating the SVM model has strong feature extraction ability. These results suggest that the SPC can accurately classify standard samples. When applied in the Yamen Waterway, the SPC was able to rapidly measure water samples collected in the field and track the changes in the number of sediments, microplastic, and microalgae in different water layers over time (Fig.5). Furthermore, the number of microalgae identified by the SPC was found to have a strong correlation with the concentration of chlorophyll-a and phycoerythrin measured by the QWA (Fig.6, Tab.2). Additionally, the so-called effective time cross-section of sediments identified by the SPC was found to have a strong correlation with the turbidity value measured by the QWA (Fig.6, Tab.2), further validating the reliability of the SPC's classification results. Conclusions In this study, a suspended particle classifier was developed with the aim of classifying and counting suspended particles in water samples collected in the field. The SPC probes polarized light scattering and fluorescence signals from individual suspended particles and uses SVM to classify them based on their optical properties. The classification accuracy for standard samples of sediment, microplastics, and microalgae was over 95%. To validate the SPC's classification ability for field water samples, the SPC and QWA were deployed in the Yamen Waterway for 25 hours of synchronous testing. The SPC was able to track changes in the number of sediment, microplastic, and microalgae in different water layers over time. There was a strong correlation between the SPC and QWA measurement data, indicating the high reliability of the SPC in classifying particles in field water samples. These results demonstrate that the SPC can rapidly detect and classify suspended particles in water and has the potential to be a valuable tool for exploring aquatic ecosystems. -
Key words:
- polarized light scattering /
- fluorescence /
- suspended particles /
- field /
- rapid classification
-
图 5 崖门水道水中悬浮颗粒的分类结果。(a) 颗粒在不同水层的数量和比例;(b)~(d)不同水层处$ {N}_{5 s} $、$ {N}_{5 p} $和$ {N}_{5 a} $随时间变化曲线
Figure 5. Classification results of suspended particles in water of Yamen Waterway by the prototype. (a) Number and proportion of particles in different water layers; (b)-(d) $ {N}_{5 s} $, $ {N}_{5 p} $ and $ {N}_{5 a} $ varied with time in different water layers
图 6 SPC与WQA的数据对比。 (a)~(c)不同水层$ {N}_{5 a} $、$ Chla $和$ Pe $随时间的变化;(d)~(f) 不同水层$ {Len}_{5 s}^{2} $和$ T $随时间的变化
Figure 6. Comparison between SPC data and WQA data. (a)-(c) $ {N}_{5 a} $, $ Chla $ and $ Pe $ varied with time in different water layers; (d)-(f) $ {Len}_{5 s}^{2} $ and $ T $ varied with time in different water layers
表 1 每个类别的准确率、精确率和召回率
Table 1. Accuracy, precision and recall of each category
Sediment Microplastic Microalgac Mean Accurancy 96.2% 96.1% 98.7% 97% Precision 93.3% 94.9% 98.3% 95.5% Recall 95.3% 93.3% 97.9% 95.5% 表 2 SPC和WQA数据的Pearson相关系数
Table 2. Pearson correlation coefficient of SPC data and WQA data
Variables Surface Middle Bottom N5a and Chla 0.809 4 0.806 4 0.599 9 N5a and Pe 0.839 1 0.806 9 0.582 5 $ {Len}_{5 s}^{2} $ and T 0.044 1 0.946 0 0.964 7 -
