Unmanned aerial vehicle water conservancy exploration intelligent system and method based on multi-source data fusion

Abstract

The invention relates to the technical field of unmanned aerial vehicle water conservancy exploration, in particular to an unmanned aerial vehicle water conservancy exploration intelligent system and method based on multi-source data fusion, and the system comprises an image collection module which is used for collecting water surface information and outputting an original image signal through an airborne multispectral polarization imaging unit; and the image preprocessing module is used for executing geometric correction based on the original image signal so as to output a corrected image signal. According to the method, a fusion weight calculation mechanism is introduced among multiple spectral bands, and the target detection depth and the spectral signal-to-noise ratio model are combined, so that the adaptive weight fusion of the multi-source spectral data is realized, and the complementarity of information of different bands is effectively enhanced; the spectrum contrast ratio, the detail fidelity and the target prominence of the generated exploration image are obviously improved, and higher accuracy is provided for quantitative analysis and feature recognition of the water body environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle water conservancy survey, and particularly relates to an unmanned aerial vehicle water conservancy survey intelligent system and method based on multi-source data fusion. BACKGROUND

[0002] With the rapid development of unmanned aerial vehicle technology and optical remote sensing technology, unmanned aerial vehicles have been widely applied in the fields of water environment monitoring, resource survey and target identification in water conservancy survey. In view of the characteristics of large changes in illumination, many signal interferences and high difficulty in underwater information acquisition in complex water environment in water conservancy survey, an intelligent survey method fusing multi-source spectral and polarization data emerges as the times require. The method can realize high-precision perception and analysis of water surface and underwater environment on the unmanned aerial vehicle platform by integrating multi-spectral imaging, polarization detection, spectral inversion and data fusion and other multi-dimensional information processing technologies, and provides efficient and intelligent technical support for water resource investigation, ecological evaluation and environmental monitoring and other applications.

[0003] In the process of underwater target detection, due to the significant difference in absorption and scattering characteristics of different turbidity water bodies to each spectral band, the penetration depth of light of different bands in water is different, resulting in inconsistent effective utilization of imaging spectral information.

[0004] In the prior art, fixed or empirically set weight coefficients are usually used, which is difficult to adaptively adjust the fusion proportion of each band according to the change of water turbidity, so that the fusion result is unstable in different water environments, it is difficult to balance the spectral contrast and detection depth, and the accuracy of underwater target identification and the environmental adaptability of the system are affected. SUMMARY

[0005] In order to make up for the above shortcomings, the present application provides an unmanned aerial vehicle water conservancy survey intelligent system and method based on multi-source data fusion, aiming at improving the problem that in the prior art, fixed or empirically set weight coefficients are usually used, which is difficult to adaptively adjust the fusion proportion of each band according to the change of water turbidity, so that the fusion result is unstable in different water environments.

[0006] In the first aspect, the present application provides the following technical scheme, an unmanned aerial vehicle water conservancy survey intelligent system based on multi-source data fusion, comprising: An image acquisition module is configured to acquire water surface information and output an original image signal through an airborne multi-spectral polarization imaging unit; An image preprocessing module is configured to perform geometric correction based on the original image signal to output a corrected image signal; An optical parameter calculation module is configured to generate an effective water body off-water signal by processing the corrected image signal through a polarization decoupling model, and to output a spectral attenuation coefficient based on the effective water body off-water signal; a weight calculation module, configured to calculate and output a multispectral fusion weight based on the spectral attenuation coefficient, a preset target detection depth, and a spectral signal-to-noise ratio model; an image fusion module, configured to perform weighted fusion on the effective water-leaving signal based on the effective water-leaving signal and the multispectral fusion weight, and output an exploration image; a control module, configured to trigger the optical parameter calculation module and the weight calculation module to update according to a preset update condition during flight based on flight state and position information of the unmanned aerial vehicle.

[0007] Preferably, the water surface information is collected by the on-board multispectral polarization imaging unit and the original image signal is outputted, specifically including: a single exposure is performed on the target water area by the snapshot multispectral polarization imaging unit to instantaneously capture a mixed light signal containing target information and environmental interference; the mixed light signal is synchronously separated into a plurality of spectral channel data and a plurality of polarization analysis channel data at a pixel level space by using a micro-filter array and a micro-polarization array integrated in the focal plane of the imaging unit; the separated channel data is converted into digital electrical signals by the image sensor, and the original image signal is outputted in combination.

[0008] Preferably, the geometric correction is performed based on the original image signal, specifically including: flight state and position information of the unmanned aerial vehicle synchronized with the time when the original image signal is collected is acquired; a geometric correction model is established based on the flight state and position information and a preset camera intrinsic parameter; the original image signal is resampled by using the geometric correction model to generate and output a corrected image signal.

