1. Introduction

Oil pollution is a serious environmental problem, as wind and waves can scatter an oil spill over a wide area of sea within just a few hours [1]. It often occurs during oil transportation or exploration processes, which seriously affects the hydrological environment and biological survival [2]. Laser fluorosensors are useful instruments and they use the phenomenon that aromatic compounds in mineral oils may absorb ultraviolet excitation light and become electronically excited. This excitation is then rapidly removed through the process of fluorescence emission, primarily in the visible region of the spectrum [3]. Furthermore, laser-induced fluorescence (LIF) emissions of naturally-occurring substances, such as chlorophyll, occur at quite different wavelengths than mineral oils [4]. As different types of oil exhibit characteristic variability of their spectral distributions, it is possible to differentiate among various classes of oil [5].

Remote sensing, which is often combined with laser-induced fluorescence technology, plays an important role in the detection of bioaerosols [6], monitoring of vegetation status [7], investigation of stone monuments [8] and oil pollution response [2]. Most laser fluorosensors used for oil spill detection employ a pulsed laser operating in the ultraviolet region of 308-355 nm [9]. However, the high cost and complexity of conventional pulsed lidar systems hinder their usability to researchers [10,11]. In recent years, continuous-wave laser diodes with high-power and low-cost have been developed and widely used in Scheimpflug lidar systems based on the Scheimpflug principle, which describes the relationship between image planes and object planes when they are not parallel [10–13]. Since the Scheimpflug systems were initially developed for the purpose of remote modulation spectroscopy [10], they can only detect elastic signals instead of inelastic signals, such as fluorescence or Raman signals. In 2016, an inelastic hyperspectral lidar was realized by combining Scheimpflug lidar and hyperspectral push broom imaging techniques in Zhao’s work, which mainly focused on small organisms in water, such as phytoplankton and zooplankton [14]. However, the system design for range measurement is approximate. In our work, the theory of system parametric design is established for the first time and is adapted for aquatic applications. To our best knowledge, it is the first time that 446 nm is employed as the excitation wavelength in a laser-induced fluorescence system for range-resolved oil pollution detection, with a compromise of weak water absorption, availability to high-power LD and reasonable fluorescence excitation efficiency. Seven kinds of typical oil samples are measured and distinguished using the principal component analysis (PCA) and linear discriminant analysis (LDA) methods. Our experimental results show that not only can ultraviolet light be used as the excitation wavelength, but also that blue LDs have great potential for oil pollution discrimination.

2. Principles and method

According to the Scheimpflug principle, when the lens and object planes are not parallel, the image plane will intersect both the lens and object planes. Theoretically, an infinite focus depth can be achieved despite using a large optical aperture.

The relationship between pixel position and distance is derived to the following form [12]:

Here, d is the distance to the lens from the object plane, α is the tilt angle of the lens plane relative to the object plane, and β is the tilt angle of the image plane relative to the lens plane. $v_0$ is the distance between the origins of the image and lens planes (OʹOʹʹ), and $p_I$ is distance between the pixel position and the origin of the image plane (MʹOʹʹ).

It can be noted that some parameters of Eq. (1) are not independent. The value of $v_0$ can be calculated through the lens equation, i.e., $1/u+1/v=1/f$, as illustrated in Fig. 1:

 

Fig. 1 Scheimpflug principle: the image plane intersects both the lens and object planes when the object plane is not parallel to the lens plane. Oʹ and Oʹʹ are the origins of the lens and image planes, respectively; $p_I$- the pixel position of the image sensor on the image plane, d -the distance to the lens from the object plane, α - the tilt angle of the lens plane to the object plane, β - the tilt angle of the image plane to the lens plane. M is the object point and Mʹ is the image point accordingly, which satisfies the lens equation, i.e., $1/u+1/v=1/f$.