[1] Chen Y, Wang H, Liao R, et al. Rapidly measuring scattered polarization parameters of the individual suspended particle with continuously large angular range [J]. Biosensors, 2022, 12(5): 321-336. doi: 10.3390/bios12050321 [2] Zhang Kaidi, Li Anchun, Zhang Jin, et al. Seasonal variations in the surficial sediment grain size in the East China Sea continental shelf and their implications for terrigenous sediment transport [J]. Journal of Oceanography, 2019, 76(1): 1-14. doi: 10.1007/s10872-019-00523-8 [3] Tsang Y Y, Mak C W, Liebich C, et al. Spatial and temporal variations of coastal microplastic pollution in Hong Kong [J]. Marine Pollution Bulletin, 2020, 161: 111765-111770. doi: 10.1016/j.marpolbul.2020.111765 [4] 刘洋, 钟年丙, 陈明, 等. 准确测量微藻生物量的塑料光纤倏逝波传感器[J]. 光学精密工程, 2015, 23(10 z): 151-158. doi: 10.3788/OPE.20152313.0150. Liu Yang, Zhong Nianbing, Chen Ming, et al. Plastic optical fiber evanescent wave sensor for accurate measurement of microalgae biomass [J]. Optics and Precision Engineering, 2015, 23(10z): 151-158. (in Chinese) doi: 10.3788/OPE.20152313.0150 [5] 张东彦, 尹勋, 佘宝, 等. 多源卫星遥感数据监测巢湖蓝藻水华爆发研究[J]. 红外与激光工程, 2019, 48(7): 726004-0726004(12). doi: 10.3788/IRLA201948.0726004. Zhang Dongyan, Yin Xun, She Bao, et al. Using multi-source satellite imagery data to monitor cyanobacterial blooms of ChaohuLake [J]. Infrared and Laser Engineering, 2019, 48(7): 0726004. (in Chinese) doi: 10.3788/IRLA201948.0726004 [6] Tao Jing, Hill P S, Boss E S, et al. Variability of suspended particle properties using optical measurements within the columbia river estuary [J]. Journal of Geophysical Research-Oceans, 2018, 123(9): 6296-6311. doi: 10.1029/2018JC014093 [7] Zhou J, Gilerson A, Ioannou I, et al. Retrieving quantum yield of sun-induced chlorophyll fluorescence near surface from hyperspectral in-situ measurement in productive water [J]. Optics Express, 2008, 16(22): 17468-17483. doi: 10.1364/OE.16.017468 [8] Felix D, Albayrak I, Boes R M. In-situ investigation on real-time suspended sediment measurement techniques: Turbidimetry, acoustic attenuation, laser diffraction (LISST) and vibrating tube densimetry [J]. International Journal of Sediment Research, 2018, 33(1): 3-17. doi: 10.1016/j.ijsrc.2017.11.003 [9] Méndez-barroso L A, Rivas-márquez J A, Sosa-tinoco I, et al. Design and implementation of a low-cost multiparameter probe to evaluate the temporal variations of water quality conditions on an estuarine lagoon system [J]. Environmental Monitoring and Assessment, 2020, 192(11): 710. doi: 10.1007/s10661-020-08677-5 [10] Olson R J, Shalapyonok A, Kalb D J, et al. Imaging FlowCytobot modified for high throughput by in‐line acoustic focusing of sample particles [J]. Limnology and Oceanography: Methods, 2017, 15(10): 867-874. doi: 10.1002/lom3.10205 [11] Dashkova V, Malashenkov D, Poulton N, et al. Imaging flow cytometry for phytoplankton analysis [J]. Methods, 2017, 112: 188-200. doi: 10.1016/j.ymeth.2016.05.007 [12] Wang Yong, Liao Ran, Dai Jincai, et al. Differentiation of suspended particles by polarized light scattering at 120 ° [J]. Optics Express, 2018, 26(17): 22419-22431. doi: 10.1364/OE.26.022419 [13] Li Da, Chen Fei, Zeng Nan, et al. Study on polarization scattering applied in aerosol recognition in the air [J]. Optics Express, 2019, 27(12): A581-595. doi: 10.1364/OE.27.00A581 [14] Zhuo Zepeng, Wang Hongjian, Liao Ran, et al. Machine learning powered microalgae classification by use of polarized light scattering data [J]. Applied Sciences, 2022, 12(7): 3422-3432. doi: 10.3390/app12073422 [15] Zhao Nanjing, Zhang Xiaoling, Yin