Unmanned aerial vehicle-based reservoir ecological environment safety monitoring and early warning system

Abstract

The invention belongs to the technical field of reservoir ecological monitoring, and particularly relates to a reservoir ecological environment safety monitoring and early warning system based on an unmanned aerial vehicle, which comprises an autonomous motion control module, a jitter rule optimization module and a reservoir ecological monitoring module. According to the scheme, through the autonomous motion control module and the jitter rule optimization module, a PID controller is adopted to dynamically adjust the attitude of the unmanned aerial vehicle and sensor parameters, a BANet model is optimized to eliminate dynamic fuzziness, Kalman filtering is adopted to compensate high-frequency jitter, and multi-frame fusion is adopted to enhance and compensate low-frequency jitter. Stable acquisition of spectral data of the unmanned aerial vehicle in motion modes of target surrounding, tracking and the like is realized; through a reservoir ecological monitoring module, a random forest model and clustering fusion analysis are adopted, and a pollutant input path is traced through a historical spectrum difference index, so that automatic positioning of a pollution hot spot area and quantitative evaluation of a nutrition state are realized.

Description

Technical Field

[0001] This invention relates to the field of reservoir ecological monitoring technology, specifically to a reservoir ecological environment safety monitoring and early warning system based on unmanned aerial vehicles (UAVs). Background Technology

[0002] The reservoir ecological environment safety monitoring and early warning system based on drones is a system that uses drones equipped with spectral sensors as a technical means to provide early warning of the ecological environment safety of reservoirs.

[0003] Among the existing approximate solutions, such as CN117330714B, a regional environmental monitoring and early warning system and method based on big data, this solution addresses the technical problems of failing to detect local pollution exceeding standards due to incomplete monitoring coverage, blind allocation of drone numbers, and the inability of existing technologies to guarantee the comprehensiveness and accuracy of environmental monitoring data. This solution combines fixed stations with drones, only activating drone collaborative monitoring when pollution exceeds standards in areas not covered by the stations. It establishes a linear regression model, considering flight time and the area of ​​the uncovered area, to calculate the required number of drones, ensuring that drones can complete the monitoring in one go. This technical solution achieves the technical effect of compensating for the limitations of fixed station monitoring range. However, it still has the technical problem that existing drone monitoring systems suffer from blurred and jittery images acquired by spectral sensors in complex flight environments (such as wind speed changes and mechanical vibrations), making it difficult to stably obtain high-precision spectral data.

[0004] For example, CN112580893B describes a multi-UAV atmospheric monitoring path planning method based on an improved genetic algorithm. This method addresses the technical problem of multi-UAV monitoring path optimization in mixed areas of elevated and non-elevated pollutant point sources. It incorporates three-dimensional elevation information, i.e., point source height, into the path planning, establishes an objective function to optimize the total length of the monitoring path, flight altitude stability, and flight path turning angle, constrains the optimization conditions, and optimizes the objective function using an improved genetic algorithm. This ultimately yields the optimal path for guiding UAVs to perform monitoring tasks, achieving improved monitoring accuracy. However, it suffers from the inability to directly apply this technology to reservoir ecological monitoring applications. Furthermore, traditional reservoir ecological monitoring relies on manual sampling or single spectral index analysis, which cannot quickly locate pollution sources, quantify nutrient status, or provide dynamic early warnings. Summary of the Invention

[0005] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides a reservoir ecological environment safety monitoring and early warning system based on unmanned aerial vehicles (UAVs). Addressing the technical problem of existing UAV monitoring systems experiencing blurred and jittery images from spectral sensors under complex flight environments (such as wind speed changes and mechanical vibrations), making it difficult to stably acquire high-precision spectral data, this solution utilizes an autonomous motion control module and a jitter pattern optimization module. It employs a PID controller to dynamically adjust the UAV's attitude and sensor parameters, optimizes the BANet model to eliminate dynamic blur, uses Kalman filtering to compensate for high-frequency jitter, and employs multi-frame fusion enhancement to compensate for low-frequency jitter. This achieves stable acquisition of spectral data by the UAV in target orbiting and tracking motion modes. Furthermore, addressing the limitations of directly applying atmospheric monitoring technology to reservoir ecological monitoring, and the fact that traditional reservoir ecological monitoring relies on manual sampling or single spectral index analysis, which cannot quickly locate pollution sources, quantify nutrient status, or provide dynamic early warnings, this solution combines spectral sensor data and key indices through a reservoir ecological monitoring module. It uses a random forest model and K-means clustering fusion analysis for spatial clustering to trace pollution sources and historical spectral difference indices to trace pollutant input paths, achieving automatic location of pollution hotspots and quantitative assessment of nutrient status.