[0009] Preferably, the effective water-leaving signal is generated by processing the corrected image signal based on a polarization decoupling model, specifically including: the Stokes vector and the degree of polarization of each spectral channel are obtained by solving based on the multi-polarization state data in the corrected image signal by using the Stokes vector method; the water surface reflection interference component is estimated by using the polarization decoupling model based on the Stokes vector and the degree of polarization and the difference in polarization characteristics between the water surface reflection light and the water-leaving signal; the water surface reflection interference component is subtracted from the total light intensity signal corresponding to the corrected image signal to generate the effective water-leaving signal.

[0010] Preferably, the spectral attenuation coefficient is inversely outputted based on the effective water-leaving signal, specifically including: The effective water body's water separation signal is organized into a spectral feature vector; The spectral feature vector is input into a pre-trained inversion model, which is trained by machine learning and used to construct a nonlinear mapping relationship between the spectral morphology of the water body detachment signal and the spectral attenuation coefficient. The inversion model outputs the spectral attenuation coefficient.

[0011] Preferably, the step of calculating and outputting the multispectral fusion weights based on the spectral attenuation coefficient, the preset target detection depth, and the spectral signal-to-noise ratio model specifically includes: Obtain the vector composed of the spectral attenuation coefficients and receive the target detection depth parameters input from the outside; For each spectral channel, based on the spectral attenuation coefficient and the target detection depth parameter, the theoretical signal attenuation factor of that channel is calculated according to the exponential attenuation model with the spectral attenuation coefficient as the key parameter. The theoretical signal attenuation factors for all spectral channels are normalized to generate a set of weighting coefficients; The weighting coefficients are output as the multispectral fusion weights.

[0012] Preferably, the step of performing weighted fusion of the effective water body water-leaving signal and outputting the exploration image based on the effective water body water-leaving signal and the multispectral fusion weight specifically includes: Acquire image data of each spectral channel of the effective water body's water separation signal; The image data of each spectral channel are weighted using the multispectral fusion weights. The weighted image data from each channel are fused at the pixel level to generate an exploration image.

[0013] Preferably, triggering an update during flight based on preset update conditions specifically includes: The preset update condition is monitored, wherein the preset update condition is that the change in the spectral attenuation coefficient output by the optical parameter calculation module exceeds a preset threshold. When the preset update condition is met, an execution instruction is sent to the optical parameter calculation module and the weight calculation module. After receiving the execution command, the optical parameter calculation module and the weight calculation module recalculate and output the updated spectral attenuation coefficient and multispectral fusion weight based on the currently acquired and corrected image signal processed by the image preprocessing module.

[0014] Secondly, the present invention provides the following technical solution: a UAV-based water conservancy survey method based on multi-source data fusion, the method comprising the following steps: The airborne multispectral polarization imaging unit acquires water surface information to obtain raw image signals. Based on the original image signal, geometric correction is performed to generate a corrected image signal; Based on the corrected image signal, a polarization decoupling model is used to process and generate an effective water body water separation signal, and the spectral attenuation coefficient is output based on the effective water body water separation signal. Based on the spectral attenuation coefficient, the preset target detection depth, and the spectral signal-to-noise ratio model, the multispectral fusion weight is calculated. Based on the effective water body water-free signal and the multispectral fusion weights, weighted fusion is performed to generate exploration images; Based on the UAV's flight status and location information, the spectral attenuation coefficient and the multispectral fusion weight are recalculated during flight according to preset update conditions.

[0015] The present invention has the following beneficial effects: 1. In this invention, by introducing a fusion weighting calculation mechanism between multispectral bands and combining it with the target detection depth and spectral signal-to-noise ratio model, adaptive weighted fusion of multi-source spectral data is realized, which effectively enhances the complementarity of information in different bands. This results in a significant improvement in the spectral contrast, detail fidelity, and target prominence of the generated exploration images, providing higher accuracy for quantitative analysis and feature identification of the aquatic environment.

[0016] 2. In this invention, by introducing a technical approach that combines multispectral polarization imaging with optical inversion, high-precision separation and reconstruction of water surface and underwater environmental information are achieved. This effectively suppresses water surface glare interference, extracts real water body signals, and obtains water body optical characteristic parameters by combining quantitative inversion of spectral attenuation coefficients. This significantly improves the detection accuracy and identification capability of underwater targets and environment, ensuring that UAVs can still obtain clear and stable survey images under complex lighting and wave conditions.

