Hyperspectral Imaging-Based Marine Oil Spills Remote Sensing System Design and Implementation

Abstract

Offshore drilling platforms leak hundreds of thousands of tons of oil every year causing immeasurable damage to the marine environment, therefore it is important to be able to monitor for oil leakage. A hyperspectral camera, as an advanced device integrating spectral technology and imaging technology, can keenly capture the differences in spectral reflectance of different types of oil and seawater. This study presents the design of a hyperspectral camera covering the 400 nm–900 nm spectral band (90 bands total) and establishes a monitoring system comprising a high-precision inertial navigation system, a stabilization system, and a data acquisition system. Furthermore, this study conducted a field flight experiment using a Cessna aircraft, acquiring hyperspectral data with a one m spatial resolution of a drilling platform around the South China sea at 3000 m altitude, which effectively delineated the spectral characteristics of the oil spill area. The detection system developed in this study provides a robust means for oil spill monitoring on drilling platforms in remote sensing of the marine environment.

1. Introduction

With the rapid development of the global economy, offshore oil development activities have grown increasingly frequent. As critical facilities for oil extraction, offshore drilling platforms not only make substantial contributions to energy supply but also bring serious oil pollution problems. According to statistics, the annual oil spills caused by offshore drilling platform operations reach hundreds of thousands of tons, causing immeasurable damage to the marine ecological environment. This oil pollution not only endangers the survival of marine organisms, but also exerts adverse impacts on related industries such as fisheries and tourism [1,2,3,4].

The traditional monitoring methods for offshore oil pollution mainly include ship monitoring, satellite remote sensing monitoring, and buoy monitoring. Ship monitoring is the on-site measurement of the marine environment by carrying various monitoring equipment on board [5,6,7]. Although this method can obtain relatively accurate data, it has problems such as limited monitoring range and low efficiency. The daily monitoring area of a monitoring ship usually does not exceed a few hundred square kilometers, which makes it difficult to meet the monitoring needs of large sea areas. Satellite remote sensing monitoring is the use of sensors carried by satellites to observe the surface of the ocean. It has the advantages of wide monitoring range and fast speed, but its resolution is low, making it difficult to accurately identify small-scale oil pollution events [8,9,10]. Some low-resolution satellite images cannot clearly distinguish small oil films and other floating objects on the sea surface. Buoy monitoring is the process of placing buoys in the ocean and monitoring real-time ocean environmental parameters through sensors [11,12,13]; however, the parameters monitored by buoys are relatively single, mainly focused on conventional parameters such as temperature and salinity, and have limited monitoring capabilities for oil pollution. The airborne remote sensing platform can flexibly carry multiple sensors to obtain multidimensional detection results. By controlling flight parameters, airborne sensors can obtain high-spatial, high-temporal, and high-resolution remote sensing data. These data can be used to achieve accurate observation of oil spills at sea. The airborne platform has a range of several hours to more than ten hours and can provide continuous remote sensing monitoring of oil spill areas and visual inspection of key areas. It has the characteristics of high flexibility and maneuverability, as shown in

Table 1. Comparison of different observation platforms.

Parameter	Ship Monitoring	Satellite Remote Sensing	Airborne Remote Sensing	Buoy Monitoring
Observation coverage	Low	High	High	Low
Continuous endurance	High	Low	High	High
Environmental adaptability	High	Medium	Medium	High
Cost	Low	High	Medium	Low
Resolution	High	Low	High	High
Sensor types	High	Low	High	Low
Flexibility	Low	Low	High	Low
Timeliness	Low	Low	High	Low

. When an ocean oil spill accident occurs, the aircraft can quickly reach the oil spill contaminated area for detection [14,15,16].