[0006] The technical solution adopted by the present invention is as follows: The present invention provides a reservoir ecological environment safety monitoring and early warning system based on unmanned aerial vehicles (UAVs). The reservoir ecological environment safety monitoring and early warning system based on unmanned aerial vehicles (UAVs) includes an autonomous motion control module, a jitter pattern optimization module, and a reservoir ecological monitoring module.

[0007] The autonomous motion control module employs an autonomous motion control method to guide the movement of the UAV and the spectral sensor.

[0008] The jitter pattern optimization module uses a jitter pattern optimization method to compensate for high-frequency and low-frequency jitter.

[0009] The reservoir ecological monitoring module uses reservoir ecological monitoring methods to locate pollution hotspots, trace pollutant input paths, and issue alarms.

[0010] Furthermore, the autonomous motion control method specifically includes the following steps:

[0011] Step A1: Predefined, specifically, the motion trajectory of the UAV and the spectral sensor in 3D space is defined using control parameters. A PID controller is used to control the control parameters. Each control parameter is controlled by a corresponding PID controller. The control parameters include instantaneous UAV speed, instantaneous UAV yaw angle, instantaneous UAV pitch angle, spectral sensor focal length, and spectral sensor angle.

[0012] Step A2: Constructing an error signal. Specifically, the current motion type of the spectral sensor is obtained, which describes the type of relative motion between the UAV and the target area. An error signal is constructed by calculating the deviation between the position and area of ​​the target area in the current frame captured by the spectral sensor and the expected value.

[0013] Furthermore, the jitter pattern optimization method specifically includes the following steps:

[0014] Step B1: Jitter Repair, used to balance the jitter repair effect and repair speed of the spectral sensor on the UAV. Specifically, the blur caused by the jitter of the spectral sensor due to the influence of the flight motion mode presents a periodic amplitude change. The BANet model is used to eliminate dynamic blur. The jitter direction of the spectral sensor is extracted according to the motion type, and the strip pooling kernel corresponding to the jitter direction is added to the BANet model.

[0015] Step B2: Sensor collaboration, specifically, using a gyroscope and accelerometer to collect actual motion data from the spectral sensor, estimating the jitter direction change through Kalman filtering, comparing with the actual motion data, generating a high-frequency jitter compensation signal, synchronizing the high-frequency jitter compensation signal with the frame rate of the spectral sensor, and using the high-frequency jitter compensation signal to dynamically adjust the spectral sensor image through affine transformation, cropping the image boundary, and compensating for high-frequency jitter;

[0016] Step B3: Multi-frame fusion enhancement to compensate for low-frequency jitter. Specifically, take m consecutive image sequences captured by the spectral sensor, calculate the optical flow field of adjacent frames, where the optical flow field represents the pixel displacement vector, align the optical flow field coordinate system between image sequences, divide each image frame into blocks, use CNN to calculate the local blur of each block, use the mean of the local blur as the blur score of each frame, use the softmax function to add weights to the continuous image sequences according to the blur score, and weightedly fuse the continuous image sequences to obtain a fused image. Use the fused image as input to the BANet model, and finally output the deblurred image.

[0017] Furthermore, the reservoir ecological monitoring method specifically includes the following steps:

[0018] Step C1: Obtain remote sensing data through UAV, filter the spectral bands of the remote sensing data, extract the spectral bands to analyze spectral features, obtain the water quality parameters of the reservoir monitoring point, the water quality parameters include total phosphorus concentration, total nitrogen concentration, chlorophyll and turbidity, perform preprocessing operations on the remote sensing data, extract the reservoir area and convert it into polygon vector data;

[0019] Step C2: Calculate key spectral indices. Specifically, extract the spectral response characteristics before and after historical pollution events in the reservoir's ecological environment, identify water quality change patterns, and the key spectral indices include the normalized water index, the normalized vegetation index, and the blue-green band ratio. The spectral changes of the key spectral indices before and after historical pollution events in the reservoir's ecological environment are recorded as difference indices.

