An enteromorpha area estimation method based on a non-orthographic operation mode of a drone

Abstract

The application relates to the technical field of marine ecological disaster monitoring, and discloses a green algae area estimation method based on an unmanned aerial vehicle non-orthographic operation mode, which comprises the following steps: firstly, spectrum energy analysis is used to determine a sea state mode and to inversely calculate a basic wave height, a vertical non-uniform atmospheric extinction model is constructed to perform physical defogging; secondly, a fluid-solid coupling damping coefficient is inversely calculated based on texture entropy and glare characteristics, and a green algae microscopic equivalent mean square slope is corrected; finally, pixel-level macroscopic slant range projection, a wave shielding compensation factor and a microscopic surface area correction factor are comprehensively used, and a Sigmoid smoothing truncation weight function is combined to calculate a green algae physical total surface area. The application solves the measurement error caused by the non-uniformity of atmospheric medium, wave shielding and the absence of microscopic geometric characteristics under non-orthographic observation, and significantly improves the monitoring accuracy under complex sea conditions.

Description

Technical Field

[0001] This invention relates to the field of marine ecological disaster monitoring technology, specifically a method for estimating the area of ​​*Ulva prolifera* based on the non-orthophoto operation mode of unmanned aerial vehicles (UAVs). Background Technology

[0002] Unmanned aerial vehicle (UAV) aerial remote sensing is an important means of routine monitoring of nearshore seaweed blooms, offering advantages over shore-based monitoring and satellite remote sensing, including high timeliness, maneuverability, and resolution. In practical operations, to improve the monitoring coverage and efficiency of a single flight, UAVs typically employ wide-angle oblique photography (non-orthophoto) mode. However, in this non-orthophoto UAV observation scenario, existing image processing and area measurement methods struggle to adapt to the complex physical environment of the air-sea interface. Nearshore waters are often accompanied by breaking waves, resulting in a significantly uneven vertical distribution of salt spray aerosols in the near-surface atmosphere. Traditional defogging algorithms often rely on idealized assumptions of uniform atmospheric distribution, leading to deviations in transmittance calculations along the UAV's oblique observation path, which severely impacts the accurate inversion of scene depth information.

[0003] Furthermore, under low-angle tilt observation conditions, the dynamic fluctuations of sea waves can obstruct targets floating in the troughs. Existing technologies typically simplify the sea surface as a static rigid plane for homography geometric projection, failing to consider the blind spots caused by wave obstruction and ignoring the area increase brought about by the surface roughness of seaweed as a viscoelastic float at the microscale. This results in systematic errors in the final calculated monitoring data, making it difficult to meet the quantitative requirements for accurate disaster prevention and mitigation.

[0004] Therefore, this invention proposes a method for estimating the area of ​​*Ulva prolifera* based on the non-orthophoto operation mode of unmanned aerial vehicles (UAVs) to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for estimating the area of ​​*Ulva prolifera* based on the non-orthophoto mode of UAV operations. This method solves the problem of low accuracy in measuring the area of ​​*Ulva prolifera* caused by non-uniform distribution of atmospheric medium, dynamic wave occlusion, and lack of microscopic surface geometric features in non-orthophoto observation scenarios using UAVs.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for estimating the area of ​​*Ulva prolifera* based on the non-orthophoto operation mode of unmanned aerial vehicles (UAVs), comprising the following steps:

[0007] The frequency domain analysis of the raw non-orthophoto image acquired by the monocular camera is performed using the spectral energy classification locking module to extract the spectral energy density of the long-wave swell frequency band. The spectral energy density is compared with the preset calm sea state energy threshold to determine the sea state mode, and the basic wave height parameter is calculated or set according to the sea state mode.

[0008] Using the air-sea interface coupled depth field reconstruction module, a vertically non-uniform atmospheric extinction coefficient distribution model is constructed based on the basic wave height parameters. The atmospheric extinction coefficient distribution model is non-uniformly integrated along the line of sight to solve the scene atmospheric transmittance. The scene atmospheric transmittance is then used to perform physical dehazing on the original non-orthophoto image to generate a restored image.

[0009] The texture entropy features and glare statistical features of the restored image are extracted using the fluid-structure interaction parameter inversion module. The fluid-structure interaction damping coefficient of the seaweed to the wave is inverted based on the texture entropy features, and the micro equivalent mean square slope of the seaweed surface is corrected by combining the glare statistical features.

[0010] Using a multi-factor weighted area integration module, the macroscopic projected area, wave occlusion compensation factor, and microscopic surface area correction factor at the pixel level are comprehensively calculated. Combined with a smooth truncation weight function based on the Sigmoid function, the macroscopic projected area, wave occlusion compensation factor, microscopic surface area correction factor, and the weight value calculated by the smooth truncation weight function are weighted and accumulated to output the total physical surface area of ​​Ulva prolifera.

[0011] Preferably, the step of performing frequency domain analysis on the original non-orthophoto image acquired by the monocular camera using the spectral energy classification locking module includes: performing a two-dimensional discrete Fourier transform on the original non-orthophoto image to obtain a frequency domain image; constructing an adaptive low-pass filter, wherein the cutoff frequency of the adaptive low-pass filter is set according to the projection pixel period of the minimum physical wavelength of the target surge on the image plane; filtering the frequency domain image using the adaptive low-pass filter to extract the low-frequency component corresponding to the long-wave surge frequency band; calculating the mean square amplitude of the low-frequency component as the spectral energy density; if the spectral energy density is greater than the steady sea state energy threshold, it is determined to be a wind and wave mode; if the spectral energy density is less than or equal to the steady sea state energy threshold, it is determined to be a steady sea state mode.

[0012] Preferably, the steps of calculating or setting the basic wave height parameter according to the sea state mode include: when the wind and wave mode is determined, performing an inverse transform on the low-frequency component to recover the spatial domain swell image, detecting the pixel spacing between adjacent wave peaks on the spatial domain swell image, using perspective imaging geometry and camera tilt angle to invert the pixel spacing into physical wavelength, and calculating the basic wave height parameter according to a preset wave steepness coefficient; when the static steady mode is determined, directly setting the basic wave height parameter to a preset small value so that the subsequent atmospheric extinction coefficient distribution model reverts to a standard state containing only background aerosols.

[0013] Preferably, the step of constructing a vertically non-uniform atmospheric extinction coefficient distribution model based on the basic wave height parameter includes: establishing an extinction function driven by the vertical height variable, wherein the extinction function is composed of the superposition of background atmospheric extinction term and droplet aerosol extinction term; wherein the initial intensity of the droplet aerosol extinction term is positively correlated with the basic wave height parameter, and the intensity of the droplet aerosol extinction term decreases exponentially with the increase of vertical height; the non-uniform integration of the atmospheric extinction coefficient distribution model along the line-of-sight path refers to establishing the mapping relationship between the line-of-sight path and the vertical height based on the line-of-sight incident angle of the pixel and the slant distance Euclidean distance corresponding to the pixel, and performing line integration of the extinction function along the line-of-sight path to obtain the scene atmospheric transmittance that varies with the slant distance Euclidean distance.