Faced with the increasingly severe problem of offshore oil pollution, the limitations of traditional monitoring methods are becoming more apparent, and there is an urgent need for a more efficient and accurate monitoring technology to meet practical needs. SAR (Synthetic Aperture Radar) imaging can work in adverse weather conditions, penetrate clouds, be sensitive to changes in oil film surface roughness, and cover large areas of the sea. But it is susceptible to interference from sea waves (missed detection at low wind speeds, false detection at high wind speeds), fails to distinguish oil film types (such as crude oil and emulsified oil), and has low spatial resolution [17]. Polarization imaging technology enhances the contrast between oil films and background seawater through polarization information and can distinguish smooth oil films from waves; it is also sensitive to thin oil film. However, it relies on lighting conditions (requiring sunlight), is susceptible to interference from solar flares, has complex data processing, and limited quantitative analysis capabilities [18]. Thermal imaging technology detects oil spills (oil films absorb heat during the day and dissipate heat at night) through the difference in thermal radiation between the oil film and seawater. It is suitable for night work, but its disadvantage is that it is greatly affected by environmental temperature, sunlight, and oil film thickness. Thick oil films may mask thermal signals and make it difficult to distinguish oil types [19]. Ultraviolet imaging technology is sensitive to thin oil films (due to the oil film fluorescence effect) and can detect trace oil films that are missed by other technologies. The disadvantage is that it strongly relies on solar ultraviolet radiation and is susceptible to weather interference [20].

As a new type of optical remote sensing equipment, hyperspectral cameras, have brought new hope for offshore oil pollution monitoring with their unique advantages. Hyperspectral imaging technology covers visible light, near-infrared and short-wave infrared ranges, accurately distinguishing oil film types (such as crude oil, diesel, emulsified oil) through spectral features. Based on the absorption/reflection characteristics of oil, false detections can be reduced. Spectral features reduce false alarms caused by environmental factors such as wind, waves, and lighting, which is superior to SAR and thermal imaging and suitable for scenarios that require refined management [21,22,23].

2. Airborne Hyperspectral Imaging System Methods

In order to achieve the effective detection of oil spills on drilling platforms, this study presents the development of a hyperspectral detection system covering 400–900 nm, along with a supporting attitude stabilization and data storage control system. The system was mounted on a Cessna aircraft platform and flew at an altitude of 3000 m to monitor oil spills in the drilling platform and surrounding waters, the whole system is shown in Figure 1.

The entire system mainly consists of a hyperspectral camera, a high-precision inertial navigation system, a stabilization system, and a data acquisition and control system, as shown in Figure 2. The hyperspectral camera is installed on a stabilization platform to ensure that the optical axis of the hyperspectral camera is always vertically downward during the scanning process. After integration, the hyperspectral camera, stabilization platform, and stabilization control system are installed in the optical mechanical structure and fixed on the Cessna aircraft platform to achieve ground observation through the belly window area. The high-precision inertial navigation and data acquisition control system are separately installed on the internal platform, and the whole system is powered by a high-capacity uninterruptible power supply.

2.1. Hyperspectral Camera Design

2.1.1. Optical Analysis

To reconcile the detection efficiency of hyperspectral cameras for marine oil spills with the fidelity of subsequent image processing, the designed system achieves a spatial resolution exceeding 1 m @ 3 km and the imaging width surpassing 2000 m @ 3 km.

For the design and development of hyperspectral cameras, detector selection is crucial, as it directly affects the imaging performance, volume, weight, and implementation difficulty of the camera. The selected detectors need to have advantages such as high frame rate, wide spectral range, high spectral response, and low readout noise. In this paper, the GSENSE400BSI from Gpixel is selected, which has the advantages of high dynamic range, low readout noise, and high quantum efficiency. GSENSE400BSI has been optimized for sensitivity at 275 nm, 420 nm, and 560 nm, with quantum efficiencies of 77% @ 275 nm, 86% @ 420 nm, and 95% @ 560 nm, as shown in Figure 3. The effective pixel is 2048 × 2048, and the pixel size is 11 µm × 11 µm. It can achieve up to 46 fps at full resolution and can meet the requirements of higher frame rates (300 frames) under windowing operation. The main technical parameters are shown in

Table 2. Main technical parameters of GSENSE400BSI.