[0020] The formula for calculating the normalized water index is as follows: ;

[0021] In the formula, Indicates the normalized water quality index. Indicates green band reflectivity. Indicates near-infrared reflectance;

[0022] The formula for calculating the normalized vegetation index is as follows: ;

[0023] In the formula, Represents the normalized vegetation index. Indicates the reflectivity of the red band;

[0024] The formula for calculating the blue-green band ratio is: ;

[0025] In the formula, Indicates blue band reflectivity. Indicates the blue-green band ratio;

[0026] Step C3: Spatial classification of water quality features, used for model training and water quality feature identification. Specifically, a random forest algorithm model is used to establish the mapping relationship between spectral features and water quality parameters. An attention mechanism is used to optimize the weight allocation of spectral bands to simulate the spatial distribution of water quality parameters. The spectral bands, water quality parameters, and key spectral indices are used as training data. The training data is divided into unsupervised groups using K-means clustering. Finally, the random forest algorithm model outputs eutrophication regions and turbidity regions.

[0027] Step C4: Parameter correlation verification, used to verify the correlation of parameters for dividing eutrophication areas. Specifically, the consistency between the output results of the random forest algorithm model and the K-means clustering results is quantified by the error matrix and Kappa coefficients. The stages of reservoir pollution accumulation are verified by comparing eutrophication areas with the difference index through time series analysis.

[0028] Step C5: Quantitative assessment of trophic status, used to quantitatively assess the nutrient enrichment level and ecological health status of the reservoir. Specifically, the reservoir trophic status index is calculated based on the total phosphorus concentration. The calculation formula is as follows: ;

[0029] In the formula, Indicates the reservoir's trophic status index. Indicates the scaling factor. , Indicates the reference offset. , Indicates total phosphorus concentration. This represents the logarithmically transformed mean of the total phosphorus concentration within the region. This represents the standard deviation of the concentration within the region after logarithmic transformation;

[0030] Step C6: Correct the spectral bands to correct and confirm the indicative ability of spectral features to indicate the pollution status. Specifically, clarify the nutrient pollution level of the reservoir, determine the nutrient status of the reservoir based on the reservoir nutrient status index, and make a quantitative comparison between the nutrient pollution level and spectral reflectance to correct the key spectral band selection.

[0031] Step C7: Pollution source tracing. Specifically, based on the polygonal vector data of the reservoir, the neighborhood radius and minimum number of points are set. Spatial clustering algorithm is used to perform aggregate analysis on the areas corresponding to each nutrient pollution level to locate pollution hotspots, trace the input path of pollutants, and issue an alarm.

[0032] The beneficial effects achieved by the present invention using the above solution are as follows:

[0033] (1) In view of the technical problem that existing UAV monitoring systems suffer from blurred and jittery images acquired by spectral sensors in complex flight environments (such as wind speed changes and mechanical vibrations), making it difficult to stably acquire high-precision spectral data, this solution uses an autonomous motion control module and a jitter law optimization module. It adopts a PID controller to dynamically adjust the UAV attitude and sensor parameters, optimizes the BANet model to eliminate dynamic blur, uses Kalman filtering to compensate for high-frequency jitter, and uses multi-frame fusion to enhance and compensate for low-frequency jitter. This achieves stable acquisition of spectral data of UAV in motion modes such as target circling and tracking.

[0034] (2) In view of the technical problems that atmospheric monitoring technology cannot be directly applied to reservoir ecological monitoring, and that traditional reservoir ecological monitoring relies on manual sampling or single spectral index analysis, which cannot quickly locate pollution sources, quantify nutrient status and provide dynamic early warning, this solution combines spectral sensor data and key indices through the reservoir ecological monitoring module, and uses random forest model and K-means clustering fusion analysis to achieve pollution source tracing through spatial clustering. By tracing the pollutant input path through historical spectral difference index, the automatic location of pollution hotspots and quantitative assessment of nutrient status are realized. Attached Figure Description

[0035] Figure 1 The present invention provides a module connection diagram for a reservoir ecological environment safety monitoring and early warning system based on unmanned aerial vehicles (UAVs).

[0036] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0038] Example 1: See Figure 1 This embodiment provides a reservoir ecological environment safety monitoring and early warning system based on unmanned aerial vehicles (UAVs). The UAV-based reservoir ecological environment safety monitoring and early warning system includes an autonomous motion control module, a jitter pattern optimization module, and a reservoir ecological monitoring module.

[0039] The autonomous motion control module employs an autonomous motion control method to guide the movement of the UAV and the spectral sensor.

[0040] The jitter pattern optimization module uses a jitter pattern optimization method to compensate for high-frequency and low-frequency jitter.