[0014] Preferably, the step of inverting the fluid-structure interaction damping coefficient of seaweed to waves based on the texture entropy feature includes: segmenting the restored image into a seaweed target region and a seawater background region, and calculating the local texture entropy of the seaweed target region and the local texture entropy of the seawater background region respectively; constructing a mapping relationship between the fluid-structure interaction damping coefficient and the difference ratio of texture entropy; if the local texture entropy of the seaweed target region is significantly less than the local texture entropy of the seawater background region, then the fluid-structure interaction damping coefficient is determined to be close to a fully damped state; if the local texture entropy of the seaweed target region is close to the local texture entropy of the seawater background region, then the fluid-structure interaction damping coefficient is determined to be close to an undamped state.

[0015] Preferably, the step of correcting the micro-equivalent mean square slope of the Ulva surface by combining the statistical characteristics of the glare includes: statistically analyzing the variance of the brightness distribution of the glare region in the restored image, and inverting the full-spectrum mean square slope of the seawater using the Cox-Munk model; using the fluid-structure interaction damping coefficient to attenuate and correct the full-spectrum mean square slope of the seawater to obtain the micro-equivalent mean square slope of the Ulva surface; and calculating the micro-surface area correction factor using the micro-equivalent mean square slope according to the random surface geometry theory, wherein the micro-surface area correction factor is used to characterize the area increment of the micro-roughness relative to the horizontal projection surface.

[0016] Preferably, the step of calculating the pixel-level macroscopic projected area includes: determining the slant distance Euclidean distance from the pixel to the camera's optical center and the line-of-sight incident angle based on the pinhole camera imaging model; calculating the ratio of the square of the slant distance Euclidean distance to the camera's normalized focal length, and dividing the ratio by the cosine of the line-of-sight incident angle to obtain the macroscopic projected area; wherein the cosine of the line-of-sight incident angle is used to project and convert the cross-sectional area perpendicular to the line of sight onto the horizontal sea surface.

[0017] Preferably, the step of calculating the wave occlusion compensation factor includes: constructing an occlusion model based on probability statistics, wherein the wave occlusion compensation factor is positively correlated with the basic wave height parameter, positively correlated with the tangent of the line of sight incident angle corresponding to the macroscopic projected area, and inversely proportional to the slant distance Euclidean distance of the pixel; and using the wave occlusion compensation factor to numerically compensate for the area of ​​seaweed that was not captured by the camera due to wave crest occlusion.

[0018] Preferably, the steps of combining the smooth truncation weight function based on the Sigmoid function include: defining an effective observation distance cutoff threshold and a transition band width parameter; constructing a Sigmoid-type weight function with the slant range Euclidean distance as the independent variable; the Sigmoid-type weight function tends to retain the original value in the near-range region where the slant range Euclidean distance is much smaller than the effective observation distance cutoff threshold, tends to suppress to zero in the far-range region where the slant range Euclidean distance is much larger than the effective observation distance cutoff threshold, and exhibits a smooth decrease in the transition region near the effective observation distance cutoff threshold; and using the smooth truncation weight function to weight and suppress the area calculation results of far-end low signal-to-noise ratio pixels to prevent numerical divergence.

[0019] Preferably, the output of the total physical surface area of ​​*Ulva prolifera* is achieved in the following way: for each pixel identified as *Ulva prolifera* in the restored image, the macroscopic projected area of ​​the pixel, the wave occlusion compensation factor, the microscopic surface area correction factor, and the weight value calculated according to the smoothing truncation weight function are multiplied by four terms to obtain the physical surface area contribution value of the pixel; the physical surface area contribution values ​​of all *Ulva prolifera* pixels are accumulated to obtain the total physical surface area of ​​*Ulva prolifera*.

[0020] This invention provides a method for estimating the area of ​​*Ulva prolifera* based on the non-orthophoto operation mode of unmanned aerial vehicles (UAVs). It has the following beneficial effects:

[0021] 1. This invention locks the sea state mode by spectral energy classification and inverts the basic wave height, constructing a vertical non-uniform atmospheric extinction model that includes the near-surface droplet gradient. This model overcomes the shortcomings of traditional defogging algorithms that ignore the uneven vertical distribution of sea surface salt fog aerosols, and can dynamically adjust the extinction coefficient according to the real-time wave height, effectively improving the transmittance calculation accuracy and defogging restoration effect of UAV non-orthophoto images under complex meteorological conditions accompanied by sea fog and swells.

[0022] 2. This invention employs a macroscopic projection calculation based on slant range and a wave occlusion compensation mechanism. To address geometric distortion caused by non-orthophoto observations, a modified slant range projection formula replaces the traditional vertical depth formula, eliminating systematic biases in far-end area calculations. Simultaneously, an occlusion compensation factor based on probability statistics is introduced to numerically compensate for the area of ​​*Ulva prolifera* that was not captured by the camera due to wave crest occlusion, solving the problem of target omission during low-angle observations.

[0023] 3. This invention utilizes fluid-structure interaction parameter inversion and microscopic surface area correction techniques to achieve a mapping from two-dimensional image projection to three-dimensional physical surface area. By analyzing the texture entropy differential inversion to obtain the damping coefficient of *Ulva prolifera* on waves, and correcting the microscopic equivalent mean square slope accordingly, the area increment caused by microscopic roughness on the *Ulva prolifera* surface can be quantified, significantly improving the physical authenticity and measurement accuracy of *Ulva prolifera* disaster monitoring data. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0025] Figure 2 This is a flowchart illustrating the spectral energy hierarchical locking module of the present invention;

[0026] Figure 3 This is a schematic diagram illustrating the principle of the air-sea interface coupled depth field reconstruction and non-uniform atmospheric extinction model of the present invention.

[0027] Figure 4 This is a schematic diagram of the fluid-structure interaction effect of seaweed and waves and the principle of surface roughness correction of the present invention;

[0028] Figure 5 This is a schematic diagram of the multi-factor weighted area integral geometric model of the present invention, which takes into account wave occlusion compensation.

[0029] Figure 6 This is a schematic diagram of the original hazy non-orthophoto image of the present invention;

[0030] Figure 7 This is a schematic diagram showing the restoration comparison images after physical defogging treatment according to the present invention. Detailed Implementation

[0031] 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.

