Intelligent analysis system for soil erosion data based on unmanned aerial vehicle multi-spectral monitoring

Abstract

This invention relates to the field of soil and water monitoring technology, specifically to an intelligent analysis system for soil erosion data based on UAV multispectral monitoring. This system aims to solve problems such as inconsistent brightness between multispectral image frames, pixel saturation failure, and the inability to intervene in flight behavior in real time due to fluctuations in ambient light. The technical solution includes a data acquisition module that uses hardware synchronization circuitry to ensure that the upward-facing light sensor and the downward-facing multispectral camera are triggered simultaneously, outputting raw multispectral data frames; an analysis and processing module that uses gradient energy comparison to divide the data into a steady-state queue and an abnormal queue, and interpolates and repairs saturated pixels in the abnormal queue along the texture direction; and an early warning output module that stitches the repaired frames with the steady-state frames, calculates the vegetation index to identify soil erosion areas, and converts the area coordinates into flight control commands for the UAV. This invention can eliminate light interference, repair failed pixels, achieve a closed loop of perception and flight control, and improve the accuracy and real-time performance of soil erosion monitoring.

Description

Technical Field

[0001] This invention relates to the field of soil and water monitoring technology, and more specifically, to an intelligent analysis system for soil erosion data based on multispectral monitoring by unmanned aerial vehicles (UAVs). Background Technology

[0002] Soil and water loss monitoring is a fundamental aspect of soil and water conservation work. Traditional monitoring methods mainly rely on manual ground surveys or satellite remote sensing image interpretation. Although manual surveys have high accuracy, their coverage is limited and the cycle is long, making it difficult to meet the needs of high-frequency monitoring in large areas. Although satellite remote sensing images have a wide coverage, there is an inherent contradiction between spatial resolution and revisit cycle, and they are severely affected by cloud cover and weather, making it impossible to obtain real-time data at key rainfall nodes.

[0003] In recent years, using drones equipped with multispectral cameras for low-altitude remote sensing has become a compromise solution. For example, by acquiring surface images using drones equipped with multispectral cameras and segmenting and extracting bare soil areas using the normalized vegetation index threshold, this method can obtain good results when the lighting conditions are stable. However, in actual operations, due to changes in drone flight attitude and fluctuations in ambient light intensity caused by cloud drift, there are obvious brightness inconsistencies between different frames of images, which causes the vegetation index calculation results to deviate and ultimately affects the accurate identification of bare soil areas.

[0004] Another approach attempts to reduce ambient light interference by simultaneously installing an uplink light sensor and a downlink camera on the drone through light intensity compensation. However, its compensation strategy simply divides the pixel value of the downlink camera by the light intensity value read by the uplink light sensor at the same time. This approach generates a large number of pixels with abnormal reflectivity when the camera's photosensitive element experiences local saturation or nonlinear response. If these abnormal pixels are directly input into subsequent vegetation index calculations without differentiation, false erosion patches will be generated, leading to false alarms.

[0005] Furthermore, existing drone monitoring systems typically only output image analysis results with coordinate annotations after discovering suspected soil erosion areas. They do not form a closed-loop linkage with the drone flight control system, making it impossible to physically intervene in the flight trajectory based on the analysis results to conduct close-range detailed investigations or targeted data re-collection. This results in an increase rather than a decrease in the workload of subsequent manual interpretation.

[0006] In view of this, an intelligent analysis system for soil erosion data based on UAV multispectral monitoring is proposed. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent analysis system for soil erosion data based on UAV multispectral monitoring, in order to solve the technical problems of inconsistent brightness between multispectral image frames, pixel saturation failure, and inability to intervene in flight behavior in real time due to fluctuations in ambient light.

[0008] To address the aforementioned technical problems, this invention provides an intelligent analysis system for soil erosion data based on UAV multispectral monitoring, comprising:

[0009] It includes a data acquisition module, an analysis and processing module, and an early warning output module, which are connected sequentially via an inter-module communication bus, wherein:

[0010] The data acquisition module includes a hardware synchronization circuit and a reflectivity conversion calculator. The hardware synchronization circuit is connected to the pulse signal pin of the UAV flight control system and is used to trigger the airborne upward-facing light sensor and the downward-facing multispectral camera to perform exposure acquisition at the same time when a level transition signal is received. The reflectivity conversion calculator is used to divide the image pixel value acquired by the downward-facing multispectral camera by the light intensity value acquired by the upward-facing light sensor and output the original multispectral data frame that eliminates ambient light interference.

