Design of Polarization Spectroscopy Integrated Imaging System

Abstract

To simultaneously acquire the spectral and polarization information of the target and achieve the monitoring and identification of the target object, a polarization spectral integrated imaging system is proposed in this paper. Firstly, the structural principle of the polarization spectral integrated imaging system is introduced. The relationship between the spatial resolution, spectral resolution, and the system’s structural parameters is analyzed. The design of the optical part of the polarization spectral integrated imaging system is completed, along with the tolerance analysis. Secondly, the mechanical structure of the polarization spectral integrated imaging system is designed. Finally, by using a drone to carry the polarization spectral integrated imaging system, a simulation experiment for sea surface oil spill monitoring is conducted, and the hyperspectral and polarization information of the ocean, crude oil, fuel oil, palm oil, diesel, and gasoline are obtained. The polarization and spectral information were integrated. The integration of hyperspectral and polarization data yields remarkable enhancement outcomes, allowing for the clear delineation of previously challenging-to-identify crude oil contamination areas against the marine background in the fused images, characterized by sharper boundaries and improved discriminability. This accomplishment underscores the feasibility of our system for the rapid identification of large-scale oil spill events.

1. Introduction

Hyperspectral imaging technology stands out for its capability to concurrently gather spatial and spectral information reflected by targets, integrating imagery and spectral data seamlessly. This technology exhibits heightened sensitivity to the species, material, and composition of targets, effectively differentiating them from their backgrounds based on material discrepancies. While hyperspectral imaging has found applications in marine target monitoring, offering useful capabilities in oil spill detection and broad classification, its effectiveness can be hindered by sea surface glare, resulting in low signal-to-noise ratios [1]. Conversely, polarization imaging technology captures both spatial and polarization information of targets, demonstrating sensitivity to surface structures and textures [2]. Notably, it excels in identifying targets even when they share the same intensity reflectivity with the background, effectively mitigating interference from sea fog and glare [3]. Evidently, both hyperspectral and polarization imaging have their respective strengths and limitations in monitoring typical marine targets [4]. However, the “spectral + polarization” multidimensional hyperspectral imaging mechanism harnesses the complementary strengths of these detection methods. By acquiring spectral information at every point in the image and polarization information at each wavelength, it tackles the challenge of degraded image quality caused by varying or even contrasting polarization characteristics across different wavelengths. This approach offers a multi-faceted understanding of target characteristics, significantly enhancing the detection and identification performance of typical marine targets. It holds immense potential for achieving breakthroughs in oil spill detection, oil type identification, thickness inversion, and other challenges, thereby bolstering China’s precision marine monitoring capabilities.

In the context of scientific research advancements, it is noteworthy that in 2008, researchers at the University of California introduced a polarization grating-based imaging spectrometer, which incorporated multilayer gratings for spectral dispersion. However, this design led to reduced optical throughput and signal-to-noise ratio, posing challenges for target detection in complex environmental conditions [5]. Advancing further, in 2016, Yu Xun and colleagues developed a polarization spectral imaging system leveraging liquid crystal tunable filters. This innovative system was applied to detect concealed fake plants amidst green vegetation through polarization spectral analysis. The outcomes demonstrated a significant improvement, with the polarization spectral fusion images exhibiting a 72% increase in information entropy and a remarkable 250% enhancement in average gradient compared to standard spectral fusion images. Nevertheless, the operational deployment of liquid crystal tunable filters necessitated an increase in energy consumption, rendering this approach impractical for applications lacking a consistent power supply [6]. More recently, in 2018, Bai Caixun from Nanjing University of Science and Technology presented a sophisticated spectral polarization synchronous imaging technique utilizing birefringent shear interference combined with ferroelectric liquid crystal high-speed polarization modulation. This method enabled the achievement of a broad spectral range from 400 to 1000 nm, encompassing 128 spectral channels, with a spectral resolution superior to 5 nm. Despite these advancements, it was observed that the optical performance of the liquid crystal components in this system was adversely affected by variations in humidity, introducing potential measurement inaccuracies, particularly unsuitable for deployment in maritime environments where atmospheric conditions can be unpredictable [7]. In 2023, Qi Chen proposed a compact polarization spectral imaging method based on linear gradient filters and pixelated polarization modulation. The system operates within a spectral range of 430–880 nm, with a spectral resolution of 10 nm and a spatial resolution of 0.215 mrad. This system achieves miniaturization, simultaneous acquisition of polarization and spectral information, and simplified reconstruction of multidimensional information. However, the linear gradient filters used in the system face challenges in ensuring consistent manufacturing quality [8].