Parameter	Index
effective pixel	2048 × 2048
pixel size	11 µm × 11 µm
frame rate	46 fps @ Standard mode
quantum efficiencies	77% @ 275 nm, 86% @ 420 nm, 95% @ 560 nm
dynamic range	>64 dB

.

According to the specifications, the pixel resolution should be better than 1 m and the imaging width should be more than 2000 m at an altitude of 3000 m, the indicators of the hyperspectral camera are analyzed and designed. The pixel resolution is specified as 0.98 m @ 3000 m, with a detector pixel size of 0.011 mm. Correspondingly, the total focal length of the optical system is 33.7 mm, the field of view of the hyperspectral camera is 37°, with a slit length of 22.7 mm and an imaging width of 2007 m, as shown in

Table 3. Designed parameters of the hyperspectral camera.

Parameter	Index
pixel resolution	0.98 m @ 3000 m
focal length	33.7 mm
field of view	37°
imaging width	2007 m @ 3000 m

.

2.1.2. Optical Design

Based on the aforementioned system analysis, the front mirror optical system design of the hyperspectral camera adopted an off-axis Schwarzchild two mirror system, mitigating energy loss caused by light obstruction via off-axis design. The off-axis angle was set to −25°, and both the primary and secondary mirrors were quadratic surfaces. Given the absence of a physical aperture stop in the system, a virtual aperture stop was positioned proximate to the back surface of the primary mirror to achieve image eccentricity. Two mirrors share a common optical axis, without any eccentric tilting elements, significantly simplifying system installation and calibration. During the optimization process, mirror spacing, curvature radius, and aspherical coefficients were employed as optimization variables. The optimized design result optical path is shown in Figure 4.

The spectral imaging system in this study employs an improved Dyson–Fery spectral imaging architecture. In this design, the Fery prism was used as a spectral dispersive element, and two Fery prisms were used to correct the nonlinearity of the spectrum. Concurrently, two spherical lenses were appended behind the Dyson lens to modulate the system’s optical power and augment the distance between the incident slit, image plane, and Dyson lens. During optimization, variables include the curvature radius, center thickness, air gap, and angle between the front/rear surfaces of the Fery prism and the optical axis for all lenses.

The final designed spectrometer structure is shown in Figure 5. The light enters the Dyson lens and two spherical mirrors through a slit, subsequently reaching two Fery prisms with opposite directions and different materials. Owing to the angle between the Fery prism’s front/rear surfaces and the optical axis, light of varying wavelengths disperses along the direction perpendicular to the slit. The aperture stop of the entire spectrometer is situated on the rear surface of the second Fery prism, which is fully coated with a reflective film. Following reflection, the light traverses the two Fery prisms, two spherical lenses, and the Dyson lens before converging on the image plane. At this juncture, the distance between the image plane and the Dyson lens is 32.9 mm, the distance between the incident slit and the Dyson lens is 36.9 mm, and the vertical distance between the image plane and the incident slit is 18 mm. This configuration provides ample space for the installation of the incident slit and detector, facilitating the integration and testing of the entire system.

The spectrometer and front mirror are cascaded with their respective incident slits to form the complete hyperspectral camera. The light emitted by the target object is converged by the front mirror and imaged at the slit. The light enters the spectrometer through the slit and is dispersed by the Fery prism until it reaches the image plane. The spatial dimensions of the hyperspectral imaging system are 315 mm × 95 mm × 95 mm. The optical path of the system is shown in Figure 6 and the three-dimensional structure is illustrated in Figure 7.

2.1.3. Optical Quality Analysis

The imaging quality of the integrated system is assessed using single wavelength modulation transfer function (MTF) and spot diagrams. From the spot diagrams of each monochromatic light and their corresponding MTF curves, the MTF values across the entire spectrum at a spatial sampling frequency of 46 lp/mm are approximately 0.5 across the full field of view, as shown in Figure 8, Figure 9 and Figure 10. The root mean square (RMS) diameter of the diffuse spots formed by the light passing through the optical system is within the range of 11 μm, as shown in Figure 11, Figure 12 and Figure 13, indicating that the geometric aberration has been substantially corrected and the imaging performance is excellent.