[0011] The analysis and processing module includes a brightness fluctuation calculation unit, a data routing unit, and a blind spot repair unit. The brightness fluctuation calculation unit calculates the brightness difference between two consecutive frames of the original multispectral data frame, pixel by pixel. The data routing unit compares this brightness difference with a preset uniformity threshold, storing pixel regions with brightness differences less than the threshold as steady-state multispectral data frames in a steady-state data queue, and storing pixel regions with brightness differences greater than or equal to the threshold in an abnormal data queue. The blind spot repair unit locates pixels in the abnormal data queue that reach the camera's photosensitive extreme value as failure blind spots, calculates the texture extension line direction of normal pixels surrounding the failure blind spot, extracts the reflectance of normal pixels along this line direction, performs distance-weighted summation, fills the failure blind spot with the summation result, and outputs the repaired multispectral data frame.

[0012] The early warning output module includes a data standardization unit, an erosion risk assessment unit, and a flight control command encapsulator. The data standardization unit stitches together steady-state multispectral data frames from the steady-state data queue with the repaired multispectral data frames, calculates the current average brightness and contrast of the stitched image, and performs a linear scaling operation on the image based on locally stored baseline average and baseline contrast: subtracting the current average brightness, dividing by the current contrast, multiplying by the baseline contrast, and adding the baseline average. The erosion risk assessment unit performs classification processing on the linearly scaled image and outputs analysis results including the coordinates of the soil erosion area. The flight control command encapsulator converts the area coordinates into a control data packet containing waypoint deflection parameters and sends physical trajectory intervention actions to the UAV flight control system via an external communication interface.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0014] 1. In this intelligent analysis system for soil erosion data based on UAV multispectral monitoring, the hardware synchronization circuit eliminates the lighting and image pairing error caused by inconsistent triggering times; the brightness fluctuation calculation and data diversion mechanism isolates pixel areas affected by sudden changes in lighting to prevent abnormal data from contaminating the steady-state calculation process; the blind spot repair unit uses texture direction interpolation to fill saturated pixels and restore the lost true surface reflectivity information; the early warning output module converts the analysis results into flight control commands in real time, enabling the UAV to automatically adjust its trajectory based on on-site data to carry out supplementary data collection or close-range observation, forming a closed-loop system of perception, analysis, decision-making, and action.

[0015] 2. In this intelligent analysis system for soil erosion data based on UAV multispectral monitoring, the brightness fluctuation calculation unit adopts a fluctuation measurement method based on gradient energy comparison. This method is not sensitive to image translation and can accurately reflect the changes in texture structure caused by changes in illumination, thus avoiding false alarms caused by UAV translation.

[0016] 3. In this intelligent analysis system for soil erosion data based on UAV multispectral monitoring, blind spot repair adopts projection weighted interpolation along the texture extension direction. Compared with simple neighborhood mean filling, it can better maintain the continuity of the edge of the ground feature, especially for the boundary restoration of linear bare soil patches.

[0017] 4. In this intelligent analysis system for soil erosion data based on UAV multispectral monitoring, the baseline mean and baseline contrast in the data standardization unit are derived from the soil erosion training set, ensuring that the normalized baseline is consistent for data under different lighting conditions and different flights before vegetation index calculation, thus improving the stability of cross-task analysis.

[0018] 5. In this intelligent analysis system for soil erosion data based on UAV multispectral monitoring, the physical anti-counterfeiting device detects mirror reflection by contrast ratio, preventing erroneous information from entering the analysis link at the data source. At the same time, it actively adjusts the gimbal angle to obtain valid data, enhancing the system's robustness in complex surface environments. Attached Figure Description

[0019] Figure 1 This is an overall system block diagram of the present invention.

[0020] Figure 2 This is a system block diagram of the data acquisition module of the present invention.

[0021] Figure 3 This is a system block diagram of the analysis and processing module of the present invention.

[0022] Figure 4 This is a system block diagram of the early warning output module of the present invention. Detailed Implementation

[0023] The technical solutions of 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.

[0024] Example

[0025] Please see Figures 1 to 4 As shown, the purpose of this invention is to provide an intelligent analysis system for soil erosion data based on UAV multispectral monitoring. The system includes a data acquisition module, an analysis and processing module, and an early warning output module connected sequentially through an inter-module communication bus. The inter-module communication bus can be an SPI bus, an I2C bus, or a parallel data bus to achieve high-speed data exchange between modules, ensuring that the transmission delay of multispectral image frames in the acquisition, processing, and analysis stages is less than one frame period.

[0026] Furthermore, all modules of the system will be made public.

