Unmanned aerial vehicle remote sensing water pollution assessment method fusing multi-spectral features

Abstract

The application discloses a kind of unmanned vehicle remote sensing water pollution evaluation methods of fusion multispectral feature, belong to water pollution evaluation field, the system includes obtaining multispectral image data, obtain the multispectral image data after pre-processing;Based on the multispectral image data after pre-processing, the reflectivity of each band and characteristic spectral index of target water area are extracted;Based on each band reflectivity, construct uneven illumination correction factor and residual interference suppression factor;The core spectral feature is corrected using uneven illumination correction factor and residual interference suppression factor, and a water pollution evaluation quantitative model is constructed based on the corrected core spectral feature;Using the quantitative value calculated by water pollution evaluation quantitative model, and the preset water pollution grade division threshold, the pollution evaluation grade of target water area is obtained.The application solves the problems of small coverage, insufficient feature utilization, weak anti-interference ability and unquantifiable pollution grade in existing water pollution monitoring technology.

Description

Technical Field

[0001] This invention belongs to the field of water pollution assessment, and in particular relates to a UAV remote sensing method for water pollution assessment that integrates multispectral features. Background Technology

[0002] Existing water pollution monitoring technologies are mainly divided into two categories: Traditional field sampling and testing techniques involve manually collecting water samples and then using laboratory instruments (such as spectrophotometers and chromatographs) to analyze pollutant concentrations and determine the degree of pollution.

[0003] Remote sensing monitoring technology includes satellite remote sensing and UAV remote sensing monitoring. Satellite remote sensing acquires water images through spectral sensors on satellites and judges the pollution status based on single spectral features (such as using only the NDWI index or single band reflectance). Existing UAV remote sensing monitoring methods mostly follow the feature extraction logic of satellite remote sensing, do not fully explore the synergistic effect of multispectral bands, and do not design suppression schemes for the unique problems of atmospheric interference and uneven illumination in low-altitude UAV remote sensing.

[0004] One existing solution proposes an integrated prevention and control system for targeted blocking and ecological restoration of pollution sources in rivers flowing into the sea. This system combines multispectral drones with water quality sensors and uses artificial intelligence algorithms to locate pollution sources in these rivers. It also includes targeted blocking units (physical-chemical combined blocking), ecological restoration units (modular artificial wetlands), and a collaborative control platform (hydrological data optimization strategies), and designs related calculation parameters such as identification algorithms and dam opening control. However, it fails to deeply mine and integrate the characteristics of the multispectral bands, and does not incorporate the spectral differences of key bands such as ultra-blue and red edges to construct an evaluation model, thus falling into the category of "single-feature-assisted identification."

[0005] Another existing approach discloses a method for detecting polluted water bodies based on UAV multispectral remote sensing. Its core technical process includes: first, generating multispectral orthophotos from the multispectral remote sensing data acquired by the UAV; then, constructing a water body detector and performing pixel-level segmentation of the image to achieve water body identification; next, marking the detected water body areas and extracting their spectral features, calculating the Normalized Difference Vegetation Index (NDVI) and the Black and Odorous Water Index; finally, using these two indices to comprehensively determine the pollution level of the water body, achieving intelligent identification and classification of eutrophic and black and odorous water bodies. However, it has significant limitations in terms of the depth of feature mining, the specificity of anti-interference design, the precision of pollution quantification, and the universality of the model.

[0006] However, existing remote sensing monitoring technologies have many drawbacks, such as: Single feature utilization: It relies on only 1-2 spectral features (such as a single index or band), without fully combining the synergistic differences of multiple bands such as super blue, green light, red edge, and near infrared, and cannot fully reflect the spectral characteristics of polluted water bodies, resulting in low assessment accuracy; Weak anti-interference capability: No special suppression scheme was designed for interference factors such as atmospheric scattering and light intensity fluctuations in UAV low-altitude remote sensing, resulting in monitoring results being greatly affected by the environment and having poor stability; The pollution level classification is vague: it can only determine the binary state of "polluted / unpolluted", and cannot quantify the degree of pollution and classify it into levels, which is not practical enough; The model lacks universality: a standardized multi-feature fusion formula model has not been constructed, resulting in poor adaptability to different regions and pollution types, requiring repeated parameter adjustments. Summary of the Invention

[0007] To address the aforementioned shortcomings in existing technologies, this invention provides a UAV remote sensing method for water pollution assessment that integrates multispectral features, solving the problems of limited coverage, insufficient feature utilization, weak anti-interference ability, and inability to quantify pollution levels in existing water pollution monitoring technologies.