[0017] 3. In this invention, by dynamically monitoring the flight status and position information during the flight of the UAV, an adaptive update mechanism for the spectral attenuation coefficient and fusion weight is set up. This mechanism can automatically adjust the imaging parameters when the ambient light, attitude or area changes, so as to maintain the timeliness and consistency of the survey results, thereby improving the system's environmental adaptability and the level of intelligence in mission execution in complex scenarios. Attached Figure Description

[0018] Figure 1 This is an architecture diagram of an intelligent unmanned aerial vehicle (UAV) water conservancy survey system based on multi-source data fusion proposed in this invention. Figure 2 This is a flowchart of a UAV-based water conservancy survey method based on multi-source data fusion proposed in this invention. Detailed Implementation

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1:

[0020] In the first embodiment of the present invention, the present invention provides an intelligent unmanned aerial vehicle (UAV) water conservancy survey system based on multi-source data fusion, such as... Figure 1 As shown, it includes: The image acquisition module is used to acquire water surface information and output raw image signals through an airborne multispectral polarization imaging unit; Furthermore, the acquisition of water surface information and output of raw image signals via an airborne multispectral polarization imaging unit specifically includes: A single exposure is performed on the target water area using a snapshot-type multispectral polarization imaging unit to instantly capture a mixed light signal containing target information and environmental interference; By utilizing microfilter arrays and micropolarizer arrays integrated on the focal plane of the imaging unit, the mixed optical signal is synchronously separated into multiple spectral channel data and multiple polarization analysis channel data in pixel-level space; The image sensor converts the separated channel data into digital electrical signals and combines them to output the original image signal.

[0021] Specifically, the image acquisition module is mounted on the payload platform of the UAV and is used to acquire multi-dimensional optical information of the target water area in real time during flight. The image acquisition module includes a snapshot-type multispectral polarization imaging unit, an attitude synchronization acquisition unit, and signal buffering and transmission circuitry. According to mission planning instructions, the UAV triggers the imaging unit to perform a single-exposure imaging operation at a predetermined flight position. The imaging unit adopts a snapshot design, with the exposure time... Simultaneously, it completes the acquisition of imaging data for all pixels, avoiding the time delay and displacement blurring problems of traditional pushbroom imaging, and ensuring that each frame of image corresponds to the optical state of the water surface at the same moment.

[0022] During exposure, the incident light from the target area consists of three parts: light reflected from the water surface, light scattered from the air-water interface, and the signal from the water leaving the water body. These together form a mixed light signal. The mixed light passes sequentially through a microfilter array and a micropolarizer array before entering the focal plane of the imaging unit. The microfilter array consists of multiple filter units with different center wavelengths, arranged periodically at the pixel level, allowing adjacent pixels to selectively respond to different wavelengths of light. The micropolarizer array is located on the same plane... The four directions are arranged alternately to separate light signals with different polarization directions. For any pixel... Its output light intensity can be expressed as: ; in, Indicates the first The center wavelength of each spectral channel Indicates the first Each polarization direction angle This indicates the light response intensity of the pixel at that wavelength and polarization angle. The imaging unit can acquire a value containing [the following information] after a single exposure. Each spectral channel and A two-dimensional data cube with polarization channels, whose data structure can be represented as follows: ; The data cube is converted into a digital signal output by a high dynamic range CMOS image sensor. The sensor employs pixel-by-pixel adaptive gain technology, automatically adjusting the amplification factor according to the incident light intensity. The output signal strength can be expressed as: ; in, The dynamic gain coefficient is used to reduce the risk of overexposure in areas of strong reflection and improve signal discernibility in dark areas. The output signal is quantized into a digital electrical signal by an on-chip analog-to-digital converter (ADC) and transmitted to the onboard processing unit via a high-speed cache circuit and data bus. The attitude synchronization acquisition unit simultaneously records the UAV's flight attitude, geographic coordinates, and timestamp information, and embeds a unified identifier in the output data to ensure data synchronization between multispectral and polarization channels.

[0023] Through this snapshot-type multispectral polarization imaging design, the system can simultaneously acquire high-dimensional optical data in both spectral and polarization dimensions in a single frame exposure, thereby enabling instantaneous optical sampling of water surface and shallow water bodies in dynamic flight environments.

[0024] By separating the spectral and polarization at the pixel level, spatial misalignment and temporal drift errors caused by UAV movement and water surface disturbance are reduced. This provides high signal-to-noise ratio and high stability basic data for subsequent geometric correction, polarization decoupling, and spectral inversion, significantly improving the imaging clarity and survey accuracy of UAVs in complex aquatic environments.

[0025] The image preprocessing module performs geometric correction based on the original image signal to output the corrected image signal; Furthermore, based on the original image signal, geometric correction specifically includes: Acquire UAV flight status and position information synchronized with the time of raw image signal acquisition; A geometric correction model is established based on flight status and position information, as well as preset camera intrinsic parameters; The original image signal is resampled using a geometric correction model to generate and output the corrected image signal.

[0026] Specifically, the image preprocessing module is used to process the raw image signal output by the image acquisition module. Perform geometric correction. The original image signal is a data cube containing multispectral and multipolarity channels, and its pixel coordinates... It exists in an uncorrected sensor focal plane coordinate system. Due to the airflow disturbances that cause attitude changes such as pitch, roll, and yaw during the flight of the UAV, as well as the constant changes in flight altitude and position, the raw image signal acquired contains geometric distortions caused by changes in viewpoint and attitude, such as tilt, scale changes, and pixel displacement.