[0041] The reservoir ecological monitoring module uses reservoir ecological monitoring methods to locate pollution hotspots, trace pollutant input paths, and issue alarms.

[0042] Example 2: See Figure 1 This embodiment is based on the above embodiment, and the autonomous motion control method specifically includes the following steps:

[0043] Step A1: Predefined, specifically, the motion trajectory of the UAV and the spectral sensor in 3D space is defined using control parameters. A PID controller is used to control the control parameters. Each control parameter is controlled by a corresponding PID controller. The control parameters include instantaneous UAV speed, instantaneous UAV yaw angle, instantaneous UAV pitch angle, spectral sensor focal length, and spectral sensor angle.

[0044] Step A2: Constructing an error signal. Specifically, the current motion type of the spectral sensor is obtained, which describes the type of relative motion between the UAV and the target area. An error signal is constructed by calculating the deviation between the position and area of ​​the target area in the current frame captured by the spectral sensor and the expected value.

[0045] Example 3: See Figure 1 This embodiment is based on the above embodiment, and the jitter pattern optimization method specifically includes the following steps:

[0046] Step B1: Jitter Repair, used to balance the jitter repair effect and repair speed of the spectral sensor on the UAV. Specifically, the blur caused by the jitter of the spectral sensor due to the influence of the flight motion mode presents a periodic amplitude change. The BANet model is used to eliminate dynamic blur. The jitter direction of the spectral sensor is extracted according to the motion type, and the strip pooling kernel corresponding to the jitter direction is added to the BANet model.

[0047] Step B2: Sensor collaboration, specifically, using a gyroscope and accelerometer to collect actual motion data from the spectral sensor, estimating the jitter direction change through Kalman filtering, comparing with the actual motion data, generating a high-frequency jitter compensation signal, synchronizing the high-frequency jitter compensation signal with the frame rate of the spectral sensor, and using the high-frequency jitter compensation signal to dynamically adjust the spectral sensor image through affine transformation, cropping the image boundary, and compensating for high-frequency jitter;

[0048] Step B3: Multi-frame fusion enhancement to compensate for low-frequency jitter. Specifically, take m consecutive image sequences captured by the spectral sensor, calculate the optical flow field of adjacent frames, where the optical flow field represents the pixel displacement vector, align the optical flow field coordinate system between image sequences, divide each image frame into blocks, use CNN to calculate the local blur of each block, use the mean of the local blur as the blur score of each frame, use the softmax function to add weights to the continuous image sequences according to the blur score, and weightedly fuse the continuous image sequences to obtain a fused image. Use the fused image as input to the BANet model, and finally output the deblurred image.

[0049] Example 4: See Figure 1 This embodiment is based on the above embodiment, and the reservoir ecological monitoring method specifically includes the following steps:

[0050] Step C1: Obtain remote sensing data through UAV, filter the spectral bands of the remote sensing data, extract the spectral bands to analyze spectral features, obtain the water quality parameters of the reservoir monitoring point, the water quality parameters include total phosphorus concentration, total nitrogen concentration, chlorophyll and turbidity, perform preprocessing operations on the remote sensing data, extract the reservoir area and convert it into polygon vector data;

[0051] Step C2: Calculate key spectral indices. Specifically, extract the spectral response characteristics before and after historical pollution events in the reservoir's ecological environment, identify water quality change patterns, and the key spectral indices include the normalized water index, the normalized vegetation index, and the blue-green band ratio. The spectral changes of the key spectral indices before and after historical pollution events in the reservoir's ecological environment are recorded as difference indices.

[0052] The formula for calculating the normalized water index is as follows: ;

[0053] In the formula, Indicates the normalized water quality index. Indicates green band reflectivity. Indicates near-infrared reflectance;

[0054] The formula for calculating the normalized vegetation index is as follows: ;

[0055] In the formula, Represents the normalized vegetation index. Indicates the reflectivity of the red band;

[0056] The formula for calculating the blue-green band ratio is: ;

[0057] In the formula, Indicates blue band reflectivity. Indicates the blue-green band ratio;

[0058] Step C3: Spatial classification of water quality features, used for model training and water quality feature identification. Specifically, a random forest algorithm model is used to establish the mapping relationship between spectral features and water quality parameters. An attention mechanism is used to optimize the weight allocation of spectral bands to simulate the spatial distribution of water quality parameters. The spectral bands, water quality parameters, and key spectral indices are used as training data. The training data is divided into unsupervised groups using K-means clustering. Finally, the random forest algorithm model outputs eutrophication regions and turbidity regions.