[0027] The data acquisition module includes a hardware synchronization circuit and a reflectivity conversion calculator. The hardware synchronization circuit is connected to the pulse signal pin of the UAV flight control system, used to trigger the onboard upward-facing light sensor and the downward-facing multispectral camera to perform exposure acquisition simultaneously upon receiving a level transition signal. The reflectivity conversion calculator divides the image pixel values ​​acquired by the downward-facing multispectral camera by the illumination intensity values ​​acquired by the upward-facing light sensor, outputting a raw multispectral data frame free from ambient light interference. The specific implementation is as follows:

[0028] Considering that the ambient light intensity may fluctuate instantaneously during drone flight due to cloud cover, changes in solar altitude angle, or adjustments in drone attitude, if there is any time difference between the exposure times of the upward-facing light sensor and the downward-facing multispectral camera, the collected light intensity value and the pixel value of the ground image will correspond to different lighting conditions at different times. When performing reflectivity conversion later, this time difference will directly translate into reflectivity calculation error. Traditional solutions use software commands to trigger the two sensors sequentially. Due to the jitter of the operating system's task scheduling, there will be a time difference of several milliseconds to tens of milliseconds. For a drone flying at a speed of 10 meters per second, this time difference is enough to cause a ground sampling position shift of several centimeters to tens of centimeters. Coupled with the coupling of light changes, the error cannot be ignored.

[0029] To address the aforementioned issues, the specific wiring structure of the hardware synchronization circuit in this embodiment is as follows: the level capture pin of the upward-facing light sensor and the shutter trigger pin of the downward-facing multispectral camera are connected in parallel to the same general-purpose input / output pulse pin of the UAV flight control system in the hardware circuit. The acquisition actions of the two sensors are synchronously triggered by the same voltage transition edge on the same wire. Specifically:

[0030] The level capture pin of the overhead light sensor and the shutter trigger pin of the over-ground multispectral camera are connected in parallel on the printed circuit board to the same general-purpose input / output pulse pin of the UAV flight control system. When the flight control system needs to acquire a frame of data, its program instruction pulls the GPIO pin level from low to high or from high to low, generating a voltage transition edge. Since the trigger terminals of the two sensors are physically shorted to the same wire, this transition edge will arrive at the trigger detection circuits of the two sensors simultaneously within the electrical signal propagation delay (nanosecond level). The exposure start action of the two sensors is synchronously triggered by the same voltage transition edge on the same wire, achieving hardware synchronization accuracy on the order of nanoseconds.

[0031] Through the above technical solution, the incident light irradiance value measured by the Chaotian light sensor and the ground object reflected radiant flux value received by the same pixel of the Chaodi multispectral camera strictly correspond to the solar illumination conditions at the same time. This eliminates the registration error caused by ambient light interference due to inconsistent triggering times. When the reflectivity conversion calculator performs division operation, the illumination conditions of the numerator and denominator are time-consistent, thus obtaining an accurate reflectivity value.

[0032] The reflectivity conversion operator can be implemented as an embedded digital signal processor or a hardware divider within an FPGA. Its operation is as follows: for each band of image pixel value output by the North Korean multispectral camera, divide by the quantized value of illumination intensity output by the North Korean light sensor in the same band channel; the pixel value of the North Korean camera is denoted as... The value of the upward-facing light sensor Then the reflectivity factor of that pixel is output. If the sensor has been radiometrically calibrated, then The value is proportional to the radiance, and the quotient is approximately the surface reflectance factor; this operation is performed in parallel pixel by pixel, and the raw multispectral data frame is output in units of frames.

[0033] The output of the data acquisition module is also equipped with a data traceability security node. This node has a built-in direct memory access (DMA) controller independent of the central processing unit. Within a single clock cycle after the original multispectral data frame is generated, it directly transfers the underlying binary code of the data frame to the onboard solid-state drive for storage, generating an untampered, tamper-proof copy of the original evidence. Specifically:

[0034] This node has a built-in direct memory access controller independent of the main processor. After the raw multispectral data frame is output by the reflectance conversion unit, the DMA controller directly moves the underlying binary code of the data frame to a preset sector of the onboard solid-state drive for storage within a single clock cycle, without any software processing. The reason for this setting is that in subsequent analysis and processing, the raw data may be modified by blind spot repair, standardization, and other processes. If there is a dispute over the authenticity of the data or if it is necessary to trace the original observation state, there must be a tamper-proof copy of the original evidence that has not been repaired or cropped. Using hardware DMA to move data does not occupy the main processor's computing resources and completes storage in a very short time after the data is generated, minimizing the risk of data being tampered with or lost in memory.