[0027] To eliminate these distortions, the image preprocessing module first obtains the data from the airborne attitude synchronization acquisition unit at the same time as the acquisition of the original image signal. Fully synchronized drone flight status and position information. This information specifically includes the drone's three-dimensional position coordinates in the world coordinate system. For example, the WGS-84 coordinate system, and the three attitude angles, namely the roll angle. Pitch angle Yaw angle At the same time, the module retrieves the pre-calibrated camera intrinsic parameter matrix from local storage. and distortion coefficient .

[0028] Based on the above information, the image preprocessing module establishes a geometric correction model, which is based on the collinearity equation and describes the distance from the target water surface feature point. To sensor pixels The projection relationship between them can be expressed as: ; in: As a scale factor, For ideal, undistorted pixel coordinates, This is the camera intrinsic parameter matrix, which includes the focal length. and principal point coordinates These are the preset parameters. The rotation matrix from the world coordinate system to the camera coordinate system is derived from the attitude angles acquired in real time. Calculations show that The real-time location of the drone is obtained from GPS and INS systems. The three-dimensional coordinates of the target water surface point.

[0029] In the actual correction process, the image preprocessing module employs back projection resampling. First, a corrected image grid is defined, corresponding to a standard geographic reference coordinate system, such as the UTM projected coordinate system. For each pixel in the corrected image... The image preprocessing module first maps it to the corresponding land or water surface coordinates. And combined with the preset average water surface elevation Then, using the collinearity equations described above, the projected coordinates of the feature point on the original image sensor are calculated in reverse. The coordinates It is usually a non-integer.

[0030] Next, the module... Correction is performed using a preset lens distortion model, which uses distortion coefficients. The original pixel coordinates after distortion are obtained. Finally, the image preprocessing module employs a bicubic interpolation algorithm, based on non-integer coordinates. and surrounding pixels in the original image signal The value in the original data cube is used to calculate the precise signal strength at that point. This interpolation process must be performed across all spectral and polarization channels of the original data cube. The above-mentioned geometric correction process is performed synchronously to ensure pixel alignment accuracy between channels after correction. The corrected image signal is then output through this process. .

[0031] This effectively eliminates image geometric distortion caused by changes in drone attitude and terrain undulations. In this scenario, the distortion is mainly caused by water surface ripples and changes in flight altitude, converting the original sensor center projection image into an orthophoto image with a uniform scale and geographic reference.

[0032] The optical parameter calculation module, based on the corrected image signal, processes it through a polarization decoupling model to generate an effective water body water separation signal, and outputs the spectral attenuation coefficient based on the effective water body water separation signal. Furthermore, based on the corrected image signal, the generation of an effective water separation signal through polarization decoupling model processing specifically includes: Based on the multi-polarization state data in the corrected image signal, the Stokes vector method is used to calculate the Stokes vector and degree of polarization for each spectral channel. Based on the Stokes vector and degree of polarization, and taking advantage of the difference in polarization characteristics between the reflected light from the water surface and the signal leaving the water body, the interference component of the reflected light from the water surface is estimated through a polarization decoupling model. The water surface reflection interference component is subtracted from the total light intensity signal corresponding to the corrected image signal to generate an effective water body departure signal.

[0033] Furthermore, the spectral attenuation coefficient retrieved from the effective water body's water separation signal specifically includes: Organize the effective water body's water separation signal into a spectral feature vector; The spectral feature vector is input into a pre-trained inversion model, which is trained by machine learning and used to construct a nonlinear mapping relationship between the spectral morphology of the water body detachment signal and the spectral attenuation coefficient. The spectral attenuation coefficient is output from the inversion model.

[0034] Specifically, the optical parameter calculation module is used to process the corrected image signal output by the image preprocessing module in real time. The optical parameter calculation module first separates the water component from the mixed optical signal using a polarization decoupling model to remove glare interference from the water surface. Then, it infers the spectral attenuation coefficient of the water body based on this water component signal, thus providing key parameters for subsequent physical modeling and environmental analysis.

[0035] Polarization decoupling processing: For each pixel and its spectral channels in The module representing the band number extracts four polarization angles. The light intensity data is used to construct a Stokes vector to describe the polarization state of the light: ; ; ; in, These represent the polarization angles respectively. The measured value of light intensity at that time, Indicates total light intensity. and This represents the linearly polarized component, used to describe the polarization state of light. Glare reflected from water surfaces exhibits strong linear polarization characteristics, while the polarization degree of signals leaving the water is extremely low. Based on this physical characteristic, this module uses Fresnel's equations to calculate the theoretical degree of polarization of light reflected from the water surface. (in To observe the zenith angle and estimate the water surface reflection components: ; in This represents the estimated intensity of light reflected from the water surface. This represents the observed intensity amplitude of linearly polarized light.

