A method for underwater hyperspectral pushbroom image geometric correction

Abstract

This invention relates to the fields of computer vision and remote sensing image processing technology, and discloses a method for geometric correction of underwater hyperspectral pushbroom images. The method includes: deploying an underwater pushbroom platform equipped with adjustable support legs; using an acoustic ranging array to sense local terrain and automatically level it to ensure the optical axis of the hyperspectral imager is perpendicular to the seabed; simultaneously fusing data from the inertial measurement unit, depth sensor, and acoustic ranging to generate a six-degree-of-freedom pose time sequence; performing spatial inversion and affine transformation resampling on the line-by-line spectral image based on this sequence, stitching them together to generate a geometrically corrected hyperspectral data cube, and supplementing it with attitude angle-driven radiometric cosine correction. This invention improves the spatial consistency and radiometric accuracy of underwater hyperspectral images, and is suitable for fine-grained detection in complex seabed environments.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and remote sensing image processing, and specifically relates to a method for geometric correction of underwater hyperspectral pushbroom images. Background Technology

[0002] Hyperspectral imaging technology has been widely used in marine exploration in recent years due to its significant value in underwater target identification, ecological environment monitoring, and resource exploration. Pushbroom hyperspectral imaging constructs a three-dimensional data cube line by line scanning, and its imaging quality is highly dependent on the motion stability and attitude accuracy of the observation platform. However, the underwater environment is complex and variable, and the platform is susceptible to water flow disturbances, topographic undulations, and its own propulsion system during operation, leading to attitude shifts and positional drifts. This, in turn, causes geometric misalignment between image frames, seriously affecting subsequent image stitching, spectral consistency analysis, and ground feature interpretation.

[0003] Pushbroom underwater hyperspectral imaging mainly relies on two types of observation platforms: one is to deploy a fixed guide rail on the seabed to achieve uniform linear motion of the spectrometer. Although this can ensure the geometric stability of the imaging, the deployment cost is high, the applicable range is limited, and it is difficult to implement in complex seabed environments due to the uneven seabed topography. The other is to mount the hyperspectral meter on an underwater robot for autonomous pushbrooming. Although this improves the flexibility of operation, the robot has difficulty maintaining a constant attitude during hovering or movement. Especially when there are ocean currents or seabed disturbances, the roll and pitch angles fluctuate significantly, causing nonlinear distortion of the pushbroom trajectory and greatly increasing the difficulty of geometric correction.

[0004] Although existing technologies have attempted to reduce platform sway through self-scanning structures or landing-based frame robots, rigid support structures still struggle to adaptively level themselves when facing tilted or uneven seabed surfaces. This results in the hyperspectral imager being unable to maintain a vertical observation posture, making it difficult to guarantee image quality.

[0005] Existing technologies generally suffer from poor adaptability to seabed topography, insufficient platform attitude control precision, and limited geometric distortion correction capabilities. Particularly in unstructured seabed environments, traditional methods cannot predict topographic features before landing and lack mechanisms for real-time adjustment of support structures to maintain instrument verticality, resulting in severe geometric distortion and even loss of usability of pushbroom images. Therefore, there is an urgent need for a geometric correction method for underwater hyperspectral pushbroom images that integrates terrain perception, adaptive leveling, and high-precision attitude control to overcome the adverse effects of complex underwater environments on imaging stability and improve the geometric accuracy and application value of underwater hyperspectral data. Summary of the Invention

[0006] This invention provides a method for geometric correction of underwater hyperspectral pushbroom images, aiming to solve the problems of severe geometric distortion, degraded image quality, and difficulty in effective visual interpretation and subsequent data processing caused by the instability of underwater observation platform attitude and complex seabed topography in existing technologies. This invention constructs an underwater pushbroom platform system with adaptive topographic balancing capabilities, and combines multi-source sensor information fusion and a real-time attitude feedback control mechanism. While ensuring that the hyperspectral imager is always perpendicular to the seabed reference plane, it synchronously acquires high-precision position and attitude data, thereby achieving frame-by-frame geometric correction of the pushbroom image sequence, ultimately generating an underwater hyperspectral data cube with high spatial consistency and accurate geometric structure.

[0007] This invention provides a method for geometric correction of underwater hyperspectral pushbroom images, comprising:

[0008] Deploy an underwater push-broom platform, which includes a frame body, at least three independently telescopic and adjustable support legs, a push-broom hyperspectral imager, an inertial measurement unit, a depth sensor, an acoustic ranging array, a main controller, and a watertight power module.

[0009] After the underwater push-broom platform contacts the seabed, the acoustic ranging array is used to measure the local topographic height of the seabed below each support leg in real time, and obtain the vertical distance data of each support point relative to the bottom reference plane of the frame body.