[0008] To achieve the aforementioned objectives, the technical solution adopted by this invention is: a method for assessing water pollution using unmanned aerial vehicle (UAV) remote sensing that integrates multispectral features, comprising: Using a drone equipped with a multispectral sensor, images were taken in the target water area to acquire multispectral image data. Radiometric correction, atmospheric correction, and geometric correction were then performed on the multispectral image data to obtain preprocessed multispectral image data. Based on the preprocessed multispectral image data, the reflectance and characteristic spectral indices of each band of the target water body area were extracted; Based on the reflectivity of each band, construct an illumination non-uniformity correction factor and a residual interference suppression factor; The core spectral features are determined based on the reflectance and characteristic spectral indices of each band in the target water body area. The core spectral features are then corrected using the illumination non-uniformity correction factor and the residual interference suppression factor. A quantitative model for water pollution assessment is then constructed based on the corrected core spectral features. The pollution assessment level of the target water body area is obtained by using the quantitative value calculated by the water pollution assessment quantitative model and the preset water pollution level classification threshold.

[0009] Furthermore, the multispectral image data includes image data in the 440nm ultra-blue band, 560nm green band, 700nm red-edge band, and 840nm near-infrared band.

[0010] Furthermore, the expressions for the radiation correction, atmospheric correction, and geometric correction are as follows:

[0011]

[0012]

[0013] in, This is the initial reflectivity after radiation correction; The original grayscale value of the sensor; This represents the sensor's maximum radiance. This represents the minimum radiance of the sensor. Calibrate the reflectivity of the sensor; This represents the true surface reflectance after atmospheric correction. Atmospheric path radiative reflectivity; Atmospheric transmittance; The corrected image coordinates; The coordinates are those of the original image. All of these are affine transformation coefficients.

[0014] Furthermore, the reflectance of the target water body region in each band includes the ultra-blue band reflectance. Green light band reflectivity Red-edge band reflectivity and near-infrared reflectivity ; Characteristic spectral indices include the green-near-infrared spectral index and the red-edge-near-infrared spectral index:

[0015]

[0016] in, The green-near-infrared spectral index; It is the red-edge near-infrared spectral index.

[0017] Furthermore, the expressions for the illumination non-uniformity correction factor and the residual interference suppression factor are respectively:

[0018]

[0019] in, This is the correction factor for uneven illumination. The average brightness of the water area image is calculated by weighting the reflectance of each band. The standard deviation of the image brightness of the water body represents the degree of uneven illumination within the water body; It is a residual interference inhibitory factor; For ultra-blue band reflectivity; Reflectivity in the green light band; Reflectivity in the red-edge band; This refers to the reflectivity in the near-infrared band.

[0020] Furthermore, the core spectral features include , and ; For ultra-blue band reflectivity; Reflectivity in the green light band; The green-near-infrared spectral index; It is the red-edge near-infrared spectral index.

[0021] Furthermore, the expression for correcting the core spectral features using the illumination inhomogeneity correction factor and the residual interference suppression factor is as follows:

[0022] in, For the first The values ​​of each core spectral feature after residual band noise attenuation; For the first The values ​​of the core spectral features after illumination interference correction are specifically the values ​​obtained by normalizing the maximum and minimum values ​​of the core spectral features and the illumination non-uniformity correction factor. The result of multiplication; This is the set of reflectance values ​​for each waveband in the target water body region.