[0035] The analysis and processing module includes a brightness fluctuation calculation unit, a data routing unit, and a blind spot repair unit. The brightness fluctuation calculation unit calculates the brightness difference between two consecutive frames of the original multispectral data frame, pixel by pixel. The data routing unit compares this brightness difference with a preset uniformity threshold, storing pixel regions with brightness differences less than the threshold as steady-state multispectral data frames in a steady-state data queue, and storing pixel regions with brightness differences greater than or equal to the threshold in an abnormal data queue. The blind spot repair unit locates pixels in the abnormal data queue that reach the camera's photosensitive extreme value as failure blind spots, calculates the texture extension line direction of normal pixels surrounding the failure blind spot, extracts the reflectance of normal pixels along this line direction, performs distance-weighted summation, fills the failure blind spot with the summation result, and outputs the repaired multispectral data frame. The specific implementation method is as follows:

[0036] The brightness fluctuation calculation unit is used to calculate the brightness difference between two consecutive frames of the original multispectral data frame pixel by pixel. Its purpose is to identify pixel areas where the reflectance calculation value deviates abnormally from the steady state due to sudden changes in illumination. Sudden changes in illumination are manifested in the image as drastic changes in brightness in local areas. If these abnormal pixels directly participate in subsequent stitching and vegetation index calculation, false erosion patches will be introduced.

[0037] Because simply comparing the difference in pixel brightness values ​​is too sensitive to scene changes caused by drone translation—for example, when the drone flies forward, the pixels at the edge of a tree may correspond to different ground features in different frames, resulting in a large difference in brightness—this is a normal scene change and should not be classified as abnormal flow. Therefore, the brightness fluctuation calculation unit introduces gradient energy values ​​as a comparison criterion. The specific implementation method is as follows:

[0038] The brightness fluctuation calculation unit includes a spatial difference operator, a delay buffer, and a subtraction operator.

[0039] The spatial interpolation operator hardware circuit implements the Sobel or Prewitt operator convolution kernel. For the target pixel position in the current data frame, it reads the values ​​of its eight neighboring pixels in each spectral band and calculates the brightness difference in the horizontal and vertical directions respectively. For example, for the red band, the horizontal gradient... Vertical gradient The gradient energy value of the pixel in that band is obtained by squaring Gx and Gy and then adding them together. The gradient energy values ​​of all bands are then summed to generate the current gradient energy value of the target pixel. .

[0040] The delay buffer is a one-frame memory that reads the historical gradient energy values ​​of pixels at the same geographic coordinates in the previous captured frame. Since the drone is in flight, the geographic correspondence of pixels in consecutive frames can be geometrically corrected and registered using the positioning and orientation data and digital elevation model transmitted by the flight control system. Only after registration can pixels of the same geographic point be comparable in consecutive frames. This registration operation can be completed by an independent geometric correction unit before brightness fluctuation calculation.

[0041] The subtraction operator calculates the absolute value of the difference between the current gradient energy value and the historical gradient energy values. The absolute value of this difference is the basis for performing the split comparison. The principle is that if the ambient lighting is stable, the calculated reflectance value of the same feature point at different times should be basically consistent, its image texture structure remains unchanged, and the gradient energy value fluctuates very little. However, if cloud shadows suddenly cover or leave the area, the light intensity changes drastically, but the feature texture itself remains unchanged. The reflectance conversion operation will reflect this change in light intensity in the pixel values, causing a significant jump in the gradient energy value. Therefore… It can effectively reflect the impact of fluctuations in lighting conditions on the calculated reflectance value, while filtering out false alarms caused by normal changes in scene content.

[0042] The data routing unit is used for comparison. The preset uniformity threshold is determined as follows: Before system deployment, multiple consecutive frames of static scene multispectral data are collected under standard stable lighting conditions, and the uniformity threshold of all pixels is calculated. The statistical distribution is used, and the 95th percentile of the cumulative probability distribution is taken as the initial threshold value. This initial threshold value is selected based on the background noise variance σ of the sensor under laboratory calibration. Specifically, 3σ is set as the lower limit of the uniformity threshold to exclude pure hardware noise interference. Typical values ​​are between 0.015 and 0.030, which aims to ensure that data diversion can be triggered in a timely manner when reflectivity fluctuations exceed 5% due to sudden changes in illumination, thereby ensuring the reflectivity accuracy of steady-state data. If the value is less than this threshold, it indicates that the data in this pixel region has not been affected by sudden changes in illumination and belongs to steady-state data. Therefore, this pixel region data is stored as a steady-state multispectral data frame in the steady-state data queue. If the value is greater than or equal to the threshold, it indicates that the pixel area is severely affected by sudden changes in illumination, and it is stored in the abnormal data queue to await repair.

[0043] Through the above-described diversion process, the steady-state data queue contains high-quality, reliable reflectance data that can be directly used for subsequent stitching and analysis; the abnormal data queue contains problem areas that need to be repaired. This diversion mechanism ensures that bad data caused by sudden changes in illumination will not contaminate the overall analysis results, while the isolated abnormal areas still have the opportunity to be repaired, rather than being simply discarded.