[0010] Based on the vertical distance data, the main controller calculates the required extension and retraction of each support leg and drives each support leg to perform extension and retraction actions, so that the frame body is adjusted to a horizontal state and the optical axis of the pushbroom hyperspectral imager is perpendicular to the local seabed reference surface.

[0011] Before and during the push-broom hyperspectral imager begins its push-broom operation, the inertial measurement unit collects the platform's three-axis attitude angle data in real time, the depth sensor obtains the water depth of the platform, and the acoustic ranging array continuously monitors the relative height change between the platform and the seabed.

[0012] The three-axis attitude angle data, water depth data, and relative height change data are synchronized with the coordinate system to form a six-degree-of-freedom pose timing sequence of the platform during the push-broom process;

[0013] During the push-broom process, the push-broom hyperspectral imager acquires spectral image data line by line according to a preset line frequency, with each line of image corresponding to a timestamp;

[0014] For each row of spectral image data, spatial location inversion is performed to calculate the actual projected position of that row of image data in the global geographic coordinate system;

[0015] An inter-row geometric transformation model is constructed, and the original pushbroom image sequence is resampled and stitched row by row to generate a geometrically corrected hyperspectral image data cube.

[0016] Preferably, after the underwater push-broom platform contacts the seabed, the acoustic ranging array is used to measure the local seabed topography height below each support leg in real time, obtaining the vertical distance data of each support point relative to the bottom reference plane of the frame body, including:

[0017] The system emits sound beams downwards at a preset tilt angle using multiple miniature sonar probes, receives the echoes, and calculates the slant range based on the flight time.

[0018] By combining the sound velocity profile data with temperature and salinity compensation for the slant distance, the accurate sound wave propagation speed is obtained.

[0019] Based on the sound wave propagation speed, sound beam tilt angle, and installation height, the slant distance is converted into the vertical terrain height value at each support point.

[0020] Preferably, based on the vertical distance data, the main controller calculates the required extension / retraction amount for each support leg and drives each support leg to perform an extension / retraction action, thereby adjusting the frame body to a horizontal state, including:

[0021] Using the spatial coordinates of each support point and the corresponding vertical terrain height as input, the current local seabed reference surface is determined through a least squares plane fitting algorithm.

[0022] Calculate the ideal height that the ends of each support leg need to reach, so that the adjusted support points are coplanar and the normal vector is in the vertical direction;

[0023] Based on the difference between the ideal height and the actual height, the extension and retraction of each support leg is determined, and the hydraulic telescopic rod is driven to perform the leveling action through closed-loop control.

[0024] Preferably, the support leg includes a hydraulic telescopic rod, a servo motor, a displacement encoder, and an end pressure sensor;

[0025] The cylinder of the hydraulic telescopic rod is fixed at the bottom node of the frame body, and the piston rod end is provided with a ball universal joint.

[0026] The servo motor drives the hydraulic pump through a reduction gear set to adjust the extension and retraction stroke of the piston rod.

[0027] The displacement encoder provides real-time feedback of the telescopic displacement.

[0028] The end pressure sensor is used to detect whether the support leg is in stable contact with the seabed and to prevent overload.

[0029] Preferably, the three-axis attitude angle data, water depth data, and relative height change data are synchronized with the coordinate system to form a six-degree-of-freedom pose timing sequence of the platform during the push-broom process, including:

[0030] Temperature drift compensation and non-orthogonal error correction are performed on the raw angular velocity, acceleration and magnetic field strength data output by the inertial measurement unit;

[0031] The corrected data is input into an extended Kalman filter, which fuses the vertical position constraints provided by the depth sensor and the relative height information of the acoustic ranging array to calculate the roll angle, pitch angle and yaw angle.

[0032] The calculated attitude angles, the fused vertical position, and the horizontal position obtained from the velocity integral are unified into the optical center coordinate system of the imager to generate a six-degree-of-freedom pose timing sequence.

[0033] Preferably, the inertial measurement unit includes a three-axis gyroscope, a three-axis accelerometer, and a three-axis magnetometer;

[0034] The origin of the coordinate system of the six-degree-of-freedom pose timing sequence is set at the optical center of the pushbroom hyperspectral imager, and the coordinate axis directions are aligned with the coordinate system of the imager body.