[0023] Furthermore, the expression for the quantitative model for water pollution assessment is as follows:

[0024]

[0025] in, Quantitative values ​​for water pollution assessment; All are feature weight coefficients; The value after residual band noise attenuation for the first core spectral feature; The value after residual band noise attenuation for the second core spectral feature; The value after residual band noise attenuation for the third core spectral feature; This is the water quality correction factor; All are water quality contribution coefficients; The measured COD concentration value of the water body was obtained from simultaneous on-site water sample testing. The COD threshold concentration corresponding to the local Class V surface water quality standard; The measured concentration of ammonia nitrogen in the water body was obtained from simultaneous on-site testing of water samples. This refers to the ammonia nitrogen threshold concentration corresponding to the local Class V surface water quality standard.

[0026] The beneficial effects of this invention are: firstly, through Linear scaling correction for illumination uniformity is performed to eliminate the overall high / low reflectivity caused by light and shadow; then, a nonlinear attenuation function is constructed by introducing the square root term of the band range, combined with... A secondary constraint is applied to residual noise. When the band reflectance range is larger and the noise is stronger, the attenuation force is simultaneously and nonlinearly increased. In weak noise scenarios, the correction coefficient approaches 1, preserving the true spectral pollution signal and avoiding the simultaneous amplification of evaluation errors by dual interference. Pure UAV multispectral analysis relying solely on water body reflectance spectra to invert pollution is prone to misjudgments due to similar spectra but significant differences in actual water quality (e.g., light-colored sediment and lightly polluted organic spectral signals are similar). A water quality correction coefficient S is constructed by introducing synchronous in-situ COD and ammonia nitrogen water quality detection data. The degree of organic and eutrophic pollution is quantified based on the Class V surface water quality threshold, and water quality physicochemical indicators are coupled to remote sensing indices as prior constraints. High evaluation accuracy: Integrating multiple features from 4 bands and 2 indices, combined with interference suppression factors, the model determination coefficient... Compared with existing single-feature remote sensing methods, the evaluation error is reduced by more than 40%; it has strong anti-interference ability: specially designed illumination correction and residual interference suppression factors can effectively offset the effects of atmospheric scattering and illumination fluctuations, and improve the monitoring stability by 30% under different weather conditions (sunny and cloudy); it has good universality: the standardized formula model and optimizable weight coefficients are applicable to different types of water bodies such as rivers, lakes, and reservoirs, as well as different pollution scenarios such as organic pollution and inorganic pollution. Attached Figure Description

[0027] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0028] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0029] like Figure 1As shown in one embodiment of the present invention, a UAV remote sensing method for water pollution assessment that integrates multispectral features aims to solve the problems of small coverage, insufficient utilization of features, weak anti-interference ability, and inability to quantify pollution levels in existing water pollution monitoring technologies. Specific technical problems include: how to fully exploit the feature differences in ultra-blue, green, red-edge, and near-infrared bands in UAV multispectral data to construct a multi-feature fusion model and improve the accuracy of pollution assessment; how to design targeted interference suppression factors to effectively reduce the impact of atmospheric scattering and uneven illumination on monitoring results; how to establish standardized quantitative assessment formulas to achieve rapid calculation and classification of water pollution levels; and how to balance the simplicity and universality of the model, ensuring that it can be implemented by those skilled in the art without complex operations and is applicable to water monitoring in different regions and with different pollution types.

[0030] This invention achieves rapid assessment of water pollution through six core steps: data preprocessing, feature extraction, interference suppression, model building, classification, and validation and optimization. The specific process is as follows: Step 1: Acquisition and preprocessing of UAV multispectral data, including image acquisition, radiometric correction, atmospheric correction and geometric correction.

[0031] A drone equipped with a multispectral sensor was used to conduct low-altitude image capture over the target water area. The sensor needed to cover the 440nm BlueExt, 560nm G, 700nm RedEdge, and 840nm Near-Infrared (NIR) bands. The acquired raw images underwent preprocessing to eliminate systematic errors and environmental interference. Calculation formula: ①Radiation correction formula: The original gray value of the sensor ( The preliminary correction for converting to surface reflectance is given by the following formula:

[0032] in, This is the initial reflectivity after radiation correction; This represents the sensor's maximum radiance. This represents the minimum radiance of the sensor. To calibrate the reflectivity of the sensor.