[0044] The blind spot repair unit is used to handle failed blind spots in the abnormal data queue. Failed blind spots refer to pixels that have reached the camera's light sensitivity limit. The main reason is that when the clouds dissipate rapidly and the direct sunlight suddenly intensifies, the reflected radiation flux of some high reflectivity features, such as bare soil, sand, or the specular reflection area of ​​water bodies, may exceed the full-capacity of the ground-facing camera. The pixel value saturates and stops at the maximum value. At this time, the reflectivity value obtained by division is a false value, and the pixel becomes a failed blind spot. If it is not repaired, these saturated pixels will show abnormally low vegetation index values ​​in the subsequent vegetation index calculation and be misjudged as bare soil.

[0045] The specific logic of the blind spot repair unit performing distance-weighted summation is as follows: A search window is defined centered on the failed blind spot; the grayscale distribution gradient of normal pixels within the window is extracted; the direction with the smoothest grayscale change is locked as the texture extension line direction; the spatial straight-line distance from each normal pixel within the search window to the failed blind spot is calculated; the projection length of each normal pixel in the texture extension line direction is divided by the spatial straight-line distance, and the quotient is used as the interpolation weight for each normal pixel; the reflectivity of each normal pixel is multiplied by its corresponding interpolation weight using a multiply-accumulate operator, and the sum is accumulated; the accumulated result is divided by the sum of all interpolation weights within the search window; the final calculated value is written to the storage address of the failed blind spot. The specific implementation method is as follows:

[0046] First, a rectangular search window is defined centered on the failure blind spot, with a window size of 7×7 or 9×9 pixels. Multispectral reflectance values ​​of all unsaturated normal pixels within the window are extracted to construct a grayscale distribution matrix. The direction and magnitude of the grayscale gradient are calculated, specifically using the structure tensor method: the horizontal gradient Ix and vertical gradient Iy of each normal pixel within the window are calculated, a structure tensor matrix is ​​constructed, and its eigenvalues ​​are obtained. The direction indicated by the eigenvector with the smaller eigenvalue is the direction of the most gradual grayscale change, and this direction is locked as the direction of the texture extension line. This direction represents the natural orientation of the ground texture, such as the boundary line of bare soil patches or the direction of vegetation strips.

[0047] Then, for each normal pixel in the search window, calculate the Euclidean distance d from it to the center of the failure blind spot; at the same time, calculate the projection length p of the vector connecting the normal pixel and the failure blind spot in the direction of the texture extension line; divide the projection length p by the spatial line distance d, and the quotient w=p / d is used as the interpolation weight of the normal pixel.

[0048] The reason for using this weighting structure is that if a normal pixel is located exactly on the texture extension line, its projected length is equal to the spatial distance, and its weight is 1; if it deviates from the direction of the line, the weight decreases with the cosine of the deviation angle. In this way, the weight takes into account both the distance of the spatial distance and the consistency of the direction, so that the interpolation result strictly follows the linear extension law of the texture and avoids mixing irrelevant pixel information in the vertical texture direction into the interpolation.

[0049] The reflectance values ​​of each normal pixel are multiplied and added together using a multiply-accumulate operator. Its corresponding weight Multiply and then sum to get a weighted sum. Simultaneously, summing all weights yields... Ultimately Divide by The calculated value is written to the storage address of the failure blind spot, completing the repair.

[0050] The repaired multispectral data frame is sent to the early warning output module and spliced ​​with the frames in the steady-state data queue.

[0051] The early warning output module includes a data standardization unit, an erosion risk assessment unit, and a flight control command encapsulator. The data standardization unit stitches together steady-state multispectral data frames from the steady-state data queue with repaired multispectral data frames, calculates the current average brightness and contrast of the stitched image, and performs a linear scaling operation on the image based on locally stored baseline average and baseline contrast: subtracting the current average brightness, dividing by the current contrast, multiplying by the baseline contrast, and adding the baseline average. The erosion risk assessment unit performs classification processing on the linearly scaled image and outputs analysis results including the coordinates of the soil erosion area. The flight control command encapsulator converts the area coordinates into a control data packet containing waypoint deflection parameters and sends physical trajectory intervention actions to the UAV flight control system via an external communication interface. The specific implementation is as follows:

[0052] The data standardization unit is used to stitch together the steady-state multispectral data frames in the steady-state data queue with the repaired multispectral data frames. Before performing spatial stitching, the data standardization unit uses the system timestamp attached to the original multispectral data frames when they were generated to time-align the data frames that have experienced computational delays due to blind spot repair with the data frames in the same time period in the steady-state data queue, based on the system timestamp attached to them, to eliminate the phase difference caused by dual-channel processing. Subsequently, the stitching operation aligns the pixels of the same geographic area to a unified coordinate grid based on the positioning and orientation data of each frame. However, due to the slow changes in illumination conditions over time and the differences in atmospheric conditions between different flights, the overall average brightness and contrast of the stitched image may still be inconsistent. This inconsistency will directly affect the calculation of spectral derived indicators such as vegetation index, resulting in non-universal thresholds across tasks.