In this paper, we design a slit polarization imaging spectrometer which can obtain both polarization and spectral information at the same time. It can realize high spectral resolution, high spatial resolution, and real-time multi-polarization angle imaging. The polarization spectrum image of the simulated target of oil spill monitoring is obtained by using this system, and the feasibility of the polarization spectrum imager in the field of oil spill monitoring is preliminarily verified.

2. Basic Theory

In order to ensure the acquisition of spectral information and polarization information at the same time, the system adopts the polarization spectrum integration design. Figure 1 is the schematic diagram of the polarization spectrum integration system. The telescope system images the oil spill image information from the sea surface into the slit. The two-dimensional image at the slit is segmented into one-dimensional spatial information, and then collimated by the collimation system, incident on the chromatic dispersion element. The chromatic dispersion element disperses the light of different colors, and then forms an image on the polarization detector through the imaging lens group, so as to obtain one-dimensional spatial information, one-dimensional spectral information, and four-dimensional polarization information. The polarization detector uses a micro-polarizer array covered on a CMOS sensor to separate polarized light in different vibration directions, so as to obtain four-dimensional polarization information. Finally, the spectrum and polarization information of the two-dimensional target are collected simultaneously by pushing and sweeping along the direction perpendicular to the slit. Finally, through a fusion algorithm, a fused image combining polarization and spectral information of the two-dimensional spatial image is obtained.

The polarization detector adopts a focal-plane polarization detector, as shown in the figure below. In front of the CMOS sensor, a combination of micro polarizers with micro- and nano-processing is set, so that the light intensity of different polarization directions can be obtained on each pixel of the CMOS. After reasonable polarization element arrangement and mathematical calculation, the Stokes parameter information of the target can be obtained.

According to the mathematical meaning of Stokes parameters,

$$\begin{array}{l}{\mathrm{S}}_0=I_{0^{\circ }}+I_{{90}^{\circ }}\\ S_1=I_{0^{\circ }}-I_{{90}^{\circ }}\\ S_2=I_{{45}^{\circ }}-I_{{135}^{\circ }}\end{array}$$

(1)

The polarization information of the target is obtained by obtaining the polarization intensities in various directions, as shown in Figure 2.

3. Optical System Design

3.1. Design of Telescope System

In the design of imaging spectrometer, the structure of the front-end telescope system has an important effect on the imaging quality and spectral resolution. In order to ensure the accuracy of the system in the calculation and application of detecting the target area and reduce the spectral line bending of the system, the image square and centroid structure is adopted in the telescope system [9]. This system adopts 3 × 3 pixel combination. The pixel size of the detector is 3.45 μm × 3.45 μm, the pixel number is 2248 × 2048, the calculation shows that the pixel size of this system is 10.35 μm, and the zoom ratio of the adopted spectral system is 1. The spatial resolution relation is as follows:

$$0.21 \mathrm{mrad}={\displaystyle \frac{10.35 \mathsf{\mu }\mathrm{m}}{f}}$$

(2)

In the formula: $f$ is the focal length of the telescope system.

It can be obtained that $f=49 \mathrm{m}\mathrm{m}$. In order to match the telescope system with the slit and the beam splitting system, the image plane of the telescope system should be the same length as the slit. The reduction ratio of the optical splitting system in this design is 1, so the slit width is consistent with the detector space dimension $H$, that is, the slit length $H=7$ mm. It can be obtained that the half-field angle of the telescope system is:

$$\omega =\arctan \left({\displaystyle \frac{H}{2f}}\right)=4{.1}^{\circ }$$

(3)

That is, the field of view angle of the telescope system is 2ω = 8.2°. Reducing the F-number can improve the ability of optical system to collect light energy, but it will also increase the difficulty of the optical system’s design. After comprehensive consideration, the F-number of the system is determined to be 3. The optical path diagram of the telescope system is shown in Figure 3, the point list diagram of the telescope system is shown in Figure 4, and the MTF curve of the telescope system is shown in Figure 5. It can be seen from Figure 5 that the MTF of the system is greater than 0.5 at 48 lp/mm, which is close to the diffraction limit.

The RMS radius dimensions of each field of view of the telescope system are shown in

Table 1. RMS radius of the telescope system in each field of view.