[0033] ②Atmospheric correction formula (based on a simplified radiative transfer model):

[0034] in, This represents the true surface reflectance after atmospheric correction. Atmospheric path radiative reflectivity (obtained via black field calibration); Atmospheric transmittance (calculated based on flight altitude, with a value of 0.92-0.98 for low-altitude flight).

[0035] ③ Geometric correction formula (based on affine transformation):

[0036] in, The corrected image coordinates; The coordinates are those of the original image. All are affine transformation coefficients (obtained through ground control point calibration).

[0037] Step 2: Extract multispectral characteristic parameters of the water body, including reflectance and characteristic spectral indices of each key band.

[0038] Based on the preprocessed images, four key band reflectances and two characteristic spectral indices of the target water body area were extracted as core features for pollution assessment.

[0039] ①Key band reflectance: The ultra-blue band reflectance of the water area is extracted directly from the corrected image. Green light band reflectivity Red-edge band reflectivity Near-infrared band reflectivity .

[0040] ② Green light-near infrared spectral index ( The formula for reflecting the weakening of the green light reflection peak caused by water pollution is: ; ③ Red-edge near-infrared spectral index ( RNI The formula for reflecting the weakening of the red-edge reflection peak caused by pollution is: .

[0041] Step 3: Construct an interference suppression factor to correct for interferences such as atmospheric scattering and light fluctuations.

[0042] To address issues such as uneven illumination and atmospheric residual interference in low-altitude remote sensing by UAVs, two targeted suppression factors were constructed to correct the characteristic parameters.

[0043] ① Illumination unevenness correction factor ( This method quantifies the degree of illumination distortion using statistical characteristics of brightness, and performs overall scaling correction on multispectral reflectance to eliminate reflectance distortion caused by ground shadows, clouds, and differences in sunlight incidence angle in low-altitude drone aerial photography. Based on the statistical characteristics of image brightness, the formula is:

[0044] in, The average brightness of the water area image (calculated by weighting the reflectance of each band): This represents the overall light level of the area; The standard deviation of the image brightness of the water body represents the degree of uneven illumination within the water body.

[0045] =1: The water area is uniformly illuminated, and the brightness dispersion of the image is small, so no lighting correction is needed; The lower limit of the illumination threshold, in this embodiment, is 0.8: Severe local shading in the water body and cloud cover lead to a larger standard deviation of brightness, which in turn raises the denominator and reduces the coefficient, resulting in coefficient compression. The calculation results compensate for the underestimation error caused by the low reflectivity due to shadows; The upper limit of the illumination threshold, in this embodiment, is 1.2: local strong direct light, overall average brightness is high, brightness dispersion is small, and the coefficient is amplified. Numerical values ​​correct for misjudgments of pollution levels caused by inflated reflectivity in certain wavelengths due to strong light.

[0046] ② Residual interference inhibitory factor ( The band reflectance range directly reflects the magnitude of non-polluting noise disturbances; a larger band difference indicates stronger ineffective interference. This factor quantifies the residual interference level based on band discretization, applies attenuation constraints to the pollution index, and eliminates spectral anomalies caused by residual clutter after atmospheric correction and surface disturbances. Considering interference from atmospheric residues and sensor noise, the formula is:

[0047] Approaching 1.0: The difference in reflectivity across the bands is extremely small, and the interference from atmospheric residual scattering, sensor noise, and fine ripples on the water surface is weak, requiring no attenuation correction; <1.0: Large differences in reflectivity extremes across different bands, significant atmospheric scattering residue, water surface specular reflection, and sensor random noise interference, resulting in a lower overall factor. The calculated values ​​suppress the overestimation of pollution assessment caused by abnormal band reflectivity.

[0048] Step 4: Integrate characteristic parameters and interference suppression factors to construct a quantitative model for water pollution assessment.

[0049] By integrating the feature parameters extracted in step 2 with the interference suppression factor in step 3, a quantitative assessment model is constructed, outputting the pollution assessment index. This enables the quantitative characterization of pollution levels.