In addition to single-wavelength imaging performance, hyperspectral cameras must achieve high-fidelity spectral data consistency in push-broom imaging, specifically, the slit image for any wavelength must maintain linearity and parallelism with each pixel row. The slit image of any wavelength must maintain a straight line and be parallel to each row of pixels. In this design, the main rays of seven wavelengths of 400 nm, 500 nm, 600 nm, 700 nm, 800 nm, and 900 nm, as well as five fields of view (0,0), (0,0.3), (0,0.5), (0,0.7), and (0,1), were traced. Results indicated that the maximum spectral line bending at all bands and different field of view positions was less than two μm, accounting for less than 20% of the pixel size, thus satisfying the technical requirements of the spectral imaging system, as shown in Figure 14.

Furthermore, the dispersed spectrum of each target point in the slit field of view must maintain linearity and perpendicularity to the slit direction. Tracing the same seven wavelengths and five fields of view revealed that the maximum color distortion across all bands and field-of-view positions was less than 1.5 μm, also less than 20% of the pixel size, ensuring the accuracy of spectral retrieval, as shown in Figure 15.

2.2. Stable System Design

The hyperspectral camera in this study utilizes push-broom imaging, which necessitates a high-precision stabilization system. This system incorporates both pitch and roll control to compensate for the aircraft’s attitude deviations in these two axes, ensuring the camera’s field of view remains vertically downward.

The high-precision stable system comprises a dual-degree-of-freedom gyro-stabilized platform (roll and pitch). Both frames are directly driven by torque motors, with a gyroscope as the feedback element to sense the angular velocity of the frame and an encoder as the angular position sensor. Stabilization of the frame via dual-axis gyroscopes enables the system to maintain the camera’s vertical downward field of view.

The stable platform framework is outfitted with torque motors, photoelectric encoders, bearings, and bearings-electromechanical components that support the framework and enable rotation, ensuring stable target imaging within a defined detection range. The pitch and roll frame of the stable platform are equipped with mechanical limit blocks to constrain the hyperspectral camera’s movement to a limited angular range. Both frames have a mechanical limit angle of ±6°. In practice, to avoid rotating to the mechanical limits, the motion of the pitch and roll frame is controlled by software with an electrical limit of ±5°. The composition of the stable platform components is shown in Figure 16.

The gyroscope stabilization platform is a key component of the servo control subsystem, directly dictating equipment performance. The stability loop of the follow-up control subsystem primarily suppresses external disturbances to maintain the optical axis’s spatial stability.

The stabilization control subsystem enables functions such as the stabilization and scanning of the photoelectric system. The follow-up control subsystem isolates the carrier disturbance. The disturbance of the carrier will apply a disturbance torque to the stable platform, causing it to deflect. A high-precision gyroscope mounted on the stable platform detects this deflection and outputs an electrical signal corresponding to the deflection angle. The controller collects this deviation signal, processes it via algorithms, and outputs a control signal to the drive system, which actuates the motor to eliminate deflection, thereby suppressing the interference of the carrier disturbance on the aiming line and keeping the sensor in a stable state. The stable algorithm diagram is shown in Figure 17.

2.3. Control and Data Acquiation Design

In order to realize the hyperspectral image display and storage, this paper employs a high-performance reinforced industrial control computer MXC6401D as the data storage device. It features two hot-swappable hard drive slots and two Peripheral Component Interconnect Express (PCIE) interfaces and can accommodate two Camera Link acquisition boards to capture images from the hyperspectral camera. The Camera Link acquisition board utilizes DALSA’s MX4, supporting 1-channel base, medium, and full mode Camera Link image transmission with an interface clock of up to 80 MHz. In this system, one acquisition card handles image acquisition and storage for the hyperspectral camera.