Field	−4.1°	−2.87°	−1.43°	0°	1.43°	2.87°	4.1°
RMS Radius/μm	4.106	3.502	3.553	3.749	3.553	3.502	4.106

.

3.2. Design of Spectroscopic System

Due to the opposite spectral line curvature characteristics of prisms and gratings [10], a prism–grating (P–G) combination is employed as the dispersion system in this spectroscopic system [11]. The design must ensure both imaging quality and spectral performance. The collimation system and the imaging lens group are crucial components of the P–G hyperspectral imaging spectrometer. The collimation system serves to collimate the light from the incident slit into a parallel beam, which is then dispersed by the P–G combination chromatic dispersion system. Therefore, aberrations affecting beam collimation characteristics, primarily spherical aberration, must be corrected. The imaging system, on the other hand, focuses the dispersed light beam and forms an image on the image plane, necessitating the correction of all aberrations impacting imaging quality [12]. Additionally, to reduce design and manufacturing costs, the collimation system and the imaging system are designed with the same structure, where the collimation system adopts the inverted structure of the imaging system. This combined structure exhibits optical symmetry, facilitating the correction and elimination of system aberrations [13].

Based on the angular dispersion characteristics of the P–G combination dispersion system, the difference in dispersion angle of the light beam ranging from 410 to 900 nm originating from the center of the incident slit after passing through the P–G combination dispersion system can be determined. Given that the length of the spectral dimension of the imaging plane is $V$, the image side focal length of the imaging system can be calculated:

$$f^{\prime }={\displaystyle \frac{V/2}{\tan {\omega }_1}}$$

(4)

The image plane size of the CMOS detector adopted in this system is 8.445 × 7.066 mm, and the spectral-dimension size $V=7\mathrm{m}\mathrm{m}$. From this, the image-side $f^{\prime }$ length of the imaging system can be obtained as 49 mm. The structural form of the imaging system is shown in Figure 6.

The collimation system is designed as the inverted structure of the imaging lens, thus the constructed imaging system has a lateral magnification of 1. The incident slit is located at the object-side focal plane of the collimation system, and the aperture stop is placed at the grating surface. The optical axis of the imaging system is inclined along the dispersion direction, and the image plane is inclined along the dispersion direction. The optical splitting system consists of a 360 lp/mm transmission grating and a refracting prism with an apex angle of 8.18°. The overall structure of the spectroscopic system shown in Figure 7:

3.3. Combined Optimization of the Complete System

3.3.1. System Structure

We combine the telescope system with the spectroscopic system, and then perform joint optimization to ultimately obtain a system with a spectral resolution of 2 nm. The overall optical path diagram of the system is shown in Figure 8, as follows:

3.3.2. Spot Diagram and MTF of the Complete System

The spot diagram and the MTF of the complete system are shown in Figure 9 and Figure 10, respectively. The RMS radius sizes of each field of view of the complete system under 410 nm, 650 nm, and 900 nm are shown in

Table 2. The RMS radius of each field of view of the complete system at 410 nm, 650 nm, and 900 nm wavelengths.

Field	410 nm/μm	650 nm/μm	900 nm/μm
−4.1°	2.675	3.162	6.536
−2.87°	2.874	1.899	4.068
−1.43°	2.957	3.413	3.005
0°	3.137	4.368	2.535
1.43°	2.957	3.413	3.005
2.87°	2.874	1.899	4.068
4.1°	2.675	3.162	3.536

.

3.3.3. Spectral Resolution

Spectral resolution analysis was performed on the system for 410 nm and 412 nm, 650 nm and 652 nm, and 900 nm and 898 nm. The corresponding spot diagrams are shown in Figure 11. It can be seen from the figure that the light spots are clearly separated, indicating that the spectral resolution of the system reaches 2 nm.

3.3.4. Tolerance Analysis

After the system design is completed, a tolerance analysis needs to be conducted to account for relevant errors that may arise during actual processing and assembly, and to assign appropriate tolerance components [14]. During this tolerance analysis process, reasonable tolerances must be allocated to all optical elements to ensure that the detection performance of the optical detection system meets the required standards while minimizing the costs of the optical components, assembly, and calibration, thereby optimizing the overall performance of the system [15]. Based on the actual processing levels referenced, the tolerances for the optical system are shown in Figure 12.

