Recognition of multi-component compounds based on occurrence time of secondary crest in the fluorescence lifetime attenuation curve

Fluorescence lifetime is the main characteristic parameter of fluorescence. It is a widely used to draw fluorescence lifetime attenuation curves and to fit fluorescence lifetime parameters by using gated detection methods to identify the species of substances. However, the fluorescence attenuation of each fluorophore in a multi-component compound interferes with one another, affecting the accuracy of identification. In this paper, we propose a method to accurately identify substances by using the occurrence time of the secondary crest of the fluorescence lifetime attenuation curve based on the principle of gated detection to measure the fluorescence lifetime. Furthermore, we design a fluorescence lifetime imaging measurement system and select the same areas of interest in the images for analysis and comparison. The average lifetime of the fluorescence and the occurrence time of the secondary crest are considered as the characteristic parameters. We use five commercially available motor engine oils as the experimental samples and compare the recognition performance of different kernel functions based on a support vector machine (SVM). The radial basis kernel function presents the best performance in terms of recognition accuracy and speed. The recognition rates of the SVM model with the average fluorescence lifetime and the occurrence time of the secondary crest in the attenuation curve of the fluorescence lifetime as a feature vector are 76.24% and 74.65%, respectively. The recognition rate of the SVM model which combines them as feature vectors reaches 91.88%. The experimental results demonstrate that the occurrence time of the secondary crest in the attenuation curve of the fluorescence lifetime can be employed as the basis for substance identification in the analysis of the fluorescence characteristics of multi-component compounds, whose recognition accuracy is similar to the average fluorescence lifetime parameter. Moreover, the occurrence time of the secondary crest of the fluorescence lifetime attenuation curve can be implemented to identify multi-component compounds when it is used as a characteristic parameter.

Fluorescence lifetime is the main characteristic parameter of fluorescence.It is a widely used to draw fluorescence lifetime attenuation curves and to fit fluorescence lifetime parameters by using gated detection methods to identify the species of substances.However, the fluorescence attenuation of each fluorophore in a multi-component compound interferes with one another, affecting the accuracy of identification.In this paper, we propose a method to accurately identify substances by using the occurrence time of the secondary crest of the fluorescence lifetime attenuation curve based on the principle of gated detection to measure the fluorescence lifetime.Furthermore, we design a fluorescence lifetime imaging measurement system and select the same areas of interest in the images for analysis and comparison.The average lifetime of the fluorescence and the occurrence time of the secondary crest are considered as the characteristic parameters.We use five commercially available motor engine oils as the experimental samples and compare the recognition performance of different kernel functions based on a support vector machine (SVM).The radial basis kernel function presents the best performance in terms of recognition accuracy and speed.The recognition rates of the SVM model with the average fluorescence lifetime and the occurrence time of the secondary crest in the attenuation curve of the fluorescence lifetime as a feature vector are 76.24% and 74.65%, respectively.The recognition rate of the SVM model which combines them as feature vectors reaches 91.88%.The experimental results demonstrate that the occurrence time of the secondary crest in the attenuation curve of the fluorescence lifetime can be employed as the basis for substance identification in the analysis of the fluorescence characteristics of multi-component compounds, whose recognition accuracy is similar to the average fluorescence lifetime parameter.Moreover, the occurrence time of the secondary crest of the fluorescence