Download Full Size | PDF

Then the tilt angle β can be deduced from:

Furthermore, the imaging position can be determined when the initial parameters d, α, f_{and z are set. The point M on the object plane is projected onto the image plane Mʹ, as illustrated in Fig. 1. The coordinate of Mʹ can be given by:}

When applied to underwater applications, refraction of the laser beam must be taken into consideration. MN on the object plane in the water is equivalent to MNʹ in the air according to the refraction law, as illustrated in Fig. 2(a). In accordance with the previous derivation, the imaging plane can be determined when the initial parameters are configured. The corrected distance can be calculated from trigonometric functions:

 

Fig. 2 Scheimpflug principle applied to the underwater environment. (a) The light path which indicates that refraction of the laser beam must be taken into consideration. (b) The relationship between pixel number and distance with optical parameters: d = 0.306 m, f = 55 mm, α = 84°.

Download Full Size | PDF

Here, z is the uncorrected distance, which is deduced from Eq. (1), and $z_M$ is the distance of the interface (OM). ${\theta }_1$ is the angle between the water surface and MNʹ, and ${\theta }_2$ is the angle between the water surface and MN. Figure 2(b) gives the theoretical relationship between pixel number and uncorrected (MNʹ), as well as corrected distance (MN). It can be noted that the difference between two curves (i.e. error) grows larger with increase of the distance, so refractive correction is necessary for underwater measurement. In Fig. 2(b), the data points marked by * are experimental test data, which coincide well with the theoretical saffron curve; in addition, it can be seen that the distance of interface is about 2.5 m.

3. Instrumentation

The schematic of the hyperspectral Scheimpflug lidar system is shown in Fig. 3. The laser source is a commercial blue laser diode emitting around 446 nm with 1.5 W maximum output power. The laser beam is collimated and transmitted into the air; it travels around 2.5 m then into a 2.4 m-long, tap-water-filled tank. A white board is placed at the end of the tank to terminate the beam and to increase the weak termination echo for the convenience of system adjustment. The receiver is a ø 58 mm, f = 55 mm imaging lens (Canon, Japan), and it is located at a separation of d = 0.306 m from the object plane, as shown in Fig. 2(a). The tilt angle α of the lens plane to the object plane is set to 84°. According to Eq. (3), the tilt angle β of the image plane to the lens plane can be calculated to be approximately 10°, that is, the illuminated probe volume is imaged sharply at an image plane 10° off the optical axis. Therefore, a 50-μm slit is tilted 10° to let the collected light pass through. L1 and L2 are collimated lenses, and OF is a 450 nm long-pass optical filter (FEL0450, OD = 2 at 446 nm, Thorlabs Inc., USA). A prism-grating-prism (PGP) configuration is employed to disperse the light so that the spectral and distance information can be obtained by a 2D-CMOS camera (CMOSIS CMV2000, $2048\times 1088$pixels, $5.5\mu m\times 5.5\mu m$pixel size, Lumenera, Canada). In our setup, a transmission grating (G, 300 grooves per mm, Thorlabs Inc., USA) is sandwiched between two symmetrical wedge prisms (P1 and P2). The PGP component can image the slit onto the 2D-CMOS camera (400-770 nm; 1.9-6.4 m), which is tilted 10° off the optical axis. The range calibration was done by deducing the right pixel number from the termination echo, and the range correction was performed using Eq. (6), as illustrated in Fig. 2(b).The spectral calibration was done with a six-order polynomial fit by a Mercury-argon lamp (HG-1, Ocean Optics Inc., USA). Two lines (576.96 and 579.066 nm) can be obviously distinguished such that the spectral resolution of our system is better than 2 nm.

 

Fig. 3 Schematic diagram of the inelastic hyperspectral Scheimpflug lidar system; the figure on the top left corner is the top view of the system. L1 and L2 are collimated lenses, and OF is the 450 nm long-pass optical filter. P1 and P2 are two symmetrical wedge prisms, and G is a transmission grating with 300 grooves per mm.

Download Full Size | PDF

4. Results and discussion

Seven kinds of typical oil samples were measured by our inelastic hyperspectral Scheimpflug lidar system. The sources of the oil samples are listed in Tab. 1. An excitation wavelength of 446 nm was employed to induce the fluorescence of oil samples. Each kind of sample was injected into a tube and measured 5 times at different distances to establish an accurate estimation model. This is a dark experiment and the background signal is subtracted respectively. The light-intensity distribution from the test range was captured on the 2D-CMOS camera with proper exposure time set accordingly.