[0053] To address this issue, the data normalization unit performs a linear scaling operation on the stitched image, and records the reflectance value of a pixel in the stitched image as... The current average brightness of the image is The current contrast is The contrast ratio is taken as the arithmetic square root of the luminance variance. The calculation formula is: ,in, The baseline mean Using the baseline contrast, this operation maps image data from different lighting conditions to a unified brightness and contrast space, making subsequent vegetation index calculations comparable across different frames and time periods.

[0054] The baseline mean and baseline contrast stored locally in the data standardization unit are configured with the following parameter sources: Before configuring parameters in the erosion risk assessment unit, a standard multispectral image training set containing bare land affected by soil erosion and normal vegetation is extracted. The global brightness average of the training set images is calculated as the baseline mean, and the arithmetic square root of the global brightness variance is calculated as the baseline contrast. The specific implementation method is as follows:

[0055] Before formal parameter configuration of the erosion risk assessment unit, a standard multispectral image training set containing both bare land due to soil erosion and normal vegetation was extracted from historical UAV flight missions. This training set covered typical surface conditions under different seasons and solar altitude angles. The global average brightness was calculated for all pixels in the training set, which is the... ; Calculate the arithmetic square root of the global brightness variance, which is... These two parameters are stored in the system's local memory as a standardized benchmark.

[0056] The standardized image output by the data standardization unit is sent to the erosion risk assessment unit. The specific data operation logic for classification processing in this unit is as follows: The visible light red band matrix and the near-infrared band matrix are separated from the standardized image after linear scaling; the difference matrix between the two is divided by the sum matrix to generate a normalized vegetation index feature map; areas in the normalized vegetation index feature map with pixel values ​​lower than the set vegetation degradation standard are designated as bare soil patches with soil erosion risk. The specific implementation method is as follows:

[0057] First, the visible red band matrix (Red) and near-infrared band matrix (NIR) are separated from the normalized image. These two bands are the basis for calculating the normalized vegetation index.

[0058] Secondly, the quotient of the difference matrix and the sum matrix is ​​calculated to generate a normalized vegetation index feature map: The NDIV value range is [-1, 1]. Healthy vegetation has an NDIV value close to 1 due to its high near-infrared reflectivity and strong red light absorption. Bare soil and water bodies have lower NDIV values, close to 0 or negative.

[0059] Finally, areas with pixel values ​​in the NDIV feature map lower than the set vegetation degradation standard are delineated as bare soil patches with the risk of soil erosion. The vegetation degradation standard can be set according to the local ecosystem type. For example, for the Loess Plateau region, the threshold can be set to 0.2. Areas with a value lower than 0.2 are considered to have extremely low vegetation coverage or bare ground. The analysis results output by the erosion risk assessment unit are geographic information files containing the coordinate point set of the polygonal outline of the bare soil patches.

[0060] In the early warning output module, a physical blocking anti-counterfeiting device is also connected in parallel. This physical blocking anti-counterfeiting device is located on the data transmission path between the data standardization unit and the erosion risk assessment unit. Its hardware includes a divider and a digital comparator. The divider is used to calculate the ratio of the current contrast to the reference contrast. When the ratio is greater than the safety upper limit parameter set by the digital comparator, the physical blocking anti-counterfeiting device determines that there is a large area of ​​water surface mirror reflection in the current image, cuts off the data transmission to the erosion risk assessment unit, and directly generates a pulse width modulation electrical signal containing the gimbal yaw angle adjustment increment, which is output to the UAV gimbal controller to drive the gimbal motor to forcibly change the shooting angle. The specific implementation method is as follows:

[0061] The divider calculates the current contrast. Contrast with reference ratio The digital comparator has a preset safety upper limit parameter, for example, the upper limit is set to 2.5. This upper limit is determined based on the following reasons: large-area water surface mirror reflection will produce extremely high brightness saturation areas in the image, causing the local contrast to increase abnormally, resulting in r exceeding the normal range of surface variation; when r is greater than the safety upper limit parameter, the physical blocking anti-fake device determines that there is large-area water surface mirror reflection in the current image, and immediately cuts off the transmission of data to the erosion risk assessment unit to prevent false high-contrast images caused by mirror reflection from entering the vegetation index calculation, and avoid misjudging water bodies as bare soil.