[0035] Preferably, based on the six-degree-of-freedom pose time sequence, spatial location inversion is performed on each row of spectral image data to calculate the actual projected position of that row of images in the global geographic coordinate system, including:

[0036] Based on the timestamp of the row image, the corresponding pose is obtained by interpolation in the six-degree-of-freedom pose time sequence;

[0037] The image plane coordinates are converted into object-space direction vectors using the imager's focal length, and then combined with the pose rotation matrix to be converted into the global coordinate system;

[0038] By interpolating the seabed elevation using acoustic ranging data, the intersection points of the light rays and the seabed plane are calculated, resulting in a two-dimensional spatial location grid of all pixels in that row in the global geographic coordinate system.

[0039] Preferably, based on the spatial relationship between adjacent rows of images, an inter-row geometric transformation model is constructed, and the original pushbroom image sequence is resampled and stitched row by row to generate a geometrically corrected hyperspectral image data cube, including:

[0040] The spatial relationship between two adjacent rows of images is described by an affine transformation matrix. Its parameters are calculated from the pose difference between the two rows at the acquisition time, including translation components, rotation components and scale factors.

[0041] The subsequent rows of images are resampled using bilinear interpolation to align their spatial grid with the previous row.

[0042] Resampling and stitching are performed iteratively row by row to generate a three-dimensional hyperspectral data cube with a unified geographic reference frame.

[0043] Preferably, after generating the hyperspectral image data cube, a radiometric consistency correction step is further performed. This step performs cosine correction on the illumination intensity of each pixel based on the incident angle change caused by the platform attitude. The correction formula is:

[0044] ;

[0045] This is the original radiance value. The angle between the optical axis and the local vertical direction at the moment of imaging is calculated by combining the pitch and roll angles provided by the inertial measurement unit. This is the corrected radiance value.

[0046] Preferably, during the underwater push-broom platform's push-broom process, there are translational, rotational, and dimensional changes between adjacent rows, which are described using an affine transformation model:

[0047] ;

[0048] , They are respectively the first The x and y coordinates of the geographic coordinates of a pixel in a row. , The first The predicted x and y coordinates of the corresponding points in the row. For the first The line relative to the first Okay, in Scaling and rotation components in direction. For the first The line relative to the first OK, Directional motion pair The influence components of coordinates, For the first The line relative to the first OK, Directional motion pair The influence components of coordinates, For the first The line relative to the first Okay, in Scaling and rotation components in direction. For the first The line relative to the first Okay, in Pure translation in direction, For the first The line relative to the first Okay, in Pure translation in direction.

[0049] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0050] 1. This invention integrates independently retractable support legs and an acoustic terrain sensing system to achieve autonomous leveling of the underwater push-broom platform on complex seabed terrain, ensuring that the hyperspectral imager maintains a vertical attitude throughout the operation and eliminating geometric distortion of the push-broom image caused by platform tilt.

[0051] 2. At the same time, by integrating inertial measurement, depth and acoustic ranging data, a high-precision six-degree-of-freedom pose time sequence is constructed, providing accurate spatial positioning basis for each line of spectral image, so that the geometric correction process no longer depends on external positioning system or post-processing, thus improving correction accuracy and real-time performance.

[0052] 3. The introduction of a radiation cosine correction mechanism based on attitude angle further improves the radiation consistency of the data cube and ensures the reliability of subsequent quantitative analysis.

[0053] 4. This invention does not require seabed rails, does not rely on ROV hovering stability, and is not limited by local seabed undulations. It combines high adaptability, high stability, and high imaging quality, making it suitable for high-spectral precision detection tasks in complex underwater scenes such as coral reefs, rocky areas, and shipwreck sites. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0055] Figure 2 This is a schematic diagram of the core principle framework of the present invention based on multi-source sensor fusion and adaptive terrain leveling;

[0056] Figure 3 This is a logical flowchart of the terrain perception and autonomous leveling stage of the underwater push-broom platform in this invention.

[0057] Figure 4 This is a logical flowchart of the multi-source pose data synchronous fusion and six-degree-of-freedom pose temporal sequence construction stage in this invention.

[0058] Figure 5 This is a flowchart illustrating the logical flow of the pushbroom image inter-row geometric inversion and line-by-line resampling correction stages in this invention.

[0059] Figure 6 This is a schematic diagram of the multi-level interaction relationship and data flow between the underwater push-broom platform and the external environment in this invention. Detailed Implementation

[0060] refer to Figures 1 to 6 This invention provides a method for geometric correction of underwater hyperspectral pushbroom images. The core of this method lies in using an underwater pushbroom platform system with adaptive terrain balancing capabilities to achieve real-time stabilization and vertical alignment of the hyperspectral imager in complex seabed environments. Based on this, multi-source sensor information is fused to construct a six-degree-of-freedom pose time-series sequence. This sequence is then used to perform precise spatial position inversion and geometric correction on the spectral images acquired line by line during the pushbroom process, ultimately generating a hyperspectral data cube with high spatial consistency, accurate geometric structure, and consistent radiometric characteristics. The specific implementation of this invention will be described in detail below with reference to method steps S1 to S8.