[0032] See attached document Figure 1 , Figure 1 This is a flowchart illustrating a method for estimating the area of ​​*Ulva prolifera* based on a non-orthophoto operation mode using a drone, according to an embodiment of the present invention. The present invention provides a method for estimating the area of ​​*Ulva prolifera* based on a non-orthophoto operation mode using a drone. This method is executed collaboratively by a spectral energy classification locking module, an air-sea interface coupled depth field reconstruction module, a fluid-structure interaction parameter inversion module, and a multi-factor weighted area integration module.

[0033] The spectral energy classification and locking module receives the raw non-orthophoto image acquired by a monocular camera. This module first performs a two-dimensional discrete Fourier transform on the image to obtain a frequency domain representation, and then uses a low-pass filter to extract the low-frequency components corresponding to the long-wave surge frequency band. The module calculates the spectral energy density of this low-frequency component and compares it with a preset static steady sea state energy threshold. Based on the comparison result, the module determines whether the current sea state belongs to a wind-wave mode or a static steady mode, and calculates or sets the basic wave height parameter accordingly. This basic wave height parameter is output as an initial environmental constraint to the next-level module.

[0034] The air-sea interface coupled depth field reconstruction module constructs a vertically non-uniform atmospheric extinction coefficient distribution model using the received fundamental wave height parameters. This atmospheric extinction coefficient distribution model includes a background aerosol term and a correction term for near-surface droplet aerosols generated by wave breaking. Based on the camera imaging geometry, the air-sea interface coupled depth field reconstruction module performs non-uniform integration on the variable coefficient extinction model along the line of sight to solve for pixel-level scene atmospheric transmittance and depth field. Based on the solved transmittance, the air-sea interface coupled depth field reconstruction module performs physical dehazing on the original non-orthophoto image to generate a restored image containing clear spectral features and texture details.

[0035] The fluid-structure interaction parameter inversion module extracts features and calculates parameters from the restored image. This module identifies glare regions in the image, statistically analyzes the glare brightness distribution characteristics, and uses the Cox-Munk model to invert the full-spectrum mean square slope of the seawater. Simultaneously, the module segments the target area of ​​the seaweed and the pure seawater area, calculating the local texture entropy of each. Based on the difference ratio of texture entropy, the module quantitatively calculates the fluid-structure interaction damping coefficient of the seaweed as a viscoelastic body on the waves. Using this damping coefficient, the module corrects the full-spectrum mean square slope of the seawater to obtain the microscopic equivalent mean square slope of the seaweed surface.

[0036] The multi-factor weighted area integration module integrates depth field information and micro-geometric parameters to perform the final area measurement. Based on a pinhole camera model, this module calculates the macroscopic projected area of ​​each pixel on the horizontal plane and combines the base wave height and incident angle to calculate the wave occlusion compensation factor. The module constructs a smoothing truncation weight function based on the sigmoid function according to the pixel's signal-to-noise ratio index, and uses this weight function to weight distant pixels to suppress numerical divergence. Finally, the module multiplies and accumulates the macroscopic projected area, microscopic surface area correction factor, occlusion compensation factor, and smoothing truncation weight to output the final physical total surface area of ​​the seaweed.

[0037] See attached document Figure 1 , Figure 1This is a flowchart illustrating a method for estimating the area of ​​*Ulva prolifera* based on a non-orthophoto operation mode using a drone, according to an embodiment of the present invention. The present invention provides a method for estimating the area of ​​*Ulva prolifera* based on a non-orthophoto operation mode using a drone, comprising the following steps:

[0038] S100 uses the spectral energy classification locking module to perform frequency domain analysis on the input raw non-orthophoto image, determines the sea state mode based on the spectral energy density of the long-wave swell frequency band, and calculates or sets the basic wave height parameters to initialize the atmospheric physical model accordingly.

[0039] S200 utilizes the air-sea interface coupled depth field reconstruction module to construct a vertically non-uniform atmospheric extinction coefficient distribution model based on the fundamental wave height parameter. It then uses the non-uniform integration of the line-of-sight path to invert the slant range Euclidean distance field and generate a dehazed restored image.

[0040] S300 uses the fluid-structure interaction parameter inversion module to extract the texture entropy features and glare statistical features of the restored image, calculates the fluid-structure interaction damping coefficient of seaweed to waves, and corrects the microscopic equivalent mean square slope of the seaweed surface accordingly.

[0041] The S400 uses a multi-factor weighted area integration module to integrate the macroscopic projected area, wave shading compensation factor, and microscopic surface area correction factor, combined with a smoothing truncation weight function based on the Sigmoid function, to calculate and output the total physical surface area of ​​Ulva prolifera.

[0042] To further clarify the implementation of each technical aspect of the present invention, the following will provide a detailed description of the implementation of each functional module involved above and its internal processing flow.

[0043] Before executing step S100, the system reads the camera's intrinsic parameters (focal length) in advance and obtains extrinsic parameters (drone flight altitude, gimbal pitch angle) in real time from the UAV flight control system or photo EXIF ​​information; at the same time, it obtains the current solar zenith angle information, which can be calculated based on the observation time and GPS coordinates.

[0044] See attached document Figure 2 In step S100, to address the problem that traditional methods struggle to obtain reliable initial values ​​for atmospheric models under low signal-to-noise ratio or textureless sea conditions, the energy distribution characteristics of the image frequency domain are utilized to extract long-wave swell features that penetrate atmospheric scattering from noisy images. Specifically, this includes the following steps:

[0045] S110, acquire the raw non-orthophoto image captured by the monocular camera. A two-dimensional discrete Fourier transform is performed on the image to convert it from the spatial domain to the frequency domain; the width of the image is set to... The height is pixel coordinates are Then the frequency domain image The calculation is as follows:

[0046] ;

[0047] In the formula, , These are the frequency variables in the horizontal and vertical directions, respectively. The unit is the imaginary unit; this transformation decomposes the image into a linear combination of sinusoidal fundamental waves of different frequencies, so that the wave texture, swell contour and high-frequency noise in the image can be separated in the frequency domain; for the specific implementation algorithm of the Fourier transform, those skilled in the art can use mature algorithms such as the Fast Fourier Transform (FFT), which will not be elaborated here.

[0048] S120, an adaptive low-pass filter is constructed to extract the low-frequency components corresponding to the long-wave surge frequency band. Since haze and small reflections on the water surface are mainly concentrated in the high-frequency region, while large-scale surge contours are mainly concentrated in the low-frequency region, and low-frequency signals experience less attenuation during atmospheric transmission, the long-wave component is selected as the environmental feature carrier. This embodiment uses a Gaussian low-pass filter. Its transfer function is defined as:

[0049] ;

[0050] In the formula, Midpoint in the frequency domain Euclidean distance to the center of the spectrum This is the cutoff frequency.