Considering vibration during the flight of aerial vehicles, four shock absorbers were installed between the industrial computer and the mounting surface, and the three-dimensional mechanism model is shown in Figure 18. To mitigate damage to the industrial control computer from inherent aircraft vibration and ensure stable operation, the industrial computer employs a damping and shock-absorption mounting design.

The storage control and display software running on industrial computers mainly includes the following functions, and the diagram is shown in Figure 19:

(1)

Receive and store raw data from hyperspectral camera;

(2)

Display hyperspectral pseudo color images on the interface;

(3)

Switch spectral channels for hyperspectral pseudo color images;

(4)

Receive inertial navigation data and parse required information;

(5)

Control the hyperspectral camera’s frame rate based on real-time flight parameters;

(6)

Adjust hyperspectral camera parameters such as gain and exposure time.

The software interface is shown in Figure 20, including the image display area, command area, and parameter control area.

Additionally, the auxiliary system primarily provides GPS (Global Positioning System) coordinates and inertial navigation information for the detection system and is used for data storage and display. It mainly includes a differential GPS, inertial navigation system, and power supply.

The inertial navigation sensor used in this study is NovAtel’s SPAN-CPT, a compact, integrated Global Navigation Satellite System (GNSS) + Inertial Navigation System (INS) solution. Its built-in GNSS board utilizes NovAtel’s latest hardware platform, and the Inertial Measurement Unit (IMU) comprises a three-axis fiber optic gyroscope (FOG) and three-axis microelectromechanical system (MEMS) accelerometer, suitable for diverse operational environments. SPAN-CPT leverages NovAtel’s leading satellite navigation technology to achieve centimeter-level positioning accuracy; users can select positioning modes (single point, Satellite-Based Augmentation System [SBAS], L-band, or Real-Time Kinematic [RTK]) based on their requirements to ensure optimal accuracy. The physical configuration of the SPAN-CPT sensor is shown in Figure 21, and its technical specifications are summarized in

Table 4. Technical specifications of the SPAN-CPT sensor.

Inertial Navigation Sensor	Item	Specification
Position Accuracy	Horizontal	Single point: 1.2 m SBAS: 0.6 m DGPS: 0.4 m RTK: 1 cm
	Vertical	15 cm
Velocity Accuracy	Horizontal	0.02 m/s
	Vertical	0.01 m/s
Attitude Accuracy	Roll/Pitch	0.02°
	Heading	0.027°
3-axis Accelerometers	Full Scale Bias Bias stability	−10 g to 10 g 50 mg ±0.75 mg
3-axis Gyroscopes	Full scale Bias Bias stability	−375°/s to 375°/s 20°/h ±1°/h
Data rate	IMU Raw Data INS Solution	100 Hz 200 Hz
Mechanical	Size Weight	152 mm × 168 mm × 89 mm 2.29 kg
Max Power Consumption	-	13 W

.

2.4. System Integration Design

The aircraft platform is the Cessna 208, a single-engine turboprop multi-purpose light aircraft developed by Cessna Aircraft (USA), as shown in Figure 22. This series exhibits high reliability, economic efficiency, and flexibility, operates from simple runways, and has moderate commercial carrying capacity. The Cessna 208 has a maximum flight altitude of 25,000 feet, a maximum cruising speed of 164 mile/h, and an endurance of up to 6.6 h. Equipped with professional equipment, it is highly versatile.

Given that the Cessna aircraft is only equipped with openings in its fuselage without additional optical windows, an integrated design was implemented for the hyperspectral camera. Specifically, the hyperspectral camera, stabilization platform framework, AC/DC power conversion module, and stabilization platform controller were collectively integrated into the optical fuselage, the composition of the components is illustrated in Figure 23. The stabilization platform framework was mounted on the intermediate interlayer plate of the optical fuselage to prevent collision or interference between the hyperspectral camera and the window glass during rotation. The stabilization platform control board and AC/DC conversion module were installed on the inner wall of the optical fuselage. An opening was fabricated at the bottom of the optical fuselage to accommodate a protective window glass, which not only provides an imaging field of view for the hyperspectral camera but also safeguards its lens, as shown in Figure 24.