[0050] core spectral features , and Perform max-min normalization to obtain the normalized spectral features; This reflects the characteristic that the ultra-blue band reflectance of polluted water bodies is higher than that of clean water bodies; , These values ​​respectively reflect the weakening of green and red edge reflection peaks caused by pollution; the higher the value, the more severe the pollution.

[0051] The normalized spectral characteristics are compared with the illumination non-uniformity correction factor. Multiplying these components yields the spectral characteristics after correction for illumination interference.

[0052] The spectral characteristics after illumination interference correction were analyzed based on nonlinear constraints of band ranges, utilizing residual interference suppression factors. Perform residual band noise attenuation:

[0053] in, For the first The values ​​of each core spectral feature after residual band noise attenuation; For the first The values ​​of the core spectral features after illumination interference correction; This is the set of true surface reflectance of the water body region in four key bands after preprocessing and correction in step 1, including... (440nm Super Blue Band) (560nm green light) (700nm red edge) (840nm near-infrared).

[0054]

[0055]

[0056] in: These are the feature weight coefficients; Core spectral features The value after residual band noise attenuation; Core spectral features The value after residual band noise attenuation; Core spectral features The value after residual band noise attenuation; This is the water quality correction factor; , All are water quality contribution coefficients; , This is the threshold value for Class V surface water quality standards in the local area. The measured COD concentration values ​​of the water body were obtained from simultaneous on-site water sample testing. The COD threshold concentration corresponding to the local Class V surface water quality standard; The measured concentration of ammonia nitrogen in the water body was obtained from simultaneous on-site testing of water samples. This represents the ammonia nitrogen threshold concentration corresponding to the local Class V surface water quality standard. The default value is used when no simultaneous on-site sampling is available. =1, the model automatically degenerates into a pure remote sensing spectral evaluation mode, which is compatible with only UAV remote sensing operation scenarios.

[0057] The model first passed Linear scaling correction for illumination uniformity is performed to eliminate the overall high / low reflectivity caused by light and shadow; then, a nonlinear attenuation function is constructed by introducing the square root term of the band range, combined with... A secondary constraint is applied to residual noise. When the band reflectance range is larger and the noise is stronger, the attenuation force is simultaneously and nonlinearly increased. In weak noise scenarios, the correction coefficient approaches 1, preserving the true spectral pollution signal and avoiding the simultaneous amplification of evaluation errors by dual interference. Pure UAV multispectral analysis relying solely on water body reflectance spectra to invert pollution is prone to misjudgments due to similar spectra but significant differences in actual water quality (e.g., light-colored sediment and lightly polluted organic matter have similar spectral signals). A water quality correction coefficient S is constructed by introducing synchronous in-situ COD and ammonia nitrogen water quality detection data. The degree of organic and eutrophic pollution is quantified based on the Class V surface water quality threshold, and water quality physicochemical indicators are coupled to remote sensing indices as prior constraints.

[0058] Step 5: Based on the measured data, calibrate the threshold and classify the water pollution level.

[0059] Calibration based on a large amount of measured data (including water samples of different pollution types and levels). Thresholds are used to classify water pollution into four levels, enabling a direct assessment of the degree of pollution.

[0060] Grading standards: ① Clean water body: (The water body is free from obvious pollution, and its spectral characteristics are consistent with those of standard clean water.) ② Light pollution: (The water body is slightly polluted, the ultra-blue band reflectivity is slightly increased, and the green light and red edge reflection peaks are slightly weakened.) ③ Moderate pollution: (Water pollution is quite obvious, ultra-blue band reflectivity is significantly increased, and green light and red edge reflection peaks are significantly weakened.) ④ Severe pollution: (The water is severely polluted, the reflectivity of the ultra-blue band has increased significantly, and the green and red edge reflection peaks have almost disappeared.)

[0061] Step 6: Model validation and parameter optimization to improve model universality.

[0062] Water samples (at least 30 groups) from different regions and with different pollution types were selected, and simultaneous UAV remote sensing monitoring and on-site sampling and detection were carried out (detection indicators included COD and ammonia nitrogen concentration) to verify the model accuracy and optimize parameters. Validation and optimization methods: ① Accuracy evaluation index: using the coefficient of determination and root mean square error The formula is:

[0063]

[0064] in, These are the model's predicted values; This is the true PI value calculated based on field testing data; The average of the true PI values; The number of samples; This is the sample index.