[0051] To ensure that the extracted low-frequency components accurately correspond to the characteristics of surge waves in the physical world, the cutoff frequency is... It is not a fixed constant, but a dynamic parameter that is adaptively set according to the camera's imaging parameters and the preset physical wavelength of the surge; specifically, the minimum physical wavelength of the target surge is set to... (For example, take 50 meters), the camera focal length is The drone's flight altitude is According to the principle of pinhole imaging, the physical wavelength The number of pixels occupied by a projection on the image plane is defined as the minimum pixel period. Cutoff frequency The spatial frequency is set to be the reciprocal of the pixel period, i.e. ,in This is a scaling factor; this setting ensures that the filter retains a physical scale greater than [the specified value]. The long-wave signal is filtered to remove wave textures and high-frequency noise.

[0052] The low-frequency spectrum is obtained by filtering the frequency domain image using the constructed filter. :

[0053] ;

[0054] The low-frequency spectrum This includes information on the large-scale swell geometry under the current sea state, serving as the data basis for subsequent determination of sea state energy levels and inversion of wave height parameters.

[0055] S130, calculate the spectral energy density and perform sea state mode decision; acquire the low-frequency spectrum. Subsequently, to quantify the dynamic intensity of ocean waves within the current field of view, the spectral energy density is defined. The calculation method involves obtaining the mean of the magnitude square of the low-frequency spectrum after filtering in the frequency domain.

[0056] ;

[0057] In the formula, The modulus of the complex spectrum; the energy density It directly reflects the significance of low-frequency long-wave components on the sea surface; when the sea conditions are bad and the swells are large, the undulating texture of the sea surface is obvious, corresponding to higher low-frequency energy; when the sea surface is in a calm and stable state or is obscured by dense fog, causing the texture to be lost, the low-frequency energy is extremely low.

[0058] Set the energy threshold for calm sea state ;collection Frames (e.g.) Calculate the arithmetic mean of the low-frequency spectral energy density of sea surface images under calm and stable sea conditions with wind speeds less than 2 m / s. ,set up ,in, For safety margin, the value is typically between 1.2 and 1.5 times; the system will calculate the... With threshold The comparison is performed, and the current processing flow is directed to two different parameter initialization branches based on the comparison results: the wind and wave mode branch or the static and stable mode branch. This energy-based decision mechanism can effectively identify whether the input image contains sufficient texture information to support geometric inversion, thereby avoiding the problems of model parameter calculation errors or non-convergence caused by geometric inversion under textureless conditions.

[0059] S140, perform dual-mode foundation wave height parameter inversion; based on the decision result of S130, determine the foundation wave height parameters using the following two strategies respectively. This parameter will serve as a key initial value for the subsequent construction of a non-uniform atmospheric extinction model.

[0060] like The current mode is determined to be wind and waves; at this time, the image contains valid swell texture, and the wave height can be inverted using perspective geometry principles; for the low-frequency spectrum... Perform an inverse Fourier transform to recover the spatial domain image containing only long-wavelength components. ;exist The position of the wave peak is detected on the vertical center line; let the detected peak be the first wave peak. The vertical coordinates of each peak in the image are , No. The ordinate of each peak is The spacing between adjacent peaks on the image is then... Due to the perspective effect of non-orthographic projection, waves with equal physical spacing appear larger when closer and smaller when farther away in the image, and the distance between wave crests... With the ordinate The rate of change (i.e., texture compression ratio) is a function of the physical wave height and the UAV's flight altitude; based on the imaging geometry model, a baseline wave height is established. The inversion formula; firstly, using the near-end crest spacing Calculate the physical wavelength of ocean waves Then, the wave height is estimated based on the empirical relationship of deep-water wave steepness:

[0061] ;

[0062] In the formula, Focal length (in pixels). The downward tilt angle of the camera's optical axis relative to the horizontal plane. This is the wave steepness coefficient, taking into account that nearshore waters typically exhibit a mixed sea state of swells and wind waves. The value range is [0.03, 0.05] to cover the waveform characteristics of a fully grown wind wave from a decaying swell; through this explicit geometric relationship, the basic wave height in physical space is calculated from the image texture features. .

[0063] like The system is determined to be in a statically stable mode. At this point, sea surface texture is missing or the signal-to-noise ratio is too low, making effective geometric inversion impossible. To ensure system stability, a parameter preset operation is performed, directly setting the base wave height to a preset, small value.

[0064] ;

[0065] This operation forces subsequent atmospheric physics models to ignore the droplet effect caused by waves and revert to a standard stratified atmospheric model that only considers background aerosols. This mechanism ensures that the system can still output physically reasonable depth estimates under extremely calm or extremely obstructed visibility conditions, rather than producing erroneous random values.

[0066] See attached document Figure 3 In step S200, to address the problem of severe distortion in the estimation of far-end depth and area caused by the neglect of atmospheric inhomogeneity in traditional defogging algorithms in sea surface monitoring scenarios, a coupled extinction model of the sea-atmosphere interface that conforms to physical reality is established using the basic wave height parameters obtained in the previous steps. This specifically includes the following steps:

[0067] S210, Constructing a vertical extinction model for near-surface droplet aerosols; In the non-orthophoto environment of UAV observation, the atmospheric extinction coefficient is not a spatial constant, but exhibits a significant gradient change with vertical height above the sea surface; This non-uniformity mainly stems from two parts: one is the natural stratification of background atmospheric aerosols, and the other is the ejection of salt spray droplets generated by breaking waves; This embodiment introduces a model based on the fundamental wave height. Driven vertical extinction coefficient distribution function To quantitatively characterize the physical process; define Let be the vertical height of the spatial point above sea level, then the total extinction coefficient is... This is represented as the superposition of the background component and the droplet component:

[0068] ;

[0069] In the formula, The background atmospheric extinction coefficient is usually measured by a visibility meter or estimated from the infinity point of the dark channel of an image. It represents the uniform atmospheric composition that does not change drastically with altitude. The extinction coefficient of droplet aerosols is controlled by sea state dynamics parameters.

[0070] Based on the theory of air-sea interaction, the amount of droplet generation is positively correlated with wave energy, and its concentration decreases exponentially with increasing altitude; therefore, the following droplet extinction model is established:

[0071] ;

[0072] In the formula, The base wave height obtained in step S100 is the parameter that directly determines the initial generation intensity of droplets. The droplet formation constant is determined by the salinity and surface tension characteristics of the sea area. In this embodiment, the value range is set to [1.0 × 10⁻⁶]. -5 5.0×10 -4 ]; The vertical attenuation rate of droplets characterizes the effect of gravity settling on the vertical distribution of droplets; this model clearly describes the change in sea state from calm to wind and waves (i.e., When the surface temperature increases, the extinction coefficient near the sea surface increases significantly, and this increase is limited to a certain height range near the surface. As the altitude increases, the extinction coefficient decreases further. The rise in wave height rapidly decays to the background level; this non-uniform model driven by wave height corrects the systematic bias of traditional algorithms in overestimating near-sea surface transmittance under windy and wavey weather.