Introduction
Laser-induced fluorescence (LIF) is a type of detection technology, in which a laser is used to excite the emission of fluorescence of a substance; it is based on the discrimination of the fluorescence signal generated by the interaction between the laser pulse and the medium.Owing to its high sensitivity and good selectivity; it has been widely implemented in the monitoring of industrial waste [1], marine pollution [2][3][4], medical and health industry [5,6], and atmospheric environment [7,8].With the rapid development technologies of laser, electronics and computer, and the emergence of high-power and shortpulse lasers, Marine laser fluorescence radar monitoring system has been rapidly developed in various countries to detect the distribution of oil and gas in the seabed and oil pollution on the sea surface [9][10][11].
Common laser induced fluorescence detection techniques include fluorescence intensity analysis, fluorescence spectrum analysis, fluorescence lifetime analysis and so on according to different fluorescence parameters.The fluorescence intensity is the most direct physical quantity and has been widely used to produce the fluorescence of laser-induced substances [12].Furthermore, its application for the analysis of species for classification and recognition [13] has gained considerable attention.The peak wavelength and fluorescence intensity are typically used to determine the fluorescence substance content and for substance identification.However, fluorescence intensity is not generally used as a single parameter in practical applications owing to its vulnerability to various environmental factors such as external excitation light angle, ambient temperature, solvent concentration, and so on.The fluorescence ratio method [14], double ratio classification method and characteristic wavelength intensity ratio method [15] are commonly used to improve classification accuracy, with the average fluorescence intensity and elastic scattering intensity being considered as additional classification characteristics [16].The use of fluorescence spectroscopy for substance identification can be traced back to the 1980s [17].Zhou et al [18] characterized oil components from the Deepwater Horizon oil spill in the gulf of Mexico using fluorescence EEM and PARAFAC technique.Xu et al [19] proposed a new method to identify and quantify the vegetable oils present using cluster analysis and a Quasi-Monte Carlo integral.Three-dimensional fluorescence spectrum analysis is often applied in the treatment of multi-component system materials due to the overlapping phenomenon of complex components [20].In this analysis method, the 'three dimensions' typically include the excitation wavelength, fluorescence emission wavelength, and fluorescence intensity.This method can provide rich and complete fluorescence spectral information of a measured substance and exhibits good selectivity and high sensitivity to achieve comprehensive qualitative and quantitative analysis [21,22].However, it is difficult to solve the spectral similarity or to identify different substances based on the same spectrum phenomenon when applying three-dimensional fluorescence spectrometry for oil sample measurements with a small oil spectrum difference.Furthermore, a method for the coordination of detection distance and efficiency in the measurement process is an immediate requirement.Spectral coupling occurs between each component for complex petroleum mixtures, producing certain model errors in the spectral analysis.
Fluorescence lifetime is the main characteristic parameter of fluorescence; it is unaffected by external environment factors such as the intensity of external ambient light, excitation light, and fluorescence scattering angle during the measurement process.Due to its excellent stability and measurement accuracy, it is widely used for plant remote sensing monitoring, biomedicine [23] and the protection of cultural relics [24].Rowley et al [25] presented novel Bayesian methods for the analysis of exponential decay data.Gao and Li [26] estimated fluorescence lifetimes using extended Kalman filter.Wan et al [27] proposed an improved forward iterative deconvolution algorithm to verify the fluorescence lifetime.Hegazi and Hamdan [28] measured six crude oils using a pulsed laser radiation at 250 nm and verified that both fluorescence intensity and fluorescence lifetime the identification of crude oil.Two methods are commonly used for the measurement of fluorescence lifetime: the frequency-domain method and the time-domain method.The frequency domain method can theoretically distinguish the fluorescence lifetime of the samples when used in a multi-component fluorescence system.However, the actual realization of high-precision measurement requires the calibration of the system, which is known as the fluorescence lifetime of the samples.Furthermore, the sequential measurement of each component in the fluorescence sample increases the complexity, which reduces the speed of measurement.In the time-domain method, an ultrashort pulse laser with a high repetition rate is used to excite the sample, and the fluorescence lifetime information is obtained by analyzing the fluorescence attenuation curve after the pulse.Fluorescence lifetime microscopic imaging (FLIM) measures the intensity of fluorescence at different wavelengths with time as the variable.It combines the measurement of fluorescence lifetime with microscope imaging technology.The difference between microstructure and microenvironment can be reflected by the measurement of fluorescence lifetime in the sample microenvironment.It has high spatial and temporal resolution using in many fields [29].The main time-methods include the time-correlated single-photon counter [30], time-gated detection [31], and fringe camera methods [32].The timecorrelated single-photon technique achieves an ideal photon count rate and ultra-high time resolution.However, it only records one photo per cycle, the imaging speed is slow, and it is unsuitable for field measurement and analysis.The gated detection method is often used in oil pollution identification and detection owing to its advantages of simple realization, fast imaging speed, and convenient observation.
Motor oil is primarily used for lubricating in engineering; it is typically a hydrocarbon compound and is a good fluorescence emission substance.Furthermore, it is a multicomponent compound due to the different additives added by the manufacturers [33].The peak fluorescence intensity of each fluorophore interferes with the others owing to their individual fluorescence properties, due to which it is difficult to recognize when the intensity of fluorescence is used for oil identification.This problem can be overcome by using the gate detection method to measure fluorescence lifetime.In this study, we proposed a novel method to accurately identify the fluorophore using the position and occurrence time of the secondary wave peaks in the fluorescence lifetime decay curve.
The main contributions of this paper are as follows: (1) A novel method to recognize a multi-component compounds based on fluorescence attenuation curve was proposed by considering the occurrence time of the secondary crest as the new characteristic parameter.(2) A fluorescence lifetime imaging measurement system was designed.Using five commercially available motor engine oils as the experimental samples, the recognition performance of different kernel functions was analyzed and compared based on a support vector machine (SVM).(3) Comparing the average lifetime of fluorescence and the occurrence time of secondary peaks as the characteristic parameters of classification recognition, the results show that the occurrence time of the secondary crest for multicomponent compounds can be considered as a characteristic parameter for category identification with a high recognition rate and efficiency.
The rest of the paper is organized as follows: section 2 introduces the principles of fluorescence lifetime measurement.Section 3 is the experimental design, including the construction of experimental platform and data collection.Section 4 is the analysis and discussion of the experimental results.Section 5 concludes the paper.