Table 1. Sources of the Oil Samples

View Table

The experimental results for one such distance are displayed in Fig. 4. The fluorescence signal of the oil sample can be clearly seen in the spectral range of 400-770 nm, and the position of the oil sample can be deduced from the vertical axis to approximately 3.69 m. In addition, the Raman signal of water at 523 nm was captured, corresponding to a Raman shift of about 3300 cm⁻¹, which is a strong O-H stretch characteristic peak [15]. The elastic signals of the oil sample, tap water, and acrylic plate can also been seen in Fig. 4, and the distance to the front surface of the tank is about 2.5 m according to the backscatter signal of the acrylic plate. Moreover, it can be noted that the relationship between pixel position and distance is nonlinear; that is, the range resolution is higher in near field (~mm) than far field (~cm) in our experiment. The fluorescence spectrum of the oil sample, which was averaged by the strong fluorescence signals of several adjacent rows to increase signal-to-noise ratio, was extracted for the subsequent spectral analysis of oil discrimination.

 

Fig. 4 Light distribution as it appears at one occasion on the 2D-CMOS camera. An oil sample was measured at the range of 3.69 m. The effective spectral response curve is shown on top. The fluorescence spectra are obtained without instrument spectral response compensation.

Download Full Size | PDF

Figure 5 gives a total of 35 fluorescence spectra of seven kinds of oil samples with a wavelength range of 460-750 nm. Each spectrum is normalized by its highest peak. The spectra shapes of different types of oil samples are slightly different from each other, owing to the different contents of aromatic hydrocarbons. Crude oil fluorescence returns in the region between 460 and 750 nm, with the maximum centered in the approximately 480 to 520 nm region, while other kinds of mineral oils cover narrower spectral regions. In addition, oil pollution can be easily distinguished from Chlorophyll, which yields a sharp peak at 685 nm [3]. However, judging fluorescence spectra is subjective, so it is necessary to use a statistical method, such as the PCA method, to analyze the whole spectra from a methodological point of view to obtain objective results.

 

Fig. 5 Normalized spectra of seven kinds of oil samples

Download Full Size | PDF

The PCA procedure facilitates the extraction of important information and leads to substantial reduction of redundant data features, which is widely used in chemometry [16,17]. The original data matrix obtained from our experiments is $35\times 1608$, in which 35 is the number of oil samples and 1608 is the number of spectral sampling points for each oil sample. After the PCA operation, the original data matrix can be decomposed into a series of principal components, i.e., PC1, PC2, PC3 etc., normally the first few of which carry with most of information. In our case, the first 6 PCs carry with 94.5% information. The relationship between PC1 and PC2 is shown in Fig. 6. Different types of oil samples are well-separated using this procedure. Further, LDA was employed to predict the classification of the oil sample. A leave-one-out predictive model was established in which samples are left out one by one, and the rest are used to build the model [16,18]. In the present case, the predictive accuracy is 100% if we use the first 6 PCs to represent the whole matrix, which shows that blue excitation LDs have great potential for oil pollution discrimination.

 

Fig. 6 The relationship between PC1 and PC2 of seven kinds of oil samples

Download Full Size | PDF

5. Conclusion and outlook

We have developed an inelastic hyperspectral Scheimpflug lidar system for range-resolved oil pollution detection and discrimination. To the best of our knowledge, this is the first time that the theory of system design parameters is established and corrected for aquatic applications, and the laser-induced fluorescence spectra with an excitation wavelength of 446 nm are shown to be feasible to classify oil pollutions. It indicates that violet LDs (405 nm) are likely to be reasonable choices as excitation light sources for oil discrimination in the future. Our system can discriminate between oiled and unoiled, naturally-occurring substances, such as kelp and seaweed. In addition, the application of our range-resolved inelastic system can be extended to terrestrial environments in the future. However, some limitations should be concerned. The absorption of water limits the range of aquatic lidar system, especially of which fluorescence or Raman signals is needed. Laser eye safety would also be a substantial concern, the high-power ultraviolet LDs which emit below 400 nm are in urgent need for oil pollution detection. In general, our inelastic hyperspectral Scheimpflug lidar system has been demonstrated to be feasible for oil pollution discrimination and has potential for more applications in both marine and terrestrial environments.