[0073] S220 performs non-uniform integration and transmittance calculation of the line-of-sight path; under non-orthogonal (tilted) observation geometry, the propagation path of light reflected from the object surface into the camera is a slant path, and the vertical height corresponding to different positions on this path is... It is variable; due to the extinction coefficient constructed in step S210. The extinction varies with altitude, therefore the traditional constant extinction formula (i.e., the simplified form of Beer-Lambert's law) cannot be directly applied. (This requires integration along the line of sight), and the integration must be performed along the line of sight path.

[0074] Obtain the current flight altitude of the drone. For pixels in an image The corresponding angle of incidence of the line of sight is (Defined as the angle between the line of sight and the vertical direction); Assuming the light travels in a straight line, the distance from any point on the path to the camera is... Then the vertical height of that point It can be represented as Atmospheric transmittance Defined as the light ray traveling along the path from the target (distance) The energy attenuation ratio transmitted to the camera (distance 0); by linearly integrating the instantaneous extinction coefficient along the path, the analytical expression for atmospheric transmittance in a non-uniform medium is obtained:

[0075] ;

[0076] Will Substituting the specific expression into the integral, and considering the transformation relationship of the integral variables... The above integral can be transformed into an integral with respect to the vertical height:

[0077] ;

[0078] In the formula, The height of the target point (i.e., the sea surface) is usually taken as 0; after completing the integration calculation, the transmittance and the slant range Euclidean distance are obtained. The explicit functional relationship; this integral process accurately quantifies the cumulative attenuation effect of light passing through the high-concentration salt fog layer near the sea surface and the thin background layer at high altitude, and solves the transmittance estimation error caused by the uneven density of the medium on the inclined observation path.

[0079] S230 performs depth field iterative inversion and physical dehazing restoration; it utilizes dark channel prior (DCP) theory to obtain initial transmittance estimates. Based on the transmittance physical model derived from S220, a system for determining the slant range and Euclidean distance is established. Nonlinear equations:

[0080] ;

[0081] because Due to local noise caused by image content, iterative optimization algorithms (such as guided filtering or total variational regularization) are used to optimize the depth field. Solving and smoothing constraints yields a pixel-level accurate depth distribution map.

[0082] Subsequently, the original image was restored based on an atmospheric scattering physics model; the atmospheric scattering model is described as follows:

[0083] ;

[0084] In the formula, For the observed foggy images, The haze-free radiance to be restored. The global atmospheric light value; using the calculated precise transmittance Inverse calculation of scene radiance:

[0085] ;

[0086] To prevent excessively low transmittance from causing numerical instability, a lower limit for transmittance is usually set. The final restored image It effectively removed the haze obscuring the image, restored the spectral characteristics and geometric texture of the distant *Ulva prolifera*, and provided a high-quality data source for subsequent feature extraction.

[0087] See attached document Figure 4 In step S300, to address the problem that the surface geometry of *Ulva prolifera* as a flexible float is unknown due to its wave coupling characteristics, making it difficult to accurately calculate its microscopic surface area, the physical parameters are inverted using the optical statistical features of the image itself. Specifically, this includes the following steps:

[0088] S310, Perform target region segmentation and preprocessing; in the image after dehazing and restoration... First, it is necessary to distinguish between the areas covered by seaweed and the exposed seawater areas in order to extract their features separately; then, the restored image... The color space is converted from RGB to HSV, and the unique spectral characteristics of seaweed (usually manifested as high saturation and a specific hue range) are used to set a threshold for binarization segmentation.

[0089] Specifically, the hue channel is defined as The saturation channel is Set the color range of seaweed. and saturation threshold Generate a binary mask for *Ulva prolifera*. :

[0090] ;

[0091] Correspondingly, a mask for pure seawater regions is defined. for The complement of the target area is used; to eliminate segmentation noise, morphological opening and closing operations are performed on the generated mask to obtain a clear division between the target area and the background area.

[0092] S320 is based on the statistical characteristics of glare to invert the full-spectrum mean square slope of seawater; using the Cox-Munk ocean optical model principle, the surface roughness of seawater is inferred by analyzing the brightness distribution of glare areas in the image; glare is caused by the normal direction of the sea surface micro-element reflecting sunlight into the camera lens; according to statistical optics theory, there is a direct mapping relationship between the spatial probability density function of glare brightness and the probability density function of sea surface slope.

[0093] In the restored image, detect high-brightness glare regions and calculate the histogram distribution of pixel brightness within these regions; extract the second-order central moment (i.e., variance) of this distribution. This variance characterizes the degree of dispersion in the direction of reflected light, thus reflecting the random distribution of the tilt angle of sea surface micro-element; according to the Cox-Munk model, the full-spectrum mean square slope of seawater ( The variance of the flare distribution is linearly proportional to the variance of the flare distribution.

[0094] ;

[0095] In the formula, It is only related to the zenith angle of the sun and camera observation angle The relevant geometric proportionality coefficients, which are determined by Fresnel's law of reflection and the geometry of glare, are calculated. It represents the overall roughness of the pure seawater surface unaffected by any cover under this sea condition. It includes wave energy at all scales from long waves to capillary waves, providing a physical benchmark for subsequent calculations of the damping effect of seaweed.

[0096] S330, perform fluid-structure interaction damping coefficient calculation based on texture entropy difference; since seaweed is a viscoelastic biological aggregate, it will have a significant inhibitory effect on high-frequency capillary waves when floating on the sea surface (i.e., damping effect), making the micro-roughness of the seaweed surface significantly lower than that of the surrounding exposed seawater; in order to quantify this physical phenomenon, this embodiment introduces texture entropy as a statistical descriptor to characterize the surface geometric complexity.

[0097] To eliminate the interference of uneven illumination and color differences on geometric feature analysis, the restored image is first calculated. grayscale gradient map Gradient calculation uses the Sobel operator or the Laplacian operator to extract high-frequency edge information from the image.

[0098] Subsequently, in the areas covered by the seaweed mask and seawater shield area Calculate local texture entropy; define local regions. Inner texture entropy for:

[0099] ;

[0100] In the formula, The quantization level of the gradient magnitude (e.g., 256 levels). For the first Level gradient value, For this gradient value in the region The probability density of occurrence; calculate the average texture entropy of the entire image's Ulva prolifera region. and average texture entropy of seawater region .