[0061] The method includes the following steps:

[0062] S1, deploy an underwater push-broom platform;

[0063] S2, after the underwater push-broom platform contacts the seabed, the local topographic height of the seabed below each support leg is measured in real time using an acoustic ranging array;

[0064] S3, based on the vertical distance data, the main controller calculates the required extension and retraction amount of each support leg and drives each support leg to perform the extension and retraction action;

[0065] S4, before and during the push-broom hyperspectral imager starts its push-broom operation, collects the platform's three-axis attitude angle data in real time through the inertial measurement unit, obtains the water depth of the platform through the depth sensor, and continuously monitors the relative height change between the platform and the seabed through the acoustic ranging array.

[0066] S5, synchronize the three-axis attitude angle data, water depth data and relative height change data with the coordinate system to form a six-degree-of-freedom pose timing sequence of the platform during the push-broom process;

[0067] S6, During the push-broom process, the push-broom hyperspectral imager acquires spectral image data line by line according to the preset line frequency, and each line of image corresponds to a timestamp;

[0068] S7. Based on the six-degree-of-freedom pose time sequence, perform spatial location inversion on each row of spectral image data and calculate the actual projection position of the row of images in the global geographic coordinate system.

[0069] S8, based on the spatial relationship between adjacent rows of images, constructs an inter-row geometric transformation model, performs row-by-row resampling and stitching on the original pushbroom image sequence, and generates a geometrically corrected hyperspectral image data cube.

[0070] In step S1, the deployed underwater pushbroom platform serves as the physical carrier for implementing this method. Its structural integrity and functional completeness directly determine the accuracy and reliability of subsequent geometric correction. The underwater pushbroom platform includes a frame body, at least three independently telescopic and adjustable support legs, a pushbroom hyperspectral imager, an inertial measurement unit, a depth sensor, an acoustic ranging array, a main controller, and a watertight power supply module.

[0071] The main frame is made of high-strength aluminum alloy and has a regular hexagonal layout. Support legs are mounted at the six vertices, and the central area is the equipment integration platform for fixing the pushbroom hyperspectral imager and other sensing modules.

[0072] The main frame structure is designed to meet the hydrostatic pressure requirements at a water depth of 300 meters. The wall thickness has been optimized through finite element analysis to ensure the airtightness and rigidity of the internal equipment compartment while withstanding external water pressure. The center of gravity is located 10 centimeters directly below the geometric center to enhance the platform's static stability after placement on the seabed and prevent capsizing due to water flow disturbances or sudden changes in local topography.

[0073] The pushbroom hyperspectral imager is fixed directly below the center of the main frame. Its optical axis is initially set vertically downward. The spectral range covers 400 to 1000 nanometers, with a spectral resolution of 5 nanometers and a spatial resolution of 60 line pairs per millimeter. The line frequency can be dynamically adjusted within the range of 10 Hz to 100 Hz.

[0074] The front of the imager is equipped with a sapphire waterproof optical window and an integrated automatic cleaning brush driven by a miniature DC motor. It runs every 5 minutes to remove attached plankton or sediment, ensuring the clarity of the optical path.

[0075] The watertight power module is encapsulated in a titanium alloy sealed chamber, containing a 48-volt lithium-ion battery pack with a capacity of 25 amp-hours, supporting continuous operation for more than 10 hours, and supplying power to each subsystem through waterproof connectors.

[0076] The main controller uses an embedded multi-core processor, runs a real-time operating system, and has a gigabit Ethernet interface for high-speed image data transmission. It is also equipped with multiple high-speed serial communication buses, which are connected to the inertial measurement unit, depth sensor, acoustic ranging array and servo drivers of each support leg, to achieve millisecond-level command response and data acquisition.

[0077] In step S2, the acoustic ranging array is activated and performs a seabed local topography sensing task. The acoustic ranging array consists of six miniature sonar probes, which are evenly distributed along the bottom edge of the main frame. Each probe emits a narrow beam of sound waves downward at a 45-degree angle.

[0078] The speed of sound propagation in water is affected by temperature, salinity, and pressure. Therefore, the system has a built-in sound velocity profile database, which can call the corresponding sound velocity value for time-of-flight compensation based on real-time water depth data fed back by the depth sensor. Each sonar probe operates independently, transmitting at a frequency of 50 kHz with a pulse width of 0.5 milliseconds. After receiving the echo, the arrival time of the echo is determined by a threshold detection method, and the distance to the seabed is calculated by combining the sound velocity. This distance value is expressed as the slant distance from the sonar probe installation point along the sound beam direction to the intersection point with the seabed, and needs to be converted into vertical height using trigonometric relationships.