[0036] Subtracting this component from the total light intensity yields the water body departure signal for glare reduction: ; in, Represents pixels In the band The water radiance after removing glare from the water surface.

[0037] Inversion of spectral attenuation coefficient: The optical parameter calculation module constructs spectral feature vectors from the water ionization signals of each spectral channel: ; in, for 3D column vector, The total number of spectral channels, indicated by the superscript. This indicates transpose.

[0038] This vector is input into a pre-trained inversion model. Output the spectral attenuation coefficients for each band: ; in, This represents an inversion model trained based on machine learning. This is the output spectral attenuation coefficient vector. For band The diffuse attenuation coefficient at that location, in units of This quantitatively describes the rate attenuation of light intensity with water depth.

[0039] Physical modeling using the Stokes vector method and Fresnel equations effectively separates water surface glare interference; a machine learning inversion model accurately retrieves the water's optical properties from spectral features. The final output is the water separation signal. With spectral attenuation coefficient This provides a key physical basis for subsequent weight calculation and image fusion, enabling the system to achieve comprehensive detection capabilities that are both depth-adaptive and environment-adaptive.

[0040] The weight calculation module calculates and outputs the multispectral fusion weights based on the spectral attenuation coefficient, the preset target detection depth, and the spectral signal-to-noise ratio model. Furthermore, based on the spectral attenuation coefficient, the preset target detection depth, and the spectral signal-to-noise ratio model, the multispectral fusion weights are calculated and output, specifically including: Obtain the vector composed of spectral attenuation coefficients and receive the target detection depth parameters from external input; For each spectral channel, based on the spectral attenuation coefficient and the target detection depth parameter, the theoretical signal attenuation factor of that channel is calculated according to the exponential attenuation model with the spectral attenuation coefficient as the key parameter. The theoretical signal attenuation factors of all spectral channels are normalized to generate a set of weighting coefficients; The output weighting coefficients are used as the weights for multispectral fusion.

[0041] Specifically, the weight calculation module outputs the spectral attenuation coefficient vector from the optical parameter calculation module. This module is then activated to calculate the weighting coefficients required for multispectral image fusion. It determines the contribution ratio of each spectral channel in the fusion process based on the optical attenuation characteristics of water and the user-defined exploration depth requirements, thereby enabling adaptive exploration in different water depth environments. The weighting calculation module receives the spectral attenuation coefficient vector output by the optical parameter calculation module. ; in, Indicates the first Each spectral channel at wavelength The diffuse attenuation coefficient is given by the following unit: This reflects the attenuation rate of light as it passes through water in that wavelength band. This represents the total number of multispectral channels. Simultaneously, the weight calculation module receives external input or system-preset target detection depth parameters. The unit is This parameter is set by the user according to the actual exploration task requirements and is used to indicate the depth of the water body that the light signal needs to pass through.

[0042] For each spectral channel The weight calculation module is based on the spectral attenuation coefficient. and target detection depth Based on the exponential decay model of light propagation in water, the theoretical signal attenuation factor of this channel is calculated. This is used to characterize the proportion of signal intensity remaining after propagation in water. The calculation formula is as follows: ; in, For the first The theoretical signal attenuation factor for each spectral channel is a dimensionless value between 0 and 1, representing the proportion of signal intensity remaining after two-way propagation through water: The spectral attenuation coefficient is derived from the output of the optical parameter calculation module; The target detection depth is represented by the coefficient 2, which indicates the attenuation of light through both incident and reflected paths in the water. This calculation yields a set of theoretical signal attenuation factor vectors. .

[0043] The weight calculation module then normalizes the aforementioned signal attenuation factors to generate a multispectral fusion weight vector. The normalization formula is as follows: ; in, For the first The fusion weighting coefficients of each spectral channel are the final output result; This represents the total number of spectral channels. Normalization ensures that the sum of all weight coefficients is 1, thus meeting the consistency requirement of energy distribution in multispectral fusion. The final weight vector output by the weight calculation module is... This is used to guide subsequent multispectral fusion processing.

[0044] A direct physical link was established between the determination of spectral fusion weights and the optical properties of water bodies and the target detection depth. For wavelengths with minimal light attenuation at a specified detection depth, i.e. Lower channels have lower attenuation factors. A higher value corresponds to a higher fusion weight. It is also larger. This allows for adaptive enhancement of channels with strong light transmission capabilities, effectively improving the clarity and accuracy of shallow water exploration.

[0045] The image fusion module performs weighted fusion of the effective water body's water-leaving signal based on the effective water body's water-leaving signal and multispectral fusion weights, and outputs the survey image. Furthermore, based on the effective water body water-leaving signal and multispectral fusion weights, the effective water body water-leaving signal is weighted and fused to output the exploration image, specifically including: Acquire image data of each spectral channel of the effective water body separation signal; The image data of each spectral channel are weighted using multispectral fusion weights. The weighted image data from each channel are fused at the pixel level to generate an exploration image.