[0101] Constructing fluid-structure interaction damping coefficients The mapping function; the coefficient The value used to characterize the proportion of high-frequency wave energy dissipation by seaweed is [0,1]; when When , it indicates complete damping, meaning the surface approaches a theoretically smooth plane; when When the current is undamped, it means the surface follows the wave's undulations completely; the mapping relationship is defined as follows:

[0102] ;

[0103] This formula utilizes the correspondence between image features and physical states: if the texture entropy of the seaweed region is significantly less than that of the seawater region (i.e., The smaller size indicates that the surface of the seaweed is smoother than that of seawater, resulting in a stronger damping effect. The value approaches 1; conversely, if the texture entropy of the two is close, it indicates that the damping effect is weak. The value approaches 0; through this entropy difference mechanism, the hydrodynamic parameters of the target can be dynamically inverted based solely on visual features without the need for contact sensors.

[0104] S340, perform equivalent slope correction for *Ulva prolifera* micro-surface elements; based on the damping coefficient calculated in S330... Combined with the mean square slope of the full spectrum of seawater obtained by S320 inversion The microscopic geometric parameters of the *Ulva prolifera* surface were derived; the equivalent mean square slope of the *Ulva prolifera* surface was determined. Revised to:

[0105] ;

[0106] This modified formula reflects the physical coupling mechanism: the actual roughness of the seaweed cover is the result of the background wave roughness after damping attenuation.

[0107] Furthermore, a microscopic surface area correction factor is defined. Because the undulations of the microscopic surface cause the actual physical surface area to be larger than its horizontal projected area, this ratio depends on the mean square slope of the surface; according to the theory of random surface geometry, the correction factor is calculated as follows:

[0108] ;

[0109] Microscopic surface area correction factor This will be used in subsequent steps to map and restore the two-dimensional projected area on the image plane to the real physical surface area in three-dimensional space, thereby greatly improving the physical accuracy of area measurement.

[0110] See attached document Figure 5 In step S400, to address the issues of missed measurements of *Ulva prolifera* due to wave shading under long-distance non-orthophoto observations and computational divergence caused by decreased resolution of distant pixels, a multi-factor integral model that comprehensively considers macroscopic projection, microscopic roughness, and statistical shading is established. This model includes the following sub-steps:

[0111] S410 calculates the pixel-level macroscopic projected area; based on the pinhole camera imaging model, each pixel in the image actually corresponds to a trapezoidal region on the sea level in physical space; due to the characteristics of the non-orthogonal viewing angle, the area of ​​this region increases rapidly and non-linearly with increasing distance; for the coordinates in the restored image... The pixels, using the slant range Euclidean distance obtained in step S200 (here) Defined as the straight-line Euclidean distance from a pixel to the camera's optical center (i.e., the slant distance) and camera intrinsics (focal length). ), calculate its macroscopic projected area on the horizontal sea surface. According to the principles of differential geometry and projection transformation, the physical area corresponding to a pixel is:

[0112] ;

[0113] In the formula, the denominator contains The term is used to project the cross-sectional area perpendicular to the line of sight onto the horizontal sea surface; if Defined as vertical depth The formula needs to be adjusted to In this embodiment, the slant distance is used for calculation.

[0114] S420, Construct a statistical occlusion compensation factor based on wave height and incident angle; When observing the sea surface in wind and waves at low angles (large incident angles), seaweed floating in the wave troughs is easily occluded by the wave crests in front, causing the camera to fail to capture this part of the target, thus causing a systematic negative bias in area measurement; In order to correct this physical occlusion error, this embodiment introduces a statistical occlusion compensation factor. The compensation is not based on deterministic geometric restoration (because the occluded part is not visible), but on statistical inference based on the probability distribution model of the wave height field.

[0115] Assuming sea level follows a Gaussian distribution, according to Smith's occlusion theory, the probability of a view being blocked by waves is... Angle of incidence of line of sight and root mean square wave height of sea surface Related to; root mean square wave height The basic wave height can be obtained from step S100. Estimated (usually) Define the occlusion compensation factor. The reciprocal of the probability of visibility:

[0116] ;

[0117] In the formula, For complementary error functions, The mean square slope of the sea surface. This is the occlusion function; to reduce computational complexity, intermediate variables are removed in this embodiment. The occlusion compensation factor is calculated directly using the following empirical fitting formula:

[0118] ;

[0119] In the formula, The shielding intensity coefficient typically ranges from [0.5, 2.0], depending on the average wave steepness. This formula indicates that the higher the wave's ( Larger), the more tilted the observation angle ( (Larger), the more severe the occlusion effect, the greater the compensation factor. The corresponding increase; by introducing this factor, numerical compensation was made for the area of ​​Ulva prolifera that was not captured by the image sensor due to wave crest occlusion, thus correcting the systematic geometric omission error.

[0120] S430, Constructing a smooth truncation weight function based on the Sigmoid function; In UAV non-orthophoto observation, as the line-of-sight increases, the physical size represented by image pixels grows geometrically, and the signal attenuation due to atmospheric scattering becomes more severe, leading to a sharp decrease in the signal-to-noise ratio (SNR) of distant pixels; if the area integral is directly applied to the distant pixels, the tiny pixel noise will be amplified by the huge geometric projection coefficient, causing the calculation results to diverge; to solve this numerical stability problem, this embodiment abandons the traditional hard threshold distance truncation method and instead introduces a continuously differentiable Sigmoid function (S-shaped function) to construct a smooth truncation weight. .

[0121] First, define the pixel-level overall signal-to-noise ratio metric. This index is the atmospheric transmittance calculated in step S200. Positive correlation; further, with slant range Euclidean distance For the independent variable, construct the following weight function:

[0122] ;

[0123] In the formula, This is the preset effective observation distance cutoff threshold, representing the slant range Euclidean distance position when the weight function value drops to 0.5; This is the transition band width parameter, used to control the gradient slope of the weight decay with distance.

[0124] The physical meaning of this function is: for the near-distance region ( ), weight Approaching 1, retain the original calculated value; for long-distance regions ( ), weight It rapidly approaches 0, suppressing unreliable measurements; while In the nearby transition region, the weights decrease smoothly in an "S" shape; this smooth truncation mechanism ensures the continuous differentiability of the integral boundary, effectively eliminates the area jump error caused by hard truncation, and prevents numerical divergence caused by low signal-to-noise ratio data at the far end.