Using the tolerance analysis, the MTF is taken as the evaluation criterion, and the system undergoes 500 Monte Carlo analyses. The Monte Carlo analysis results at wavelengths of 410 nm, 650 nm, and 900 nm are presented in

Table 3. Monte Carlo analysis results.

Probability	MTF at 48 lp/mm
	410 nm	650 nm	900 nm
>98%	0.698	0.535	0.459
>90%	0.751	0.633	0.525
>80%	0.779	0.664	0.551
>50%	0.814	0.725	0.602
>20%	0.833	0.765	0.634
>10%	0.841	0.778	0.655

. The analysis shows that at 48 lp/mm, the MTF of the optical system can reach 0.55 with an 80% probability surrounding the mean. This meets the requirements of the optical imaging system and demonstrates that the system has good tolerance characteristics.

3.3.5. Signal-to-Noise Ratio Analysis

The signal-to-noise ratio (SNR) is an important performance evaluation index of polarization spectral imaging remote sensors.

The main task of the system in this paper is to monitor oil spills at sea, so it uses signals within the solar spectrum range for inversion to obtain information about oil spills on the sea surface. The uncertainty of spectral inversion usually decreases with the increase of the SNR of the detector. The signal-to-noise ratio at different wavelengths of the spectrometer will be calculated based on its various parameters.

The formula for calculating the SNR in optical remote sensing imaging is as follows:

$$SNR={\displaystyle \frac{S_e}{N_e}}={\displaystyle \frac{S_e}{\sqrt{S_e+{{\sigma }_R}^2+D_e}}}$$

(5)

In the above formula, $S_e$ is the signal value, $N_e$ is the number of noise electrons, ${\sigma }_R$ is the root mean square value of the detector’s readout noise, and $D_e$ is the number of dark signal output electrons from the detector.

The number of signal electrons is the response of the detector to the spectral radiant energy obtained by the spectrometer, and its specific calculation formula is as follows:

$$S_e\left(\lambda \right)={\displaystyle \frac{\pi A_{d}L\left(\lambda \right){\tau }_o\left(\lambda \right)DE{t}_{\mathit{int}}\eta \left(\lambda \right)\Delta \lambda \lambda }{4F^{2}hc}}$$

(6)

$A_d$ is the detector pixel area; $L\left(\lambda \right)$ is the spectral radiance at the entrance pupil; ${\tau }_0\left(\lambda \right)$ is the efficiency of the optical system; $DE$ is the diffraction efficiency of the grating; $t_{int}$ is the integration time and the frame rate is selected to be 100 Hz, with a maximum integration time of 10 ms; $\eta \left(\lambda \right)$ is the quantum efficiency of the detector, $∆\lambda$ is the spectral sampling interval, and $\lambda$ is the wavelength value; $F=f/D$ is the $F/\#$ of the optical system, and $F=3$ is taken according to the design results; $h$ is Planck’s constant $6.626176\times {10}^{-34}\mathrm{J}·\mathrm{s}$, and $c$ is the speed of light.

First, calculate the spectral radiance at the entrance pupil of the spectrometer. The spectral radiant energy obtained by the airborne spectrometer during the shooting of the sea surface mainly comes from the reflected radiation of the sea surface.

$L\left(\lambda \right)$ is the spectral radiance at the target wavelength of approximately $\lambda$, which can be obtained by the following formula:

$$L\left(\lambda \right)={\displaystyle \frac{E\left(\lambda \right)}{2\pi }}{\tau }_{w1}$$

(7)

$E\left(\lambda \right)$ is the sea level solar irradiance, ${\tau }_{w1}$ is the light transmittance of seawater, and the radiation spectral curve is shown in Figure 13.

$\eta \left(\lambda \right)$ represents the quantum efficiency of the area array detector. The quantum efficiency curve of the detector is shown in Figure 14:

${\tau }_0\left(\lambda \right)$ represents the spectral radiant transfer efficiency of the optical system, which can be expressed as follows in this structure:

$${\tau }_0\left(\lambda \right)=t^n\left(\lambda \right)t_f\left(\lambda \right)G\left(\lambda \right)$$

(8)

$t_f\left(\lambda \right)$ is the transmittance of the optical filter, and $t\left(\lambda \right)$ is the transmittance of the lens surface. The transmittance curves of the anti-reflection film on the designed lens at different wavelengths are shown in Figure 15. It can be seen from the figure that the designed optical lens has a good transmittance in the range of 410–900 nm, with $t\left(\lambda \right)=95\%$; $n$ represents the number of transmission surfaces in the optical system, with $n=36$.