[0079] Specifically, let the sonar probe be installed at a height of... (Relative to the bottom reference plane of the main frame) Beam tilt angle (45 degrees), the slope distance was measured. The local seabed height The six probes correspond to the six support leg areas, thus obtaining six local terrain height values, which form an initial terrain height vector. . These are six local seabed height values. This process begins immediately after the platform touches the bottom, with a sampling period of 200 milliseconds, lasting for 3 seconds to obtain stable readings. The median of the three measurements is then taken as the final height value for each point to suppress transient noise interference.

[0080] In step S3, the main controller performs autonomous leveling of the platform based on the terrain height vector obtained in step S2. The leveling objective is to ensure that the main frame is in a horizontal position, i.e., the optical axis of the pushbroom hyperspectral imager is strictly perpendicular to the local seabed reference plane. To achieve this, the deviation of the main frame's current attitude from the ideal horizontal plane is first calculated.

[0081] The ideal contact points at the ends of the six support legs should be coplanar and parallel to the horizontal plane, but the measured height vectors indicate height differences between these points. The main controller employs a least-squares plane fitting algorithm, using the spatial coordinates of the six measurement points... For input, These are the horizontal and vertical coordinates of the measurement point. This is a measured local seabed height. The best-fit plane equation is then derived. . , , These are the plane parameters obtained through fitting. Represents the corresponding plane on the fitting plane The height of the position. The normal vector of the plane. This represents the direction of the current local seabed reference plane. To level the main frame, the lengths of each support leg need to be adjusted so that the adjusted six support points are coplanar and their normal vectors are [0,0,1]. The required extension / retraction of each support leg can then be calculated. ,in For the first The angle between the axis of each support leg and the vertical direction, due to the vertical installation of the support legs, Therefore The main controller will Send to the servo driver of the corresponding support leg.

[0082] Each support leg includes a hydraulic telescopic rod, a servo motor, a displacement encoder, and an end pressure sensor. The hydraulic telescopic rod cylinder is fixed to the bottom node of the main frame, and the piston rod end is equipped with a spherical universal joint, allowing for slight angular deflection without generating lateral force when in contact with uneven seabed. The servo motor drives a miniature hydraulic pump through a planetary reduction gear set, controlling the bidirectional flow of hydraulic oil to extend or retract the piston rod.

[0083] The displacement encoder is a magnetostrictive type with a resolution better than 0.1 mm, providing real-time feedback of the piston rod displacement to the main controller to form a closed-loop control. The end pressure sensor is a piezoelectric thin-film sensor embedded in the contact surface of the piston rod end. When the detected contact pressure exceeds a preset threshold (e.g., 500 Newtons), the extension and retraction action immediately stops to prevent overload damage to the equipment or getting stuck in a soft deposit layer. The leveling process adopts a staged strategy: the first stage completes coarse adjustment at a relatively fast speed, and the second stage performs fine adjustment at a low speed until the error is less than 0.5 mm. The entire process takes less than 30 seconds.

[0084] In step S4, after the platform completes leveling, it enters the push-broom preparation phase. The multi-source sensing system is simultaneously activated, continuously acquiring platform attitude-related data. The inertial measurement unit outputs raw data at a frequency of 250 Hz, including the angular velocity of the three-axis gyroscope, the specific force of the three-axis accelerometer, and the magnetic field strength of the three-axis magnetometer. The depth sensor is a piezoresistive pressure sensor with a range of 0 to 300 meters, an accuracy of ±0.05 meters, and a sampling frequency of 10 Hz. The acoustic ranging array switches to tracking mode, retaining only four symmetrically distributed sonar probes in operation, continuously measuring the average relative height from the platform bottom to the seabed at a frequency of 20 Hz. All sensor data is hardware-timestamped and synchronized by the real-time clock of the main controller, with a time reference accuracy better than 10 microseconds.

[0085] The raw data from the inertial measurement unit (IMU) is first compensated for temperature drift and corrected for non-orthogonal errors, then input into an extended Kalman filter for multi-sensor fusion. The filter state vector contains quaternions for position, velocity, and attitude, as well as sensor bias terms. The observation model fuses the gravity component from the accelerometer, the geomagnetic direction from the magnetometer, and the vertical position constraints provided by the depth sensor. The filter outputs are the platform's roll, pitch, and yaw angles in the northeast-northeast coordinate system, updated at a frequency of 100 Hz. Depth data is converted to geodetic elevation after sound velocity compensation, and acoustic ranging data is used to correct the cumulative error of inertial navigation in the vertical direction, resulting in a high-precision vertical position estimate.