Principles of fluorescence lifetime measurement
As mentioned in section 1, LIF technique is commonly used in many fields.In this section, the focus is given to fluorescence lifetime measurement based on gated-detection.
The atoms or molecules of matter transition from the ground state to an excited state by absorbing the excited light from the outside, and then releasing the resultant energy to  the ground state by emitting fluorescent photons.Fluorescence lifetime is defined as the time when an atom or molecule of a substance is in an excited state.It is the energy level lifetime of the ions in the atom or molecule of a substance, which changes exponentially over time [29].It is typically measured based on the time taken for the fluorescence intensity to reach 1/e of the maximum initial fluorescence intensity after the excitation stops.The fluorescence attenuation function can be expressed as:  fluorescence intensity attenuation curves, even with different initial intensities; that is, the fluorescence lifetime is equal to τ .The figure also demonstrates the stability of the fluorescence lifetime parameters, which are independent of the initial fluorescence intensity.This provides a theoretical basis for remote sensing monitoring using fluorescence lifetime parameters.
The fluorescence lifetime of oil was analyzed using the gated detection fluorescence lifetime imaging technique owing to these advantages.The fluorescence lifetime was obtained using equation (1), where the fluorescence intensity information, represented as I 1 and I 2 , was recorded in the same width opened at two different delay times, t 1 and t 2 [34].A multigated delay design was adopted for practical application and its principle is depicted in figure 3, The fluorescence intensity of multi-component compounds is calculated as the sum of the contribution of each component of fluorescence intensity.Subsequently, we performed the corresponding exponential fitting of different attenuation curves.The average fluorescence lifetime τ was calculated using as follows: Assuming a complex fluorescence system with m fluorescence components, i denotes the fluorescence lifetime of each fluorescence component obtained by fitting, and A i denotes the relative concentration of each fluorescent component.
The different transition times of the different fluorophore molecules produce fluctuations in the fluorescence attenuation curve.The identification accuracy and efficiency of oil species can be significantly improved by using the location and occurrence time of secondary crests as parameters combined with a SVM, and the problem of rapid classification and identification of oil species can be solved in terms of accident determination and pollution tracing in inland waters.

System of experiment
Based on the principle of gated detection method to measure fluorescence lifetime, this section described the laser induced fluorescence lifetime imaging system and experimental materials.Taking five kinds of oil as examples, the possibility of identifying the type of oil was analyzed verified by the secondary peak occurrence time.
The fluorescence attenuation curve and attenuation fitting according to different indices, can be used to calculate the average fluorescence lifetime, and also to determine the number and composition of fluorophores.Oil is a hydrocarbon with fluorescence characteristics; it is also a multi-component compound.Figure 4 depicts the designed system for laserinduced oil fluorescence lifetime imaging, and table 1 lists the parameters.
An ND:YAG laser with an output wavelength of 355 nm was selected as the excitation light source of the system.It contained a wavelength of 355 nm along with a fundamental wavelength of 1064 nm and a wavelength of double frequency, i.e. 532 nm.The light wavelengths of 1064 nm and 532 nm were filtered by using a 355 nm high-reflection spectroscope, and the oil sample was irradiated by adjusting the plane mirror.The concave lens broadened the laser beam and evenly irradiated the laser onto a quartz cuvette containing oil.Intensified charge coupled device (ICCD) was used as the image collector of the gated detection method, and fluorescence attenuation signals were obtained through repeated measurements with multiple delays.PD control was used to trigger the ICCD to ensure the synchronous shooting of the ICCD and laser pulse.Optical compensation was realized by using reflectors 1 and 2 [35].

Material
The five commonly used motorcycle lubricating oils of the 15W-40 series and gear oil sold in the market were selected and placed in a quartz cuvette, as shown in figure 5.They were then placed in the experimental system depicted in figure 4. Table 2 presents the names and codes of the experimental oil samples.The fluorescence intensity attenuation curves of the different oil products were analyzed through sampling.The fluorescence attenuation curves of the multi-component compounds were then fitted to obtain the fluorescence lifetime parameters [36].