$G\left(\lambda \right)$ represents the diffraction efficiency of the transmission grating. The diffraction efficiency curve of the selected volume holographic grating at different wavelengths is shown in Figure 16:

The noise values provided by the detector are shown in

Table 4. Noise value of the detector.

Noise Terms	Noise Values
Readout noise	1.6 e−
Dark count	80 e−/s/pix@35 °C

:

The estimated noise value for a single pixel is approximately 100 e⁻. The SNR calculated for the imaging spectrometer detecting wavelengths from 410 to 900 nm is shown in in Figure 17:

In 2022, Ying performed a study on a typical polarization multispectral imaging remote sensor. The signal-to-noise ratios of 441 nm, 488 nm, and 610 nm were calculated to be 43.6404, 47.1396, 45.0559, and 48.2649, respectively [16]. Ying’s research provided a valuable benchmark for the optimization of SNR in multispectral imaging sensors. Our analysis indicates that the SNR for our imaging spectrometer, when detecting wavelengths ranging from 410 to 900 nm, exceeds the values reported by Ying for the 441 nm, 488 nm, and 610 nm wavelengths. This improvement can be attributed to several factors, including advancements in sensor technology and the optimization of our optical system design. The enhanced SNR is crucial for obtaining high-quality images.

4. Mechanical Structure Design

Given the numerous lenses involved in the spectral polarization imaging system, installing them all within a single lens tube would compromise the manufacturability and make it challenging to effectively control the quality of the lens assembly process. Therefore, the spectral polarization imaging subsystem is divided into three parts: the front lens group, the middle lens group, and the rear lens group. The front lens group includes the telescope system and slit adjustment components; the middle lens group comprises the collimation system and prism–grating dispersion components; and the rear lens group consists of the imaging system and detector adjustment components. During mechanical and optical adjustments, each module can be precisely assembled and adjusted independently. Additionally, corresponding adjustment interfaces are designed between the modules to facilitate their combined assembly and adjustment. The mechanical structure is illustrated in Figure 18, and a physical image of the hyperspectral polarization imaging system is shown in Figure 19.

5. Experiment

Radiometric calibration establishes a quantitative relationship between the digital quantized values (DN) produced by each detection element of an imaging spectrometer and the corresponding radiant brightness values within its field of view. This process is fundamental to ensuring the accuracy of data acquired from the imaging spectrometer, as only data that has undergone precise radiometric calibration can accurately represent the radiative characteristics of terrestrial objects. The process of radiometric correction is shown in Figure 20.

Suppose the band number, row number, and pixel number of the data are represented by $m, l$, and $n$, respectively. In this context, we consider an example involving radiometric correction applied to the pixel data located at the $n$-th pixel in the $l$-th row within the $m$-th band channel. We assume that the calibration energy levels are denoted as $k_1$ and $k_2$, with the condition that $k_1<k_2$. Furthermore, let us denote the radiance values corresponding to these two calibration energy levels within a specific band channel as $E_{k_1}^{\mathrm{m}}$ and $E_{k_2}^{\mathrm{m}}$. Additionally, we assume that for these four calibration energy levels, the corresponding calibration data for our target pixel are represented by ${DN}_{m,n}^{k_1}$ and ${DN}_{m,n}^{k_2}$.

First, it is necessary to determine which segment of radiometric correction coefficient data corresponds to both the $m$-th band and $n$-th pixel where our data value resides.

If it is established that this value falls precisely within the $i$-th fitting segment (i.e., when ${DN}_{m,n}^{k_i-1}\le {DN}_{m,n,l}\le {DN}_{m,n}^{k_i}$), then it will be corrected using parameters $p_{k_i}^{m,n}$ and $q_{k_i}^{m,n}$ associated with this $i$-th fitting segment (hereafter referred to as $k_i$ for clarity). The relationship can be expressed through the following equation:

$$correctD{N}_{m,l,m}=p_{k_1}^{m,n}·D{N}_{m,l,n}+q_{k_1}^{m,n}$$

(9)

In the equation above, ${DN}_{m,n,l}$ denotes the original value of the $n$-th pixel in the $l$-th row within the $m$-th band channel prior to radiometric correction, while ${correctDN}_{m,n,l}$ signifies the value of the same pixel subsequent to radiometric correction.