[0059] Step C4: Parameter correlation verification, used to verify the correlation of parameters for dividing eutrophication areas. Specifically, the consistency between the output results of the random forest algorithm model and the K-means clustering results is quantified by the error matrix and Kappa coefficients. The stages of reservoir pollution accumulation are verified by comparing eutrophication areas with the difference index through time series analysis.

[0060] Step C5: Quantitative assessment of trophic status, used to quantitatively assess the nutrient enrichment level and ecological health status of the reservoir. Specifically, the reservoir trophic status index is calculated based on the total phosphorus concentration. The calculation formula is as follows: ;

[0061] In the formula, Indicates the reservoir's trophic status index. Indicates the scaling factor. , Indicates the reference offset. , represents the total phosphorus concentration, represents the mean value of the logarithmically transformed total phosphorus concentration within the region, represents the standard deviation of the logarithmically transformed concentration within the region;

[0062] Step C6: Modify the spectral bands, which is used to confirm the indication ability of spectral features for the pollution status. Specifically, clarify the nutritional pollution level of the reservoir, judge the nutritional status of the reservoir according to the reservoir nutritional status index, and make a quantitative comparison between the nutritional pollution level and the spectral reflectance to modify the screening of key spectral bands;

[0063] Step C7: Trace the pollution source. Specifically, according to the polygon vector data of the reservoir, set the neighborhood radius and the minimum number of points, perform aggregation analysis on the regions corresponding to each nutritional pollution level through the spatial clustering algorithm, locate the pollution hotspots, trace the pollutant input path, and issue an alarm.

[0064] Example Five: Refer to Figure 1 , in this example, based on the above example, in step B1, the addition of the strip pooling kernel corresponding to the dithering direction specifically refers to adding kernels in the 30° and 60° directions on the basis of the original horizontal and vertical kernels.

[0065] Example Six: Refer to Figure 1 , in this example, based on the above example, in step B3, m = 3.

[0066] Example Seven: Refer to Figure 1 , in this example, based on the above example, in step C1, the screening of the spectral bands of the remote sensing data specifically refers to adding kernels in the 30° and 60° directions on the basis of the original horizontal and vertical kernels.

[0067] Example Eight: Refer to Figure 1 , in this example, based on the above example, in step C4, during the stage of verifying the pollution accumulation in the reservoir, at the same time, the difference index of any region A shows a significant positive value, indicating that the algae growth intensifies, which is consistent with the expansion of the eutrophic region identified by the random forest algorithm model, confirming the pollution accumulation effect in this region.

[0068] Example Nine: Refer to Figure 1 , in this example, based on the above example, in step C6, the nutritional pollution level is specifically: hyper-eutrophication: RTLI-(TP)>67; eutrophication: 59<RTLI-(TP)≤63; mesotrophication: 52<RTLI-(TP)≤59; oligotrophication: RTLI-(TP)≤52.

[0069] Example Ten: Refer to Figure 1This embodiment is based on the above embodiment. In step C7, the neighborhood radius is set to 50m and the minimum number of points is set to 5 pixel units.

[0070] Example 11: See Figure 1 This embodiment is based on the above embodiment. In step A2, when the motion type of the spectral sensor is target circling, the error signal formula is as follows: ;

[0071] In the formula, Indicates the error signal. , This indicates the center coordinates of the target region in the image. Indicates the area of ​​the target region. This represents the area of ​​the desired target region. Indicates the yaw angle. Represents a symbolic function;

[0072] Example 12: See Figure 1 This embodiment is based on the above embodiment. In step A2, when the motion type of the spectral sensor is target tracking, the error signal formula is as follows: ;

[0073] Step A3: Generate control commands based on the error signal to drive the PID controller to adjust the control parameters in real time. The formula used is as follows: ;

[0074] In the formula, This represents the control command vector of the PID controller. Indicates the error signal. , and The coefficients representing proportions, integrals, and derivatives.

[0075] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0076] 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.

[0077] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A reservoir ecological environment safety monitoring and early warning system based on unmanned aerial vehicles (UAVs), characterized in that, It includes an autonomous motion control module, a jitter pattern optimization module, and a reservoir ecological monitoring module; The autonomous motion control module employs an autonomous motion control method to guide the movement of the UAV and the spectral sensor. The jitter pattern optimization module uses a jitter pattern optimization method to compensate for high-frequency and low-frequency jitter. The reservoir ecological monitoring module uses reservoir ecological monitoring methods to locate pollution hotspots, trace pollutant input paths, and issue alarms.