[0086] In step S5, the main controller performs spatiotemporal alignment and coordinate transformation on the multi-source sensor data to construct a six-DOF pose timing sequence. First, all sensor data are interpolated and aligned according to hardware timestamps to a 100Hz time grid. Second, the attitude angles calculated by the inertial measurement unit are converted into rotation matrices, and then combined with the vertical position obtained by fusing depth and acoustic ranging. and the horizontal position obtained by velocity integral. , This constitutes a complete pose description. The origin of the pose time sequence coordinate system is set at the optical center of the pushbroom hyperspectral imager, and its body coordinate system... The axis points in the direction of flight. The axis points to the right. The axis is vertically downward. To transform the pose from the platform coordinate system to the imager coordinate system, a fixed-mount offset matrix needs to be applied. This matrix, determined during factory calibration, includes translation vectors and rotation matrices. Ultimately, the six-DOF pose at each moment is represented as follows: ,in ( These are coordinate components in the global geographic coordinate system. (Transposed) represents the location of the optical center in the global geographic coordinate system. This is the rotation matrix from the imager coordinate system to the global coordinate system. The pose timing sequence is stored in a circular buffer with a capacity of 10 seconds of data to ensure accurate pose at the corresponding moment despite image processing delays.

[0087] In step S6, the pushbroom hyperspectral imager begins acquiring spectral images at a preset line frequency. The line frequency is set according to water depth, lighting conditions, and mission requirements, with a typical value of 50 Hz. For each line of images acquired, the imager triggers a hardware interrupt, sending a line synchronization signal to the main controller. The main controller immediately records the current system time as the timestamp for that line of images. Image data is transmitted in real-time to a host computer or local solid-state storage via gigabit Ethernet. Each line contains 1000 spatial pixels, each pixel corresponding to 200 spectral channels, and the data format is a 16-bit unsigned integer. The timestamp accuracy is better than 100 microseconds, ensuring strict correspondence with the pose and timing sequence.

[0088] In step S7, for each row of images, the system performs spatial location inversion calculation. Given the timestamp of a row of images, the corresponding pose is obtained through linear interpolation in the pose temporal sequence. . This is the rotation matrix from the imager coordinate system to the global coordinate system. The geometric model of pushbroom imaging assumes that each row of the image lies in a plane perpendicular to the optical axis at the instant of exposure. Let the imager focal length be... Image plane coordinates ( , The corresponding object-space direction vector In the imager coordinate system Transform this vector to the global coordinate system. Since the imaging target is located on the seabed, the seabed elevation... It can be obtained by interpolation from acoustic ranging data. Let the optical center position be... , ( (The object point is the coordinate component in the global coordinate system) satisfy: , for The object-space direction vector after transformation to the global coordinate system. of Component equals Solve for the scaling factor. ; yes of Quantity.

[0089] Substituting, we get:

[0090] ;

[0091] yes of Quantity, yes of Quantity.

[0092] Repeat this calculation for all pixels in a row to obtain a two-dimensional spatial location grid of the image in the global geographic coordinate system, with a resolution of sub-meter level and a typical value of five centimeters.

[0093] In step S8, a geometric transformation model between rows is constructed based on the position grid of adjacent rows. During the underwater push-broom platform's push-broom process, adjacent rows undergo translation, rotation, and scale changes, which are described using an affine transformation model:

[0094] ;

[0095] , They are respectively the first The x and y coordinates of the geographic coordinates of a pixel in a row. , The first The predicted x and y coordinates of the corresponding points in the row. For the first The line relative to the first Okay, in Scaling and rotation components in direction. For the first The line relative to the first OK, Directional motion pair The influence components of coordinates, For the first The line relative to the first OK, Directional motion pair The influence components of coordinates, For the first The line relative to the first Okay, in Scaling and rotation components in direction. For the first The line relative to the first Okay, in Pure translation in direction, For the first The line relative to the first Okay, in Pure translation in direction.

[0096] After obtaining the transformation matrix, for the first... The original image is resampled using bilinear interpolation to make its spatial grid consistent with the first image. Row alignment. The resampling process maintains independent processing for each channel in the spectral dimension to ensure that spectral information does not overlap. This process is iterated row by row, starting from the first row and sequentially correcting subsequent rows to a unified geographic reference frame. Finally, the data is stitched together to generate a three-dimensional hyperspectral data cube with a spatial dimension of M×N and a spectral dimension of 200. Each spatial pixel has a precise geographic coordinate label.