[0046] Specifically, the image fusion module, after processing by the optical parameter calculation module and the weight calculation module, receives the effective water body out-of-water signal data cube output from the optical parameter calculation module and the multispectral fusion weight output from the weight calculation module. These are used to perform weighted fusion operations and output an underwater environment survey image. The effective water body out-of-water signal data cube input to the image fusion module is denoted as: ;in, Represents spatial coordinates Location, wavelength The effective water radiance value of the channel, in units of , This represents the total number of multispectral channels. The corresponding multispectral fusion weight vector is: ; in, To be at wavelength The fusion weight coefficient under the channel is the output of the weight calculation module.

[0047] The image fusion module first reads the effective water separation signal image data from each spectral channel. And based on the weight of the corresponding channel. The signals are weighted to reflect the signal effectiveness of different spectra at the target detection depth. The mathematical expression for the weighting is: ; in, Indicates the weighted number of... Image data for each spectral channel.

[0048] Subsequently, the image fusion module performs fusion operations on all weighted spectral channel image data at the pixel level to obtain the fused exploration image. The calculation formula is: ; in, Indicates coordinates The pixel values ​​of the fused image are the result of weighted superposition of all spectral channels, and the units are the same as those of the input image.

[0049] Through the above calculations, the image fusion module achieves the conversion from multispectral dimensions to two-dimensional spatial images, and the output is... It can be used directly as an underwater environment survey image for subsequent tasks such as target identification, terrain reconstruction, or water quality inversion.

[0050] The image fusion module tightly integrates the effective water separation signal based on physical modeling with fusion weights based on spectral attenuation characteristics, thereby optimizing the balance between signal fidelity and spectral balance in the generated exploration images. For bands with strong water transmission and high signal-to-noise ratio, the corresponding weights... Larger, after fusion The contribution of the medium band is more significant, enhancing the visibility and detail resolution of the target area. The weight of bands with strong attenuation or greater noise influence is automatically reduced to suppress unnecessary interference signals. This allows for adaptive optimization of image spectral combination under different water depths and optical conditions, effectively improving the detection accuracy and overall imaging quality of underwater targets, and providing a reliable image foundation for UAVs to conduct rapid exploration and environmental monitoring of shallow water bodies.

[0051] The control module, based on the UAV's flight status and position information, triggers the optical parameter calculation module and weight calculation module to update according to preset update conditions during flight.

[0052] Furthermore, triggering updates during flight based on preset update conditions specifically includes: Monitor preset update conditions. The preset update conditions are that the change in the spectral attenuation coefficient output by the optical parameter calculation module exceeds a preset threshold. When the preset update conditions are met, an execution command is sent to the optical parameter calculation module and the weight calculation module. After receiving the execution command, the optical parameter calculation module and the weight calculation module recalculate and output the updated spectral attenuation coefficient and multispectral fusion weight based on the currently acquired and corrected image signal processed by the image preprocessing module.

[0053] Specifically, the control module dynamically schedules the execution of the optical parameter calculation module and the weight calculation module during flight, ensuring that the multispectral data fusion results can reflect the optical properties of the water body and environmental changes in real time. The control module continuously receives flight status and position information output by the UAV flight control unit, including current position coordinates. Flight speed and heading angle The system monitors parameters such as the spectral attenuation coefficient output by the optical parameter calculation module to determine whether an update should be triggered.

[0054] The control module sets a preset update condition to determine whether there has been a significant change in the environmental optical properties. This condition is defined as the change in the spectral attenuation coefficient between two consecutive outputs from the optical parameter calculation module exceeding a preset threshold. Specifically, for each spectral channel Let the attenuation coefficient of the output at the previous moment be... The attenuation coefficient of the output at the current moment is The formula for calculating the change in spectral attenuation is: ; The preset update condition is considered to be triggered when the following conditions are met: ; in, This indicates the value of maximum variation across all spectral channels. The preset spectral attenuation change threshold for the system, in units of Its value is determined based on experimental calibration and reflects the sensitivity to significant changes in the optical environment of the water body.

[0055] Once the above conditions are met, the control module immediately sends an execution command to the optical parameter calculation module and the weight calculation module, triggering them to perform update calculations. Upon receiving the command, the optical parameter calculation module calculates the image signal acquired at the current moment and corrected by the image preprocessing module. Recalculate the spectral attenuation coefficient The weight calculation module then synchronously calculates based on the updated decay coefficient vector. With preset target detection depth Regenerate multispectral fusion weights The mathematical relationship of the update process can be expressed as: ; ; in, This represents the spectral attenuation inversion function executed by the optical parameter calculation module. This indicates the multispectral fusion weight calculation function executed by the weight calculation module.