If the data value of a pixel exceeds the calibration value at the maximum fitting point (i.e., ${DN}_{m,n,l}\ge {DN}_{m,n}^{k_i}$), it indicates that correction should be applied using the coefficients from the last fitting segment. In other words, this can be computed according to the following formula:

$$correctD{N}_{m,l,n}=p_{k_K}^{m,n}·D{N}_{m,l,n}+q_{k_K}^{m,n}=p_{k_4}^{m,n}\cdot D{N}_{m,l,n}+q_{k_4}^{m,n}$$

(10)

In the aforementioned formula, $p_{k_k}^{m,n}$ and $q_{k_k}^{m,n}$ denote the correction coefficients associated with the final fitting segment.

Through radiometric calibration, the pixel values in the image can be more accurately correlated with spectral intensity values, thereby obtaining more precise spectral information. Visually, the images typically display a more uniform brightness distribution, as shown in Figure 21.

To verify the system’s ability to distinguish different types of oil on the sea surface, an experiment was set up for validation. An experimental site with a 6 m × 6 m seawater pool was set up, and the simulated oil types for the experimental targets included fuel oil, crude oil, palm oil, diesel, and gasoline. The layout diagram of the oil types is shown in Figure 22.

A resolution board was placed beside the pool to verify the spatial resolution of the polarization spectral imaging system. The resolution plate is shown in Figure 23:

Using a UAV equipped with an integrated polarization spectral imaging system experimental platform, the UAV flew at a height of 80 m and employed a push-broom method to collect image information from the target area. As shown in Figure 24, the purple line represents the location area of a single shot, and the direction shown by the arrow is the flight direction of the UAV, that is, the direction of picture information acquisition.

After the test flight, the airborne hyperspectral polarimetric imaging system successfully acquired 250 spectral channels of data, as shown in Figure 25. The wavelength range of these images is 410–900 nm, with a sampling interval of 2 nm, resulting in the acquired spectral data. Through polarization calculation, 250 hyperspectral polarization images were generated, as shown in Figure 26. The stable platform operated normally, and all images were highly matched in the temporal and spatial domains as hyperspectral polarization images without the need for spatial registration.

As shown in Figure 27, spectral images of the resolution target were extracted at wavelengths of 420 nm, 650 nm, and 880 nm, respectively.

The figure clearly shows a set of lines, which is the 4# resolution target. The width “a” of one set of lines is 18 mm. According to the formula for calculating spatial resolution, the spatial resolution of this system is better than 0.18 mrad.

$$0.18\mathrm{m}\mathrm{r}\mathrm{a}\mathrm{d}={\displaystyle \frac{a}{80\mathrm{m}}}$$

(11)

The acquisition of hyperspectral images with time-domain and spatial coincidence was achieved, as shown in Figure 28.

Utilizing the reflectance spectra of crude oil, fuel oil, diesel, palm oil, gasoline, and seawater, we derived the average remote sensing reflectance for these five representative oil types alongside seawater. The visible light bands exhibiting higher discrimination capabilities are identified as 435 nm to 470 nm, 550 nm to 570 nm, and 630 nm to 650 nm. Polarized images and spectral images within these specified bands were selected for fusion analysis, as illustrated in Figure 29.

By comparing the spectral images and degree of polarization images within the aforementioned bands, this study identifies 462 nm, 556 nm, and 646 nm as optimal wavelengths that exhibit lower noise levels for oil spill detection. Noise in these images may originate from factors such as particulate suspensions on the seawater surface, random surface fluctuations, and inherent noise associated with the polarimetric imaging sensor. The selection of these bands enhances the signal-to-noise ratio for oil spill detection, thereby facilitating more accurate identification of oil slick distribution on the ocean surface.

The spectral images, polarimetric images, and fused images at wavelengths of 462 nm, 556 nm, and 646 nm are presented in Figure 30. Subsequently, these three fused images underwent true-color restoration followed by a 2% color stretch to achieve the final fusion result depicted in Figure 31.