[0062] Meanwhile, the physical blocking anti-counterfeiting device directly generates a pulse width modulation (PWM) electrical signal containing the gimbal yaw angle adjustment increment. For example, it outputs a PWM waveform with a period of 20ms and a high-level pulse width varying between 1.0ms and 2.0ms, corresponding to the target angle increment of the gimbal yaw angle servo motor. This electrical signal is output to the UAV gimbal controller to drive the gimbal motor to forcibly change the shooting angle and avoid the specular reflection angle, thus forming an active defense mechanism, rather than relying solely on software-level anomaly removal.

[0063] The flight control command encapsulator converts the coordinates of bare soil patches output by the erosion risk assessment unit into control data packets, and sends physical trajectory intervention actions to the UAV flight control system through an external communication interface. The specific execution rules are as follows: extract the polygonal edge contour coordinate point set of the bare soil patches from the analysis results; retrieve the digital elevation terrain model stored internally by the system to calculate the terrain orientation of the edge contour coordinate point set; generate a sequence of ground-hugging flight waypoints extending along the terrain orientation and with a relative ground altitude lower than the current cruising altitude; and encapsulate this sequence into a data packet supported by the flight control protocol and send it out. The specific implementation method is as follows:

[0064] The system extracts the polygonal edge contour coordinates of bare soil patches from the analysis results, retrieves the digital elevation model stored in the system, and calculates the terrain orientation of the edge contour coordinates. That is, it determines the extension direction of erosion gullies or exposed slopes through terrain slope and aspect analysis, and generates a sequence of ground-hugging flight waypoints extending along the terrain orientation. The waypoints are positioned at a height lower than the current cruising altitude. For example, when the cruising altitude is 80 meters, the ground-hugging waypoints can be set to 30 meters. The waypoint sequence is encapsulated into data packets according to the MAVLink protocol or the drone manufacturer's custom protocol and sent to the flight control system via serial port or wireless link. The drone adjusts its flight trajectory accordingly to conduct low-altitude, high-resolution detailed surveys of suspected soil erosion areas, obtaining more detailed image data to support further manual or automatic verification.

[0065] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent analysis system for soil erosion data based on unmanned aerial vehicle multispectral monitoring, characterized in that, It includes a data acquisition module, an analysis and processing module, and an early warning output module, which are connected sequentially via an inter-module communication bus, wherein: The data acquisition module includes a hardware synchronization circuit and a reflectivity conversion calculator. The hardware synchronization circuit is connected to the pulse signal pin of the UAV flight control system and is used to trigger the airborne upward-facing light sensor and the downward-facing multispectral camera to perform exposure acquisition at the same time when a level transition signal is received. The reflectivity conversion calculator is used to divide the image pixel value acquired by the downward-facing multispectral camera by the light intensity value acquired by the upward-facing light sensor and output the original multispectral data frame that eliminates ambient light interference. The analysis and processing module includes a brightness fluctuation calculation unit, a data routing unit, and a blind spot repair unit. The brightness fluctuation calculation unit calculates the brightness difference between two consecutive frames of the original multispectral data frame, pixel by pixel. The data routing unit compares this brightness difference with a preset uniformity threshold, storing pixel regions with brightness differences less than the threshold as steady-state multispectral data frames in a steady-state data queue, and storing pixel regions with brightness differences greater than or equal to the threshold in an abnormal data queue. The blind spot repair unit locates pixels in the abnormal data queue that reach the camera's photosensitive extreme value as failure blind spots, calculates the texture extension line direction of normal pixels surrounding the failure blind spot, extracts the reflectance of normal pixels along this line direction, performs distance-weighted summation, fills the failure blind spot with the summation result, and outputs the repaired multispectral data frame. The early warning output module includes a data standardization unit, an erosion risk assessment unit, and a flight control command encapsulator. The data standardization unit stitches together steady-state multispectral data frames from the steady-state data queue with the repaired multispectral data frames, calculates the current average brightness and contrast of the stitched image, and performs a linear scaling operation on the image based on locally stored baseline average and baseline contrast: subtracting the current average brightness, dividing by the current contrast, multiplying by the baseline contrast, and adding the baseline average. The erosion risk assessment unit performs classification processing on the linearly scaled image and outputs analysis results including the coordinates of the soil erosion area. The flight control command encapsulator converts the area coordinates into a control data packet containing waypoint deflection parameters and sends physical trajectory intervention actions to the UAV flight control system via an external communication interface. 2.The water and soil erosion data intelligent analysis system based on unmanned aerial vehicle multi-spectral monitoring of claim 1, wherein, The specific hardware and computing architecture for the brightness fluctuation calculation unit in the analysis and processing module is as follows: It includes a spatial difference operator, which is used to extract the brightness difference between the target pixel and its eight neighboring pixels in each spectral band in the current data frame, and then sum the squares of each brightness difference to generate the current gradient energy value of the target pixel. Includes a delay buffer for reading the historical gradient energy value of pixels at the same location in the previous captured frame; It includes a subtraction operator, which calculates the absolute value of the difference between the current gradient energy value and the historical gradient energy value, and uses this absolute value as the basis for performing the split comparison.