2. The reservoir ecological environment safety monitoring and early warning system based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The autonomous motion control method specifically includes the following steps: Step A1: Predefined, specifically, the motion trajectory of the UAV and spectral sensor in 3D space is defined using control parameters, and the control parameters are controlled by a PID controller, with each control parameter controlled by a corresponding PID controller; Step A2: Constructing an error signal, specifically, obtaining the current motion type of the spectral sensor, and constructing an error signal by calculating the deviation between the position and area of ​​the target region in the current frame captured by the spectral sensor and the expected value.

3. The reservoir ecological environment safety monitoring and early warning system based on unmanned aerial vehicles (UAVs) according to claim 2, characterized in that, The jitter pattern optimization method specifically includes the following steps: Step B1: Jitter Repair, specifically, using the BANet model to eliminate dynamic blur, extracting the jitter direction of the spectral sensor according to the motion type, and adding the strip pooling kernel corresponding to the jitter direction to the BANet model; Step B2: Sensor collaboration, specifically, using a gyroscope and accelerometer to collect actual motion data from the spectral sensor, estimating the jitter direction change through Kalman filtering, comparing with the actual motion data, generating a high-frequency jitter compensation signal, synchronizing the high-frequency jitter compensation signal with the frame rate of the spectral sensor, and using the high-frequency jitter compensation signal to dynamically adjust the spectral sensor image through affine transformation, cropping the image boundary, and compensating for high-frequency jitter; Step B3: Multi-frame fusion enhancement, specifically, takes a continuous sequence of m frames of images captured by a spectral sensor, calculates the optical flow field of adjacent frames, where the optical flow field represents the pixel displacement vector, aligns the optical flow field coordinate system between the image sequences, divides each frame into blocks, uses a CNN to calculate the local blur of each block, uses the mean of the local blur as the blur score of each frame, uses the softmax function to add weights to the continuous image sequence according to the blur score, and weights the continuous image sequences to obtain a fused image. The fused image is used as the input of the BANet model, and the final output is a deblurred image.

4. The reservoir ecological environment safety monitoring and early warning system based on unmanned aerial vehicles (UAVs) according to claim 3, characterized in that, The reservoir ecological monitoring method specifically includes the following steps: Step C1: Obtain remote sensing data through UAV, filter the spectral bands of the remote sensing data, extract the spectral bands to analyze spectral features, obtain the water quality parameters of the reservoir monitoring points, perform preprocessing operations on the remote sensing data, extract the reservoir area and convert it into polygon vector data; Step C2: Calculate key spectral indices. Specifically, extract the spectral response characteristics before and after historical pollution events in the reservoir's ecological environment, identify water quality change patterns, and the key spectral indices include the normalized water index, the normalized vegetation index, and the blue-green band ratio. The spectral changes of the key spectral indices before and after historical pollution events in the reservoir's ecological environment are recorded as difference indices. Step C3: Spatial classification of water quality features. Specifically, a random forest algorithm model is used to establish the mapping relationship between spectral features and water quality parameters. An attention mechanism is used to optimize the weight allocation of spectral bands to simulate the spatial distribution of water quality parameters. The spectral bands, water quality parameters, and key spectral indices are used as training data. The training data is divided into unsupervised groups using K-means clustering. Finally, the random forest algorithm model outputs eutrophication regions and turbidity regions. Step C4: Parameter correlation verification, specifically, quantifying the consistency between the output of the random forest algorithm model and the K-means clustering results through the error matrix and Kappa coefficients, and verifying the stages of reservoir pollution accumulation by comparing eutrophication areas and difference indices through time series analysis. Step C5: Quantitative assessment of trophic status, specifically, calculating the reservoir trophic status index based on total phosphorus concentration; Step C6: Correct the spectral bands to correct and confirm the indicative ability of spectral features to indicate the pollution status. Specifically, clarify the nutrient pollution level of the reservoir, determine the nutrient status of the reservoir based on the reservoir nutrient status index, and make a quantitative comparison between the nutrient pollution level and spectral reflectance to correct the key spectral band selection. Step C7: Pollution source tracing. Specifically, based on the polygonal vector data of the reservoir, the neighborhood radius and minimum number of points are set. Spatial clustering algorithm is used to perform aggregate analysis on the areas corresponding to each nutrient pollution level to locate pollution hotspots, trace the input path of pollutants, and issue an alarm.

Images