[0097] After the initial calibration, the system further performs radiometric consistency correction. Due to potential residual attitude deviations on the platform, the angle between the optical axis and the local vertical direction at the imaging moment... This causes the intensity of incident light to change with the viewing angle. From roll angle With pitch angle Composite calculation: The raw radiance value for each pixel. Apply the cosine correction formula:

[0098] ;

[0099] The corrected radiance values ​​ensure that the radiance values ​​of all pixels are normalized to vertical observation conditions, improving the radiance consistency of the data cube and providing a reliable basis for subsequent quantitative inversion.

[0100] The underwater hyperspectral pushbroom image geometric correction system upon which the above method relies includes an underwater pushbroom platform, a terrain sensing and platform leveling module, a multi-source pose fusion module, an image acquisition and time stamping module, a geometric inversion and correction module, and a radiometric consistency correction module. As described above, the underwater pushbroom platform integrates all hardware components.

[0101] The terrain perception and platform leveling module consists of an acoustic ranging array, a support leg servo control system, and a leveling algorithm unit in the main controller, and is responsible for executing steps S2 and S3.

[0102] The multi-source pose fusion module includes an inertial measurement unit, a depth sensor, an acoustic ranging array data acquisition interface, and a Kalman filter fusion engine to implement steps S4 and S5.

[0103] The image acquisition and time stamping module consists of a pushbroom hyperspectral imager and its synchronous triggering circuit, and completes step S6.

[0104] The geometric inversion and correction module is a software module running on the main controller or host computer, which performs spatial inversion, affine transformation modeling and resampling stitching in steps S7 and S8.

[0105] The radiometric consistency correction module performs pixel-by-pixel radiometric correction on the data cube after geometric correction. All modules are interconnected via a high-speed data bus, forming a closed-loop control and processing pipeline to ensure automation and high precision throughout the entire process from platform deployment to data output.

[0106] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0107] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for geometric correction of underwater hyperspectral pushbroom images, characterized in that, include: Deploy an underwater push-broom platform, which includes a frame body, at least three independently telescopic and adjustable support legs, a push-broom hyperspectral imager, an inertial measurement unit, a depth sensor, an acoustic ranging array, a main controller, and a watertight power module. After the underwater push-broom platform contacts the seabed, the local topographic height of the seabed below each support leg is measured in real time by the acoustic ranging array to obtain the vertical distance data of each support point relative to the bottom reference plane of the frame body. Based on the vertical distance data, the main controller calculates the required extension and retraction of each support leg and drives each support leg to perform extension and retraction actions, so that the frame body is adjusted to a horizontal state and the optical axis of the pushbroom hyperspectral imager is perpendicular to the local seabed reference surface. Before and during the push-broom hyperspectral imager begins its push-broom operation, the inertial measurement unit collects the platform's three-axis attitude angle data in real time, the depth sensor obtains the water depth of the platform, and the acoustic ranging array continuously monitors the relative height change between the platform and the seabed. The three-axis attitude angle data, water depth data, and relative height change data are synchronized with the coordinate system to form a six-degree-of-freedom pose timing sequence of the platform during the push-broom process; During the push-broom process, the push-broom hyperspectral imager acquires spectral image data line by line according to a preset line frequency, with each line of image corresponding to a timestamp; For each row of spectral image data, spatial location inversion is performed to calculate the actual projected position of that row of image data in the global geographic coordinate system; An inter-row geometric transformation model is constructed, and the original pushbroom image sequence is resampled and stitched row by row to generate a geometrically corrected hyperspectral image data cube.

2. The method for geometric correction of underwater hyperspectral pushbroom images according to claim 1, characterized in that, After the underwater push-broom platform contacts the seabed, the acoustic ranging array is used to measure the local seabed topography height below each support leg in real time, obtaining the vertical distance data of each support point relative to the bottom reference plane of the frame body, including: The system emits sound beams downwards at a preset tilt angle using multiple miniature sonar probes, receives the echoes, and calculates the slant range based on the flight time. By combining the sound velocity profile data with temperature and salinity compensation for the slant distance, the accurate sound wave propagation speed is obtained. Based on the sound wave propagation speed, sound beam tilt angle, and installation height, the slant distance is converted into the vertical terrain height value at each support point.

3. The method for geometric correction of underwater hyperspectral pushbroom images according to claim 2, characterized in that, Based on the vertical distance data, the main controller calculates the required extension / retraction amount for each support leg and drives each support leg to perform extension / retraction actions, adjusting the main frame body to a horizontal state, including: Using the spatial coordinates of each support point and the corresponding vertical terrain height as input, the current local seabed reference surface is determined through a least squares plane fitting algorithm. Calculate the ideal height that the ends of each support leg need to reach, so that the adjusted support points are coplanar and the normal vector is in the vertical direction; Based on the difference between the ideal height and the actual height, the extension and retraction of each support leg is determined, and the hydraulic telescopic rod is driven to perform the leveling action through closed-loop control.