The integration of hyperspectral and polarization data yields a significant enhancement, allowing crude oil contamination zones that were previously obscured against the ocean backdrop to be distinctly delineated with sharper boundaries and improved discernibility in the fused imagery. For fuel oils, even the faintest and most widely dispersed oil films can now be identified with greater precision, thereby expediting response efforts and enhancing efficiency in managing such spillage scenarios. In true-color representations, gasoline spills are marked by more pronounced boundaries, effectively overcoming challenges posed by its lightweight nature and rapid evaporation that have traditionally impeded capture by conventional optical imaging methods. This capability enables emergency personnel to swiftly locate gasoline leaks and initiate timely mitigation measures. Equally noteworthy is the enhanced contrast for diesel, which significantly aids in distinguishing minute diesel contamination on water surfaces, ensuring comprehensive monitoring coverage while minimizing the risk of overlooking potential environmental hazards. Lastly, palm oil is prominently featured in the fused imagery, indicating that even amidst extensive seawater mixing and other confounding factors, contamination zones can be efficiently distinguished, underscoring the technique’s effectiveness in identifying palm oil spills.

The comprehensive evaluation of the true-color imagery, as outlined in

Table 5. Evaluation index parameters of true-color images.

Evaluation Indicator	MI	Qab/f
Fused Image
Raw true-color image	1.144	0.045
True-color image fused with polarization spectrum	1.563	0.325

, employs the Mutual Information (MI) metric to quantify the degree of information integration between the source images and the fused imagery. A higher MI score indicates a more effective retention of information from the source images within the fused product, signifying an enhancement in fusion quality. The data presented in the table reveal a significant 36% increase in MI value for the polarized spectral fused true-color image compared to its original counterpart, highlighting the method’s efficacy in integrating information from both source images. Moreover, the $Q^{ab/f}$ value of 0.325 for the polarized spectral fused true-color image markedly surpasses that of its original at a modest 0.045, clearly illustrating the substantial improvement provided by polarized spectral fusion techniques regarding application-specific performance. In summary, based on these two critical evaluation indices, true-color images generated through polarized spectral fusion technology demonstrate superior capabilities in preserving source image information and enhancing overall image quality within targeted application domains. This advancement strengthens the analytical capacity for oil spill assessments in complex marine environments, ultimately improving efficiency and precision in marine oil spill monitoring efforts.

Through comprehensive field evaluations, the airborne hyperspectral polarization imaging system has exhibited exceptional capabilities, achieving a ground pixel resolution of 0.46 m at an operational altitude of 2 km. It operates within a wavelength range of 0.4 to 0.9 μm and features a spectral resolution of 2 nm across 250 spectral channels, capturing high-spectral polarization images in four distinct linear polarization states. This system can acquire spectral cube data, wherein the hyperspectral and polarization information are intrinsically aligned without necessitating additional registration processes. By integrating hyperspectral and polarization imagery, the system significantly enhances accuracy in differentiating oil spills from water surfaces. This substantiates the practicality of the proposed split-focal-plane hyperspectral imaging technology, thereby providing researchers with an innovative methodology for data acquisition in optical remote sensing.

6. Discussion

Compared with other polarized imaging systems, the design of a slit polarization imaging spectrometer proposed in this paper features a simple structure without moving parts, strong environmental adaptability, and the ability to simultaneously acquire polarization and spectral information, achieving high spectral resolution, high spatial resolution, and real-time multi-polarization angle imaging capabilities. We designed the optical system structure, analyzed the signal-to-noise ratio, and conducted tolerance analysis, providing a basis for mechanical mechanism design and system assembly. The system was mounted on a UAV and used for detecting marine oil spills. The images obtained from the detection were processed, resulting in polarization images, spectral images, and their fused images. In the fused images, the oil-polluted areas against the marine background showed clear boundaries, and visibility was improved. This preliminary research has validated the feasibility of applying the proposed polarization imaging spectrometer in the field of marine oil spill monitoring.

7. Conclusions

An advanced integrated polarization spectral imaging system designed to simultaneously acquire hyperspectral and polarization data from target objects was developed. Operating within a wavelength range of 410–900 nm, the system features a field of view (2ω) of 8.2°, provides no fewer than 250 spectral channels, maintains a spectral resolution of 2 nm, and ensures spatial resolution superior to 0.21 mrad. Through comprehensive field testing, we successfully extracted spectral and polarization information from various types of oil in seawater, generating extensive datasets comprising hyperspectral and polarization images. Notably, the results indicate that oil pollution zones—often difficult to delineate against the marine backdrop—exhibit distinct boundaries and enhanced visibility in the fused imagery. This accomplishment highlights the practical significance of our system in rapidly identifying large-scale oil spill incidents.