Based on the aircraft model, all the equipment was mounted on an adapter plate and fixed to two frame crossbeams of the aircraft. The optical fuselage, industrial computer, inertial navigation sensor, and uninterruptible power supply (UPS) were installed in a linear arrangement, the design of the installation scheme is shown in Figure 25.

Based on component specifications, the hyperspectral camera has a maximum power consumption of 7 W, the IMU 13 W, the control and data acquisition module 40 W, and the stabilization subsystem 30 W. Excluding minor auxiliary modules, the theoretical total system power is approximately 90 W. During actual testing with a 28 V external power supply, the stable current is ~three A and the peak current ~four A; thus, a 250 W AC-DC adapter is utilized for power supply.

3. Results

3.1. Airborne Hyperspectral Imaging Flight Experiment Introduction

Utilizing the aforementioned hyperspectral imaging system, we conducted actual flight tests in the South China sea. A Cessna aircraft equipped with the detection system took off from Boao Airport, Hainan, flying at an altitude of 3000 m, and conducted vertical detection of the offshore drilling platform and surrounding waters. It successfully acquired hyperspectral data with a one m spatial resolution of around the drilling platform, effectively delineating the spectral signatures of the oil leakage area. This experiment obtained data from two bands with a width of 2048 pixels (~2048 m), as shown in Figure 26.

3.2. Data Processing

To process and analyze marine oil spills, the acquired image data were extracted, and only images of offshore drilling platforms and their surrounding waters were selected to form an 800 × 800 data cube, which can be used to analyze the oil spill scenario. The formed data cube is shown in Figure 27.

Preliminary analysis of hyperspectral data from oil spills around the drilling platform, reveals distinct spectral responses between different oil spill areas and seawater, as shown in Figure 28. The white line represents the spectral curve of thicker oil spills, the red line represents the spectral curve of thinner oil spills, and the green line represents the spectral curve of the sea water.

To evaluate the performance of the hyperspectral camera and the integrated system, a detailed analysis of drilling platform images was conducted. As shown in Figure 29, two helicopter landing pads on the drilling platform are clearly discernible, indicating that the hyperspectral camera and system operated normally. The camera’s frame rate was well-matched to the aircraft parameters, with no significant geometric distortion in the images. The white line represents the spectral curve of oil spills, the red line represents the spectral curve of the drilling platform, and the green line represents the spectral curve of the sea water.

4. Discussion

To achieve effective detection of oil spills on drilling platforms, this study designed a hyperspectral camera covering the 400–900 nm spectral band with a total of 90 bands and developed a monitoring system comprising a high-precision inertial navigation system, stabilization system, and data acquisition and control system. Furthermore, we conducted a flight experiment using a Cessna aircraft, acquiring hyperspectral data with a one m spatial resolution around a drilling platform in the South China Sea at an altitude of 3000 m, which effectively demonstrated the spectral characteristics of the oil spill area. The developed detection system provides a robust means for oil spill monitoring on drilling platforms in remote sensing of the marine environmental.

5. Conclusions

The airborne hyperspectral imaging based marine oil spills remote sensing system developed in this study validated its practical detection efficacy for drilling platform oil spills in actual flights. By adjusting flight parameters, this system can acquire high-spatial-resolution and high-temporal-resolution hyperspectral remote sensing data for different target areas, enabling precise observation of offshore oil spills. In the event of a marine oil spill, aircraft equipped with this system can rapidly reach the contaminated area for detection, with a flight duration of several to over ten hours they provide continuous remote sensing monitoring of the oil spill area and visual inspection of key regions.

In future work, we plan to collaborate with oil drilling platform companies and marine ecological environment management departments, arrange flights to survey key areas to acquire additional datasets, and complete refined processing and situational assessments.