[0125] S440, perform a weighted integral of the total physical surface area of ​​*Ulva prolifera*; integrate all physical parameters and correction factors obtained in the previous steps to establish a multi-factor coupled area integral model; for each pixel in the restored image... Its contribution to the total area depends on parameters in four dimensions:

[0126] The target recognition result is obtained from a binary mask. Decide;

[0127] Macroscopic geometric projection, consisting of the projected area Decide;

[0128] Microscopic and occlusion corrections are made by the microscopic surface area correction factor. and occlusion compensation factor Decide;

[0129] Data confidence is determined by smoothed truncation weights. Decide.

[0130] The above factors are multiplied at the pixel level and then summed to obtain the final total physical surface area of ​​the seaweed. :

[0131] ;

[0132] In the formula, The binary mask generated in step S310 is used to remove non-Ulva prolifera areas; The horizontal projected area calculated in step S410 represents the macroscopic scale; The statistical shading compensation factor calculated in step S420 is used to compensate for wave shading loss. The micro-surface area correction factor calculated in step S340 is used to correct the area increase caused by surface roughness using the fluid-structure interaction damping coefficient. The weights calculated in step S430 are used to ensure numerical convergence.

[0133] Through this summation formula, this embodiment achieves accurate mapping from two-dimensional planar images to three-dimensional physical surface areas, and completes the measurement of the physical surface area of ​​floating seaweed under monocular vision conditions, correcting the error caused by only calculating the visual projection area.

[0134] To further verify the effectiveness of the method of the present invention, experimental examples and verification results are given below in conjunction with specific sea state data:

[0135] Experimental Objectives and Environment:

[0136] This experiment aims to verify the practical effectiveness of the core algorithm of this invention in solving the measurement distortion problems caused by non-uniform distribution of salt spray, wave occlusion, and microscopic surface undulations when using UAV oblique photography to scan large areas of *Ulva prolifera*. The experiment obtained a dataset of UAV oblique photography data from a nearshore sea area, and simultaneously acquired high-resolution orthophotos and manually collected and weighed data from on-site harvesting as the true area values.

[0137] Example 1:

[0138] This embodiment verifies the effectiveness of the vertically non-uniform atmospheric extinction model constructed in step S200. Image sets of calm sea states and wind / wave sea states accompanied by significant swells and near-surface spray were selected. The control group used a traditional uniform atmospheric model; the experimental group used the vertically non-uniform model of this invention. Experimental data are shown in Table 1.

[0139] Table 1: Comparison of Defogging Effect and Depth Field Inversion Accuracy

[0140]

[0141] As shown in Table 1, under windy and wave conditions, the traditional uniform atmospheric model overestimates the near-surface transmittance, leading to severe topographic distortion in far-end depth inversion, with an error as high as 18.5%. The model of this invention, however, reduces the depth error to 4.2% by using a fundamental wave height-driven droplet extinction term. (Refer to the attached figures.) Figure 6 and Figure 7 A visual comparison reveals that the original image Figure 6 Due to the non-uniform salt spray at the air-sea interface, visibility is low and details are blurred; after treatment by the method of this invention... Figure 7 The image became clear and transparent, and the detailed features of the seaweed in the distance were clearly restored (such as...). Figure 7 The information bar in the upper left corner displays "Fog density: <1%", indicating a significant improvement in SSIM.

[0142] Example 2:

[0143] This embodiment verifies the effectiveness of the fluid-structure interaction damping inversion and wave shading compensation in steps S300 and S400. The camera's downward tilt angles are set to 30°, 45°, and 60°, and the true value of the physical area of ​​the seaweed in the region is set to 1000 m². 2 The control group only calculated the rigid projected area of ​​two-dimensional pixels; the experimental group combined the microscopic surface area correction factor and the statistical occlusion compensation factor of this invention. Experimental data are shown in Table 2.

[0144] Table 2: Comparison of Area Measurement Accuracy at Different Observation Tilt Angles

[0145]

[0146] As shown in Table 2, as the observation angle flattens (e.g., a large tilt angle of 60°), the wave obstruction effect intensifies. Traditional methods, by ignoring wave obstruction and micro-roughness, result in severe systematic underestimation (negative bias as high as 24.7%). Figure 6 and Figure 7 As shown by the purplish-red outline, this invention can accurately segment the target area of ​​*Ulva prolifera*. Based on this, this invention effectively compensates for the area increment hidden in troughs and microscopic undulations (such as...) by introducing multi-factor weighting. Figure 7 The system outputs "estimated area: 4.19 square kilometers", and the area estimation error remains stable within 5.5% under various extreme tilt conditions, improving the accuracy and robustness of area measurement.

[0147] Example 3:

[0148] This embodiment verifies the stability of the Sigmoid smooth truncation weight function in step S430. Pixel-level integration was performed on an aerial oblique view image with a field of view extending to the horizon. The control group used the conventional hard threshold truncation method; the experimental group used the smooth truncation weight of this invention. Experimental data are shown in Table 3.

[0149] Table 3: Numerical stability test results at the far end of the area integral.

[0150]

[0151] As shown in Table 3, when processing distant low signal-to-noise ratio pixels, the traditional hard thresholding method is easily affected by the amplification of the large slant projection coefficient, resulting in numerical jumps and exponential divergence outside the truncation boundary, leading to abnormal system calculations. This invention employs a Sigmoid smooth truncation mechanism, which allows the variance to smoothly transition in the transition region and quickly converge to zero, effectively suppressing distant noise interference and ensuring the numerical stability of high-resolution, large-area continuous integration operations.