Analysis
Five sample oils were each measured thrice to reduce the effect of random errors.The effects of background noise and dark  current were eliminated, and the average intensity value of the template with the same area (100 * 100 pixels) was obtained using the continuous images captured by the ICCD.The fluorescence intensity curve over time for the five oil samples was obtained after normalization and smoothing, as shown in figure 6.It can be observed that there are various similarities and differences in the fluorescence intensity curves of the five oil samples.Similarly, the entire attenuation period of oils a, b, c, d and e lies within 200 ns and they all exhibit the tendency to release fluorescence quickly after excitation and decay slowly after reaching the maximum value.The peak time of the fluorescence lifetime was observed in the period of 12-14 ns.However, the rate of attenuation of each oil is different to the maximum fluorescence intensity value; that is, each oil has a different fluorescence lifetime.The fluorescence intensity curve of each oil exhibits several secondary peaks, which are primarily attributed to the different additives added to each oil, the different energy level structures of the fluorophore molecules in the additives, and the superposition of the excited emission fluorescence curve of each fluorophore.
We performed repeated measurements of five lubricating oils by using the same measurement system based on fluorescence lifetime imaging, corresponding to the number of fluorophores and on the observation of the change rule of fluorescence intensity of different oil products over time.We observed that the time interval between the maximum fluorescence intensity and the second peak fluorescence intensity under the same measurement system was different, along with the intensity ratio.Furthermore, the time interval did not vary with the different in the gating delay time, as shown in figure 7.
A comparison of the occurrence times of the secondary crests after the normalized maximum fluorescence intensity  demonstrates that there are different occurrence times of secondary crests between different oil samples.The occurrence time of the secondary crest relative to the main peak was determined by selecting all the pixel points in the 100 × 100 pixels region of the fluorescence image.The lag time between the secondary crests of the different oil samples and the main peak was calculated, and the average and standard deviation were calculated within the study area.The confidence interval was determined based on three standard deviations, and the proportion of the number that lies within this interval was obtained when the secondary crest appearance time of all effective data points appeared in the total number of effective data points.Table 3 presents the statistical results.
The average occurrence time of the secondary crest varies for different oil films, and there is a certain precision for oil classification when it is used as the identification parameter, as shown in table 3. It can be clearly observed that the average fluorescence lifetime of different oil films varies along with occurrence time of the secondary crest in the fluorescence attenuation curve.Different oil species can be identified by using the SVM method by considering the occurrence time of the secondary crest in the fluorescence lifetime attenuation curve as a characteristic parameter.

Discuss
In this section, the identification rate of different oil species was verified by secondary peak occurrence time, the average fluorescence lifetime and the combination of the two methods.
The experimental results by using SVM method shown that the secondary peak occurrence time and the average fluorescence lifetime could be used as characteristic parameters for oil species identification with higher recognition rate and efficiency.A research area of size 100 × 100 pixels was selected in the fluorescence lifetime sequence images of the five lubricating oils.Each pixel can be used to represent a since it is considered as a time channel and records the change rule of the fluorescence intensity and time of a sample point.We fitted the function based on the multi-exponential decay rule of each sample and adopted the average fluorescence lifetime and the secondary crest as the feature vector.The first 50 data points of each sample were selected as the training set and the remaining characteristic quantity was selected as the test set.In the experiment, the characteristic parameters such as the fluorescence lifetime parameters were extracted for each oil based on the fluorescence attenuation curve, and the same sample and test sets were selected for comparative identification experiments that are conducted with four different kernel functions [37].
Table 4 presents the results of the recognition experiment based on the average of fluorescence lifetime.The penalty factor for all the kernel functions was set to 10 and the slack variable was set to 0.001.The kernel width of the radial basis kernel function (RBF) was set to six, the order of the polynomial kernel function was set to five, and the parameters of the Sigmoid function were considered as the default values.
The accuracy of the classification recognition of the four kernel functions is high in terms of the recognition rate.We can observe a short running time and high speed between the RBF and the sigmoid kernel function from the recognition rate.The results demonstrate that the RBF exhibits the best recognition performance and the highest classification recognition accuracy among four kernel functions.
Secondly, the identification experiment was conducted based only on the occurrence time of the secondary crest of The same sample and test sets were identified using four different kernel functions.For the kernel functions in table 5, the penalty factor was defined as 5, kernel width γ was defined as 5, and order was defined as7.
It can be observed that the difference in the recognition speed among the four types of kernel functions is insignificant.However, from the perspective of classification recognition accuracy, RBF recognition presents the highest recognition rate.
Lastly, the higher the number of feature vectors, the higher the recognition accuracy in identifying the experiments.Therefore, we combined the average lifetime and the occurrence time of the secondary crest as the characteristic parameters, and conducted identification experiments on the same sample and training sets using four different kernel functions.Table 6 presents the identification results.Among the kernel functions, the penalty factor was defined as 5, slack variable ξ was defined as 0.001, kernel width γ was defined as 1, and order was defined as 3.
It can be observed that the difference between the recognition speed and the running speed of classification recognition is insignificant when combining the average lifetime of fluorescence and the occurrence time of the secondary crest in the fluorescence attenuation curve.However, the recognition efficiency is significantly improved.When combined with the occurrence time of the secondary crest as the characteristic  parameter of SVM, oil species identification presents considerable advantages.
In the process of SVM, the selection of parameters will directly affect the efficiency and accuracy of the experiment.The utilization of different parameters can lead to variations in recognition accuracy, even when the same sample set, test set, and radial basis function are employed.The optimal parameters are usually determined by cross-validation method based on network search.It was found that the optimal solutions of penalty factor and kernel parameter are both 256.The result of parameter selection was shown in figure 8, the recognition rate could reach more than 96%.
According to the fluorescent life sequence images of five kinds of lubricating oil studied, 4 groups of results of each oil were randomly selected as training samples in all pixels in the excited region of the oil sample.The four groups of training samples are 1000, 500, 100, 50 respectively, and 100 pixels of each oil sample are randomly selected as test samples.Based on the optimized parameters, we employed the SVM classification method with RBF for conducting comparative experiments, and the results are presented in table 7.
It can be observed from table 7 that the SVM method based on RBF function is feasible and has a good recognition rate.After comparing different sample numbers, it is observed that reducing the number of training samples has minimal impact on the recognition results.This finding suggests that the SVM recognition method exhibits excellent classification performance even with relatively small sample sizes.