[0056] This ensures that the drone can adaptively adjust to real-time environmental changes during flight. For example, when the drone flies over different water areas and the concentration of suspended matter, turbidity, or lighting conditions in the water change, the fluctuation of the spectral attenuation coefficient will be detected by the control module, triggering a module update to ensure that the spectral parameters and fusion weights remain consistent with the current observation environment.

[0057] By combining dynamic monitoring of optical parameters with a module-level update triggering mechanism, adaptive control for UAV multispectral exploration missions is achieved. The control module can automatically identify environmental changes and update calculation parameters without human intervention, ensuring that subsequently generated exploration images maintain high spectral consistency and signal reliability under different optical conditions. This improves the system's imaging stability and exploration accuracy in complex shallow water environments.

[0058] Example 2: Using fixed or empirically set weighting coefficients makes it difficult to adaptively adjust the fusion ratio of each band according to changes in water turbidity. This results in unstable fusion results under different aquatic environments, making it difficult to balance spectral contrast and detection depth, thus affecting the accuracy of underwater target identification and the system's environmental adaptability. To address these issues, this invention provides a UAV-based water conservancy survey method based on multi-source data fusion, the structure of which is as follows: Figure 2 As shown. The method includes the following steps: The airborne multispectral polarization imaging unit acquires water surface information to obtain raw image signals. Based on the original image signal, geometric correction is performed to generate a corrected image signal; Based on the corrected image signal, a polarization decoupling model is used to generate an effective water body water separation signal, and the spectral attenuation coefficient is output based on the effective water body water separation signal. Based on the spectral attenuation coefficient, the preset target detection depth, and the spectral signal-to-noise ratio model, the multispectral fusion weight is calculated. Based on the effective water body water-free signal and multispectral fusion weights, weighted fusion is performed to generate exploration images; Based on the flight status and location information of the UAV, the spectral attenuation coefficient and multispectral fusion weight are recalculated during flight according to preset update conditions.

[0059] Specifically, firstly, multispectral polarization image data of the water surface area is acquired using an airborne multispectral polarization imaging unit to obtain raw image signals reflecting the characteristics of the water surface and its underlying environment. Then, geometric correction is performed on the raw image signals to eliminate spatial distortion caused by changes in UAV attitude, viewpoint distortion, and optical deviations, resulting in a corrected image signal with accurate spatial positioning.

[0060] After geometric correction, based on the corrected image signal, the reflected light component from the water surface is separated using a polarization decoupling model to extract the actual water separation signal. Furthermore, based on the water separation signal, a pre-trained optical inversion algorithm is used to calculate the spectral attenuation coefficient for each spectral band, characterizing the propagation attenuation characteristics of light in different bands within the water body, providing a physical basis for subsequent fusion calculations.

[0061] After obtaining the spectral attenuation coefficient, the multispectral fusion weights for each spectral channel are calculated based on the preset target detection depth and spectral signal-to-noise ratio model. By weighting the out-of-water signals of each band according to the fusion weights, an exploration image with enhanced spectral contrast is generated, thereby improving the ability to identify and analyze underwater targets or environmental features.

[0062] During UAV flight, based on real-time changes in flight status and position information, and when preset update conditions are met, the spectral attenuation coefficient and multispectral fusion weight are automatically recalculated to achieve adaptive adjustments to different environmental conditions. This method enables high-precision, dynamic reconnaissance of target areas in complex aquatic environments, improving the stability and reliability of UAVs in joint surface and underwater detection.

[0063] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent unmanned aerial vehicle (UAV) water conservancy survey system based on multi-source data fusion, characterized in that, include: The image acquisition module is used to acquire water surface information and output raw image signals through an airborne multispectral polarization imaging unit; The image preprocessing module performs geometric correction based on the original image signal to output the corrected image signal; The optical parameter calculation module, based on the corrected image signal, processes it through a polarization decoupling model to generate an effective water body water separation signal, and outputs the spectral attenuation coefficient based on the effective water body water separation signal; The weight calculation module calculates and outputs the multispectral fusion weights based on the spectral attenuation coefficient, the preset target detection depth, and the spectral signal-to-noise ratio model. The image fusion module performs weighted fusion of the effective water body water-leaving signal based on the effective water body water-leaving signal and the multispectral fusion weight, and outputs the exploration image; The control module, based on the UAV's flight status and position information, triggers the optical parameter calculation module and the weight calculation module to update according to preset update conditions during flight.

2. The intelligent unmanned aerial vehicle (UAV) water conservancy survey system based on multi-source data fusion according to claim 1, characterized in that, The process of acquiring water surface information and outputting raw image signals through an airborne multispectral polarization imaging unit specifically includes: A single exposure is performed on the target water area using a snapshot-type multispectral polarization imaging unit to instantly capture a mixed light signal containing target information and environmental interference; By utilizing the microfilter array and micropolarizer array integrated on the focal plane of the imaging unit, the mixed optical signal is synchronously separated into multiple spectral channel data and multiple polarization analysis channel data in pixel-level space; The image sensor converts the separated channel data into digital electrical signals and combines them to output the original image signal.