[0152] 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 estimating the area of ​​*Ulva prolifera* based on a non-orthophoto operation mode using unmanned aerial vehicles (UAVs), characterized in that... Includes the following steps: The frequency domain analysis of the raw non-orthophoto images acquired by the monocular camera mounted on the UAV flight platform is performed using the spectral energy classification and locking module. The spectral energy density of the long-wave swell frequency band is extracted, and the spectral energy density is compared with the preset calm sea state energy threshold to determine the sea state mode. The basic wave height parameter is calculated or set according to the sea state mode. Using the air-sea interface coupled depth field reconstruction module, a vertically non-uniform atmospheric extinction coefficient distribution model is constructed based on the basic wave height parameters. The atmospheric extinction coefficient distribution model is non-uniformly integrated along the line of sight to solve the scene atmospheric transmittance. The scene atmospheric transmittance is then used to perform physical dehazing on the original non-orthophoto image to generate a restored image. The texture entropy features and glare statistical features of the restored image are extracted using a fluid-structure interaction parameter inversion module. Based on the texture entropy features, the fluid-structure interaction damping coefficient of the seaweed to the waves is inverted, and the microscopic equivalent mean square slope of the seaweed surface is corrected using the glare statistical features. Specifically, the variance of the brightness distribution in the glare region of the restored image is statistically analyzed, and the full-spectrum mean square slope of the seawater is inverted using the Cox-Munk model. The full-spectrum mean square slope of the seawater is attenuated and corrected using the fluid-structure interaction damping coefficient to obtain the microscopic equivalent mean square slope of the seaweed surface. According to random surface geometry theory, a microscopic surface area correction factor is calculated using the microscopic equivalent mean square slope. This microscopic surface area correction factor characterizes the area increment of the microscopic roughness relative to the horizontal projected surface. Using a multi-factor weighted area integration module, the pixel-level macroscopic projected area, wave occlusion compensation factor, and microscopic surface area correction factor are comprehensively calculated. Combined with a sigmoid-based smoothing truncation weight function, the macroscopic projected area, wave occlusion compensation factor, microscopic surface area correction factor, and the weight values ​​calculated by the smoothing truncation weight function are weighted and accumulated to output the total physical surface area of ​​*Ulva prolifera*. The step of calculating the pixel-level macroscopic projected area includes: determining the slant distance Euclidean distance from the pixel to the camera's optical center and the line-of-sight incident angle based on a pinhole camera imaging model; and calculating the square of the slant distance Euclidean distance. The ratio of the macroscopic projected area to the normalized focal length of the camera, divided by the cosine of the line-of-sight incident angle, yields the macroscopic projected area; wherein the cosine of the line-of-sight incident angle is used to project the cross-sectional area perpendicular to the line of sight onto the horizontal sea surface; the step of calculating the wave occlusion compensation factor includes: constructing an occlusion model based on probability statistics, wherein the wave occlusion compensation factor is positively correlated with the basic wave height parameter, positively correlated with the tangent of the line-of-sight incident angle corresponding to the macroscopic projected area, and inversely proportional to the slant distance Euclidean distance of the pixel; and using the wave occlusion compensation factor to numerically compensate for the area of ​​seaweed not captured by the camera due to wave crest occlusion.

2. The method for estimating the area of ​​*Ulva prolifera* based on the non-orthophoto operation mode of unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The steps for performing frequency domain analysis on raw non-orthophoto images acquired by a monocular camera using a spectral energy hierarchical locking module include: Perform a two-dimensional discrete Fourier transform on the original non-orthophoto image to obtain a frequency domain image; An adaptive low-pass filter is constructed, wherein the cutoff frequency of the adaptive low-pass filter is set according to the projection pixel period of the minimum physical wavelength of the target surge on the image plane; The frequency domain image is filtered using the adaptive low-pass filter to extract the low-frequency components corresponding to the long-wave surge frequency band. The amplitude modulus mean of the low-frequency component is calculated as the spectral energy density; If the spectral energy density is greater than the steady sea state energy threshold, it is determined to be a wind and wave mode; If the spectral energy density is less than or equal to the steady sea state energy threshold, it is determined to be a steady mode.

3. The method for estimating the area of ​​*Ulva prolifera* based on the non-orthophoto operation mode of unmanned aerial vehicles (UAVs) according to claim 2, characterized in that, The steps for calculating or setting the basic wave height parameters based on the sea state modes include: When the wave mode is determined, the low-frequency component is inversely transformed to recover the spatial domain surge image. The pixel spacing between adjacent wave peaks on the spatial domain surge image is detected. Using perspective imaging geometry and camera tilt angle, the pixel spacing is inverted into physical wavelength. The basic wave height parameter is calculated based on the preset wave steepness coefficient. When the statically stable mode is determined, the basic wave height parameter is directly set to a preset small value so that the subsequent atmospheric extinction coefficient distribution model reverts to a standard state that only includes background aerosols.

4. The method for estimating the area of ​​*Ulva prolifera* based on the non-orthophoto operation mode of unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The steps for constructing a vertically non-uniform atmospheric extinction coefficient distribution model based on the aforementioned fundamental wave height parameters include: An extinction function driven by the vertical height variable is established, which is composed of the superposition of the background atmospheric extinction term and the droplet aerosol extinction term; The initial intensity of the droplet aerosol extinction term is positively correlated with the basic wave height parameter, and the intensity of the droplet aerosol extinction term decreases exponentially with increasing vertical height. The non-uniform integration of the atmospheric extinction coefficient distribution model along the line of sight path refers to establishing the mapping relationship between the line of sight path and the vertical height based on the line of sight incident angle of the pixel and the corresponding slant distance Euclidean distance of the pixel, and performing line integration of the extinction function along the line of sight path to obtain the scene atmospheric transmittance that varies with the slant distance Euclidean distance.

5. The method for estimating the area of ​​*Ulva prolifera* based on the non-orthophoto operation mode of unmanned aerial vehicles according to claim 1, characterized in that, The steps for inverting the fluid-structure interaction damping coefficient of seaweed on waves based on the texture entropy features include: The restored image is segmented into the seaweed target region and the seawater background region, and the local texture entropy of the seaweed target region and the local texture entropy of the seawater background region are calculated respectively. Construct a mapping relationship between the fluid-structure interaction damping coefficient and the texture entropy difference ratio; If the local texture entropy of the target area of ​​the seaweed is significantly less than that of the local texture entropy of the seawater background area, then the fluid-structure interaction damping coefficient is determined to be close to the fully damped state. If the local texture entropy of the target area of ​​*Ulva prolifera* is close to the local texture entropy of the seawater background area, then the fluid-structure interaction damping coefficient is determined to be close to an undamped state.

6. The method for estimating the area of ​​*Ulva prolifera* based on the non-orthophoto operation mode of unmanned aerial vehicles according to claim 1, characterized in that, The steps involved in combining a sigmoid-based smooth truncation weight function include: Define the effective observation distance cutoff threshold and the transition band width parameter; Construct a Sigmoid-type weight function with slant range and Euclidean distance as independent variables; The Sigmoid weighting function tends to retain its original value in the near-range region where the slant range Euclidean distance is much smaller than the effective observation distance cutoff threshold, tends to suppress to zero in the far-range region where the slant range Euclidean distance is much larger than the effective observation distance cutoff threshold, and exhibits a smooth decrease in the transition region near the effective observation distance cutoff threshold. The area calculation results of far-end low signal-to-noise ratio pixels are weighted and suppressed using the smooth truncation weight function to prevent numerical divergence.

7. The method for estimating the area of ​​*Ulva prolifera* based on the non-orthophoto operation mode of unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The output of the total physical surface area of ​​*Ulva prolifera* is achieved in the following way: For each pixel identified as *Ulva prolifera* in the restored image, the macroscopic projected area, wave occlusion compensation factor, microscopic surface area correction factor, and weight value calculated according to the smoothing truncation weight function of the pixel are multiplied by four terms to obtain the physical surface area contribution value of the pixel; the physical surface area contribution values ​​of all *Ulva prolifera* pixels are accumulated to obtain the total physical surface area of ​​*Ulva prolifera*.

Images