4. The method for geometric correction of underwater hyperspectral pushbroom images according to claim 3, characterized in that, The support leg includes a hydraulic telescopic rod, a servo motor, a displacement encoder, and an end pressure sensor; The cylinder of the hydraulic telescopic rod is fixed at the bottom node of the frame body, and the piston rod end is provided with a ball universal joint. The servo motor drives the hydraulic pump through a reduction gear set to adjust the extension and retraction stroke of the piston rod. The displacement encoder provides real-time feedback of the telescopic displacement. The end pressure sensor is used to detect whether the support leg is in stable contact with the seabed and to prevent overload.

5. The method for geometric correction of underwater hyperspectral pushbroom images according to claim 1, characterized in that, The three-axis attitude angle data, water depth data, and relative height change data are synchronized with coordinate system one to form a six-degree-of-freedom pose time sequence of the platform during the push-broom process, including: Temperature drift compensation and non-orthogonal error correction are performed on the raw angular velocity, acceleration and magnetic field strength data output by the inertial measurement unit; The corrected data is input into an extended Kalman filter, which fuses the vertical position constraints provided by the depth sensor and the relative height information of the acoustic ranging array to calculate the roll angle, pitch angle and yaw angle. The calculated attitude angles, the fused vertical position, and the horizontal position obtained from the velocity integral are unified into the optical center coordinate system of the imager to generate a six-degree-of-freedom pose timing sequence.

6. The method for geometric correction of underwater hyperspectral pushbroom images according to claim 5, characterized in that, The inertial measurement unit includes a three-axis gyroscope, a three-axis accelerometer, and a three-axis magnetometer; The origin of the coordinate system of the six-degree-of-freedom pose timing sequence is set at the optical center of the pushbroom hyperspectral imager, and the coordinate axis directions are aligned with the coordinate system of the imager body.

7. The method for geometric correction of underwater hyperspectral pushbroom images according to claim 6, characterized in that, Based on the six-degree-of-freedom pose time sequence, spatial location inversion is performed on each row of spectral image data to calculate the actual projected position of that row of images in the global geographic coordinate system, including: Based on the timestamp of the row image, the corresponding pose is obtained by interpolation in the six-degree-of-freedom pose time sequence; The image plane coordinates are converted into object-space direction vectors using the imager's focal length, and then combined with the pose rotation matrix to be converted into the global coordinate system; By interpolating the seabed elevation using acoustic ranging data, the intersection points of the light rays and the seabed plane are calculated, resulting in a two-dimensional spatial location grid of all pixels in that row in the global geographic coordinate system.

8. The method for geometric correction of underwater hyperspectral pushbroom images according to claim 7, characterized in that, Based on the spatial relationship between adjacent rows of images, an inter-row geometric transformation model is constructed. The original pushbroom image sequence is resampled and stitched row by row to generate a geometrically corrected hyperspectral image data cube, including: The spatial relationship between two adjacent rows of images is described by an affine transformation matrix. Its parameters are calculated from the pose difference between the two rows at the acquisition time, including translation components, rotation components and scale factors. The subsequent rows of images are resampled using bilinear interpolation to align their spatial grid with the previous row. Resampling and stitching are performed iteratively row by row to generate a three-dimensional hyperspectral data cube with a unified geographic reference frame.

9. The method for geometric correction of underwater hyperspectral pushbroom images according to claim 8, characterized in that, After generating the hyperspectral image data cube, a radiometric consistency correction step is further performed. This step performs cosine correction on the illumination intensity of each pixel based on the incident angle change caused by the platform attitude. The correction formula is as follows: ; This is the original radiance value. The angle between the optical axis and the local vertical direction at the moment of imaging is calculated by combining the pitch and roll angles provided by the inertial measurement unit. This is the corrected radiance value.

10. The method for geometric correction of underwater hyperspectral pushbroom images according to claim 9, characterized in that, During the sweeping process, the underwater pusher platform experiences translation, rotation, and dimensional changes between adjacent rows, which are described using an affine transformation model. ； , The first The x and y coordinates of the geographic coordinates of a pixel in a row. , The first The predicted x and y coordinates of the corresponding points in the row. For the first The line relative to the first Okay, in Scaling and rotation components in direction. For the first The line relative to the first OK, Directional motion pair The influence components of coordinates, For the first The line relative to the first OK, Directional motion pair The influence components of coordinates, For the first The line relative to the first Okay, in Scaling and rotation components in direction. For the first The line relative to the first Okay, in Pure translation in direction, For the first The line relative to the first Okay, in Pure translation in direction.

Images