Gaofang, et al. On-line analysis of algae in water by discrete three-dimensional fluorescence spectroscopy [J]. Optics Express, 2018, 26(6): A251-259. doi: 10.1364/oe.26.00a251 [16] Xiong Zhihang, Wang Hongjian, Li Jiajin, et al. Probing individual particles in aquatic suspensions by simultaneously measuring polarized light scattering and fluorescence [J]. Biosensors, 2021, 11(11): 416-425. doi: 10.3390/bios11110416 [17] Li Jiajin, Deng Hanbo, Guo Zhiming, et al. Convergence of illuminating beams suffering from scattering during the individual measurement of suspended particles [J]. Frontiers in Physics, 2022, 10: 975766-975775. doi: 10.3389/fphy.2022.975766 [18] 龙思源, 张葆, 宋策, 等. 基于改进的加速鲁棒特征的目标识别[J]. 中国光学, 2017, 10(6): 719-725. doi: 10.3788/CO.20171006.0719. Chang Chihchung, Lin Chihjen. LIBSVM: a library for support vector machines [J]. ACM Transactions on Intelligent Systems and Technology, 2011, 2(3): 1-27. (in Chinese) doi: 10.3788/CO.20171006.0719 [19] 张添福,钟舜聪,连超铭等.基于深度学习特征融合的视网膜图像分类[J].激光与光电子学进展,2020,57(24):266-273. doi: 10.3788/LOP57.241025. Zhang T, Zhong S, Lian C, et al. Deep learning feature fusion-based retina image classification [J]. Laser & Optoelectronics Progress, 2020, 57(24): 241025. (in Chinese) doi: 10.3788/LOP57.241025 [20] Liu Mingwei, Liu Xiao, Wang Weizhuo, et al. Effect of SiO2 and Al2O3 on characteristics of lightweight aggregate made from sewage sludge and river sediment [J]. Ceramics International, 2018, 44(4): 4313-4319. doi: 10.1016/j.ceramint.2017.12.022 [21] 王鹏程, 张肃, 申成彪, 等. 偏振光在椭球细粒子中多次散射传输特性[J]. 中国光学. 2023, 16(2): 1-10. doi: 10.37188/CO.2022-0144. Wang Pengcheng, Zhang Su, Shen Chengbiao, et al. Multiple scattering transmission characteristic of polarized light in ellipsoidal fine particles [J]. Chinese Optics, 2023, 16(2): 1-10. (in Chinese) doi: 10.37188/CO.2022-0144 [22] 陈都, 刘秉义, 杨倩, 等. 近180°水中悬浮颗粒物体积散射函数测量[J]. 红外与激光工程, 2021, 50(6): 20211029. doi: 10.3788/IRLA20211029. Chen Du, Liu Bingyi, Yang Qian, et al. Measurement of volume scattering function of suspended particles in water approaching to 180° [J]. Infrared and Laser Engineering, 2021, 50(6): 20211029. (in Chinese) doi: 10.3788/IRLA20211029 [23] Rybicki Marcus, Moldaenke Christian, Rinke Karsten, et al. WP-C: a step towards secured drinking water: development of an early warning system for lakes[C]//Chinese Water Systems, 2018. [24] Sadeghian A, Chapra S C, Hudson J, et al. Improving in-lake water quality modeling using variable chlorophyll a/algal biomass ratios [J]. Environmental Modelling & Software, 2018, 101: 73-85. doi: 10.1016/j.envsoft.2017.12.009 [25] Landers M N, Sturm T W. Hysteresis in suspended sediment to turbidity relations due to changing particle size distributions [J]. Water Resources Research, 2013, 49(9): 5487-5500. doi: 10.1002/wrcr.20394 [26] He Weipeng, Jun Nan. Study on the impact of particle size distribution on turbidity in water [J]. Desalination and Water Treatment, 2012, 41(1-3): 26-34. doi: 10.1080/19443994.2012.664675 [27] 叶陈军, 潘建明, 郑旻辉, 等. 浊度对活体叶绿素荧光检测的影响及修正方法[J]. 海洋学研究, 2018, 36(2): 27-35. doi: 0.3969/j.isn.1001-909 X.2018.02.004. Ye Chenjun, Pan Jianming, Zheng Minhui, et al. The influence of turbidity on the measurement of in vivo chlorophyll fluorescence and its correction method [J]. Journal of Marine Sciences, 2018, 36(2): 27-35. (in Chinese) doi: 10.3969/j.issn.1001-909X.2018.02.004