Conclusion
Fluorescence lifetime detection is a commonly used method to detect LIF.However, the fluorophore lifetime attenuation curves interfere with each other, increasing the difficulty of identification when the fluorescence lifetime attenuation curves of multi-component compounds are fitted using the time-gated detection method.Consequently, we proposed an identification method based on the occurrence time of the secondary crest combined with the characteristics of fluorescence emission of multi-component compounds.
(1) A method of drawing fluorescence lifetime decay curve by using gated detection and fluorescence imaging technique is put forward.(2) It was verified that the possibility of using fluorescence mean life as a characteristic parameter of multi-component compounds for species identification.(3) Combined with the characteristics of fluorescence lifetime decay curve, a method of multi-component substance identification was proposed based on the secondary peak occurrence time.
(4) Taking motorcycle lubricating oil as examples, combined with the method of SVM, the classification recognition rate and efficiency of the two methods were verified by experiments, the secondary peak occurrence time and the fluorescence mean life were taken as the characteristic parameters respectively.
The experimental results indicate that the occurrence time of the secondary crest for multi-component compounds can be considered as a characteristic parameter.It is similar to the average fluorescence lifetime parameter, for category identification with a high recognition rate and efficiency.The proposed method can be used for the recognition of multicomponent compounds using fluorescence parameters in a LIF technique.

Figure 2 .
Figure 2. Stimulation of fluorescence decay curve with different fluorescence intensity.
)where τ denotes the lifetime (in ns) and I 0 denotes the initial fluorescence intensity.Assuming the same fluorescence lifetime, i.e. τ = 10 ns, different initial intensities I 0 were considered at 20, 30, and 40 intensity units.Figures1 and 2depict the fluorescence attenuation curves of different simulated initial fluorescence intensities.From the graph, it can be clearly observed that the initial intensities of the fluorescence 1/e values are equal for different

Figure 3 .
Figure 3. Schematic diagram of fluorescence lifetime measurement by time-gated detection method.

Figure 4 .
Figure 4. System diagram of laser induced oil sample fluorescence lifetime imaging.

J Min et alFigure 5 .
Figure 5. Lubricating oil sample to be tested.

Table 2 .Figure
Figure Fluorescence intensity curves of five sample oils.

Figure 7 .
Figure 7. Secondary peak occurrence time of different oil sample.

Table 1 .
Parameter of laser-induced fluorescence lifetime imaging system.

Table 3 .
Secondary peak occurrence time parameters of different oil.

Table 4 .
Recognition result comparisons of different kernel function based on fluorescence mean life.

Table 5 .
Recognition comparisons of different kernel function based on secondary peak occurrence time.

Table 6 .
Compare the training effect with different kernel functions.

Table 7 .
Comparison of different sample number based SVM method with RBF kernel function.