3. The intelligent analysis system for soil erosion data based on UAV multispectral monitoring according to claim 1, characterized in that, The specific logic for the distance-weighted summation performed by the blind spot repair unit in the analysis and processing module is as follows: A search window is defined centered on the failure blind spot. The gray-level distribution gradient of normal pixels within the window is extracted, and the direction of the most gradual gray-level change is locked as the direction of the texture extension line. Calculate the spatial straight-line distance from each normal pixel in the search window to the failure blind spot, divide the projection length of each normal pixel in the direction of the texture extension line by the spatial straight-line distance, and use the obtained quotient as the interpolation weight of each normal pixel. The reflectivity of each normal pixel is multiplied by its corresponding interpolation weight using a multiply-accumulate operator, and the sum is accumulated. The accumulated result is then divided by the sum of all interpolation weights within the search window, and the final calculated value is written to the storage address of the failure blind spot.

4. The intelligent analysis system for soil erosion data based on UAV multispectral monitoring according to claim 1, characterized in that, The reference mean and reference contrast stored locally in the data standardization unit are configured with the following parameter sources: Before configuring parameters in the erosion risk assessment unit, a standard multispectral image training set containing bare land and normal vegetation is extracted. The global brightness average value of the training set images is calculated as the baseline mean, and the arithmetic square root of the global brightness variance is calculated as the baseline contrast.

5. The intelligent analysis system for soil erosion data based on UAV multispectral monitoring according to claim 1, characterized in that, The early warning output module also includes a physical blocking anti-counterfeiting device connected in parallel to the data transmission path. The physical blocking anti-counterfeiting device includes a divider and a digital comparator. The divider is used to calculate the ratio of the current contrast to the reference contrast. When the ratio is greater than the safety upper limit parameter set by the digital comparator, the physical blocking anti-counterfeiting device determines that there is a large area of ​​water surface mirror reflection in the current image, cuts off the transmission of data to the erosion risk assessment unit, and directly generates a pulse width modulation electrical signal containing the gimbal yaw angle adjustment increment, which is output to the UAV gimbal controller to drive the gimbal motor to forcibly change the shooting angle.

6. The intelligent analysis system for soil erosion data based on UAV multispectral monitoring according to claim 1, characterized in that, The specific wiring structure of the hardware synchronization circuit in the data acquisition module is as follows: The level capture pin of the upward-facing light sensor and the shutter trigger pin of the downward-facing multispectral camera are connected in parallel to the same general-purpose input / output pulse pin of the UAV flight control system in the hardware circuit. The acquisition actions of the two sensors are synchronously triggered by the same voltage transition edge on the same wire.

7. The intelligent analysis system for soil erosion data based on UAV multispectral monitoring according to claim 1, characterized in that, The specific data processing logic for the classification process performed by the erosion risk assessment unit is as follows: Separate the visible light red band matrix and the near-infrared band matrix from the image after linear scaling; Calculate the quotient of the difference matrix and the sum matrix to generate a normalized vegetation index feature map. In the Normalized Difference Vegetation Index (NDVI) feature map, areas with pixel values ​​below the set vegetation degradation standard are designated as bare soil patches at risk of soil erosion.

8. The intelligent analysis system for soil erosion data based on UAV multispectral monitoring according to claim 1, characterized in that, The specific execution rules for the flight control command encapsulator to generate control data packets are as follows: Extract the set of polygonal edge contour coordinates of bare soil patches from the analysis results; Retrieve the digital elevation terrain model stored internally in the system and calculate the terrain orientation of the edge contour coordinate point set; Generate a sequence of waypoints that extend along the terrain and are lower than the current cruising altitude relative to the ground, and encapsulate this sequence into a data packet supported by the flight control protocol and send it out.

9. The intelligent analysis system for soil erosion data based on UAV multispectral monitoring according to claim 1, characterized in that, The system also includes a data traceability security node mounted on the output end of the data acquisition module; The data traceability security node has a built-in direct memory access controller independent of the central processing unit. It is used to directly transfer the underlying binary code of the original multispectral data frame to the onboard solid-state drive for storage within a single clock cycle after the original multispectral data frame is generated, generating an untampered original evidence copy that has not undergone any repair or trimming.

Images