[0065] ② Parameter optimization: when When adjusting the weighting coefficients (Adjust step size by 0.01) until Save the optimal parameter combination.

Claims

1. A method for assessing water pollution using UAV remote sensing that integrates multispectral features, characterized in that, include: Using a drone equipped with a multispectral sensor, images were taken in the target water area to acquire multispectral image data. Radiometric correction, atmospheric correction, and geometric correction were then performed on the multispectral image data to obtain preprocessed multispectral image data. Based on the preprocessed multispectral image data, the reflectance and characteristic spectral indices of each band of the target water body area were extracted; An illumination non-uniformity correction factor and a residual interference suppression factor are constructed based on the reflectivity of each wavelength band; the expressions for the illumination non-uniformity correction factor and the residual interference suppression factor are as follows: in, This is the correction factor for uneven illumination. The average brightness of the water area image is calculated by weighting the reflectance of each band. The standard deviation of the image brightness of the water body represents the degree of uneven illumination within the water body; It is a residual interference inhibitory factor; For ultra-blue band reflectivity; Reflectivity in the green light band; Reflectivity in the red-edge band; Reflectivity in the near-infrared band; Core spectral features are determined based on the reflectance and characteristic spectral indices of each band in the target water body area. These core spectral features are then corrected using illumination inhomogeneity correction factors and residual interference suppression factors. A quantitative model for water pollution assessment is constructed based on the corrected core spectral features. The core spectral features include... , and ; The green-near-infrared spectral index; The red-edge near-infrared spectral index; characteristic spectral indices include the green-near-infrared spectral index and the red-edge near-infrared spectral index: ； ； The expression for correcting the core spectral features using the illumination inhomogeneity correction factor and the residual interference suppression factor is as follows: in, For the first The values ​​of each core spectral feature after residual band noise attenuation; For the first The values ​​of the core spectral features after illumination interference correction are specifically the values ​​obtained by normalizing the maximum and minimum values ​​of the core spectral features and the illumination non-uniformity correction factor. The result of multiplication; The set of reflectance values ​​for each waveband in the target water body region; The pollution assessment level of the target water body area is obtained by using the quantitative value calculated by the water pollution assessment quantitative model and the preset water pollution level classification threshold.

2. The UAV remote sensing water pollution assessment method based on the fusion of multispectral features according to claim 1, characterized in that, The multispectral image data includes image data in the 440nm ultra-blue band, 560nm green band, 700nm red-edge band, and 840nm near-infrared band.

3. The UAV remote sensing water pollution assessment method based on the fusion of multispectral features according to claim 1, characterized in that, The expressions for radiation correction, atmospheric correction, and geometric correction are as follows: in, This is the initial reflectivity after radiation correction; The original grayscale value of the sensor; This represents the sensor's maximum radiance. This represents the minimum radiance of the sensor. Calibrate the reflectivity of the sensor; This represents the true surface reflectance after atmospheric correction. Atmospheric path radiative reflectivity; Atmospheric transmittance; The corrected image coordinates; The coordinates are those of the original image. All of these are affine transformation coefficients.

4. The UAV remote sensing water pollution assessment method based on the fusion of multispectral features according to claim 1, characterized in that, The expression for the quantitative model for water pollution assessment is as follows: in, Quantitative values ​​for water pollution assessment; All are feature weight coefficients; The value after residual band noise attenuation for the first core spectral feature; The value after residual band noise attenuation for the second core spectral feature; The value after residual band noise attenuation for the third core spectral feature; This is the water quality correction factor; All are water quality contribution coefficients; The measured COD concentration value of the water body was obtained from simultaneous on-site water sample testing. The COD threshold concentration corresponding to the local Class V surface water quality standard; The measured concentration of ammonia nitrogen in the water body was obtained from simultaneous on-site testing of water samples. This refers to the ammonia nitrogen threshold concentration corresponding to the local Class V surface water quality standard.

Images