3. The intelligent unmanned aerial vehicle (UAV) water conservancy survey system based on multi-source data fusion according to claim 1, characterized in that, The geometric correction based on the original image signal specifically includes: Acquire the UAV's flight status and position information synchronized with the acquisition time of the original image signal; Based on the flight status and position information and the preset camera intrinsic parameters, a geometric correction model is established; The original image signal is resampled using the geometric correction model to generate and output the corrected image signal.

4. The intelligent unmanned aerial vehicle (UAV) water conservancy survey system based on multi-source data fusion according to claim 1, characterized in that, The step of generating an effective water separation signal by processing the corrected image signal using a polarization decoupling model specifically includes: Based on the multi-polarization state data in the corrected image signal, the Stokes vector method is used to calculate the Stokes vector and polarization degree of each spectral channel. Based on the Stokes vector and degree of polarization, and utilizing the difference in polarization characteristics between the reflected light from the water surface and the signal leaving the water body, the interference component of the reflected light from the water surface is estimated through a polarization decoupling model. The water surface reflection interference component is subtracted from the total light intensity signal corresponding to the corrected image signal to generate an effective water body departure signal.

5. The intelligent unmanned aerial vehicle (UAV) water conservancy survey system based on multi-source data fusion according to claim 1, characterized in that, The spectral attenuation coefficient retrieved based on the effective water body water separation signal specifically includes: The effective water body's water separation signal is organized into a spectral feature vector; The spectral feature vector is input into a pre-trained inversion model, which is trained by machine learning and used to construct a nonlinear mapping relationship between the spectral morphology of the water body detachment signal and the spectral attenuation coefficient. The inversion model outputs the spectral attenuation coefficient.

6. The intelligent unmanned aerial vehicle (UAV) water conservancy survey system based on multi-source data fusion according to claim 1, characterized in that, The calculation and output of the multispectral fusion weights based on the spectral attenuation coefficient, the preset target detection depth, and the spectral signal-to-noise ratio model specifically includes: Obtain the vector composed of the spectral attenuation coefficients and receive the target detection depth parameters input from the outside; For each spectral channel, based on the spectral attenuation coefficient and the target detection depth parameter, the theoretical signal attenuation factor of that channel is calculated according to the exponential attenuation model with the spectral attenuation coefficient as the key parameter. The theoretical signal attenuation factors for all spectral channels are normalized to generate a set of weighting coefficients; The weighting coefficients are output as the multispectral fusion weights.

7. The intelligent unmanned aerial vehicle (UAV) water conservancy survey system based on multi-source data fusion according to claim 1, characterized in that, The step of performing weighted fusion of the effective water body water-leaving signal based on the effective water body water-leaving signal and the multispectral fusion weight, and outputting the exploration image specifically includes: Acquire image data of each spectral channel of the effective water body's water separation signal; The image data of each spectral channel are weighted using the multispectral fusion weights. The weighted image data from each channel are fused at the pixel level to generate an exploration image.

8. The intelligent unmanned aerial vehicle (UAV) water conservancy survey system based on multi-source data fusion according to claim 1, characterized in that, The specific details of triggering updates during flight based on preset update conditions include: The preset update condition is monitored, wherein the preset update condition is that the change in the spectral attenuation coefficient output by the optical parameter calculation module exceeds a preset threshold. When the preset update condition is met, an execution instruction is sent to the optical parameter calculation module and the weight calculation module. After receiving the execution command, the optical parameter calculation module and the weight calculation module recalculate and output the updated spectral attenuation coefficient and multispectral fusion weight based on the currently acquired and corrected image signal processed by the image preprocessing module.

9. A method for unmanned aerial vehicle (UAV) water conservancy survey based on multi-source data fusion, characterized in that, The method for the UAV-based intelligent system for water conservancy exploration according to any one of claims 1-8, comprising the following steps: The airborne multispectral polarization imaging unit acquires water surface information to obtain raw image signals. Based on the original image signal, geometric correction is performed to generate a corrected image signal; Based on the corrected image signal, a polarization decoupling model is used to process and generate an effective water body water separation signal, and the spectral attenuation coefficient is output based on the effective water body water separation signal. Based on the spectral attenuation coefficient, the preset target detection depth, and the spectral signal-to-noise ratio model, the multispectral fusion weight is calculated. Based on the effective water body water-free signal and the multispectral fusion weights, weighted fusion is performed to generate exploration images; Based on the UAV's flight status and location information, the spectral attenuation coefficient and the multispectral fusion weight are recalculated during flight according to preset update conditions.

Images