A photovoltaic power station unmanned aerial vehicle dual-camera intelligent cruise detection device

Abstract

The present application relates to the technical field of unmanned aerial vehicle, especially to a photovoltaic power station unmanned aerial vehicle dual-camera intelligent cruise detection device, comprising a multi-source data acquisition module, which synchronously acquires photovoltaic panel images through an infrared thermal imaging camera and a visible light camera, and uses a laser range finder to acquire the vertical distance from the photovoltaic panel, wherein the visible light camera is equipped with a stepping motor and a rotatable circular polarizer; a dynamic parallax compensation module, which calculates a dynamic parallax compensation angle according to the vertical distance and the camera installation interval, drives the stepping motor to adjust the angle, and makes the optical centers of the two cameras intersect; and a spatial pose analysis module, which extracts the current unmanned aerial vehicle attitude data, position data, time data and the dynamic parallax compensation angle when detecting a suspected hot spot. In the present application, the vertical distance is acquired in real time by using a laser range finder, and the dynamic parallax compensation angle is calculated in combination with the camera installation interval, thereby effectively eliminating the dynamic parallax caused by the physical baseline.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) technology, and in particular to a dual-camera intelligent cruise detection device for photovoltaic power station UAVs. Background Technology

[0002] With the rapid development of the photovoltaic industry, it has become commonplace to use drones equipped with both infrared and visible light cameras for photovoltaic power plant inspections. During routine patrols, the system typically first uses an infrared thermal imaging camera to perform a global temperature scan to detect suspected abnormal hot spots. Then, it links with a visible light camera to capture close-up images of the target area to confirm whether there are any physical damages such as hidden cracks, breaks, or shading on the surface of the photovoltaic panel, thereby achieving rapid and automated inspection of power plant equipment.

[0003] Existing technologies employ open-loop independent dual-light linkage and blind scanning extinction based on image overexposure feedback. This causes the visible light camera to be susceptible to glare from the photovoltaic glass mirror when the drone is flying at high speed, making it unable to instantly acquire a clear underlying image. Furthermore, the system is prone to misjudging artifacts formed by the reflection of environmental heat sources as real hot spot faults in the components, thereby increasing the false alarm rate and ineffective operation costs of the inspection. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a dual-camera intelligent cruise detection device for photovoltaic power plants, aiming to improve the problem of high false alarm rate during inspections.

[0005] In a first aspect, the present invention provides the following technical solution: a dual-camera intelligent cruise detection device for photovoltaic power plants using unmanned aerial vehicles (UAVs), comprising: The multi-source data acquisition module simultaneously acquires images of the photovoltaic panel through an infrared thermal imaging camera and a visible light camera, and uses a laser rangefinder to obtain the vertical distance to the photovoltaic panel. The visible light camera is equipped with a stepper motor and a rotatable circular polarizing filter. The dynamic parallax compensation module calculates the dynamic parallax compensation angle based on the vertical distance and camera mounting distance, and drives the stepper motor to adjust the angle so that the optical centers of the two cameras intersect. The spatial pose analysis module extracts the current UAV attitude data, position data, time data, and the dynamic parallax compensation angle when a suspected hot spot is detected. The polarization extinction feedforward control module calculates the solar incident direction based on the position data and time data, and calculates the polarization direction of the reflected light by combining the UAV attitude data, dynamic parallax compensation angle and the vertical direction of the panel to obtain the target extinction angle, and drives the rotatable circular polarizing mirror to rotate to that angle to capture a visible light image. The bidirectional feature verification module identifies physical damage in the visible light image, determines the actual fault by combining the confidence level of suspected hot spots, and eliminates artifact data by combining the reflection geometric overlap.

[0006] Preferably, in the multi-source data acquisition module, the step of using a laser rangefinder to obtain the vertical distance to the photovoltaic panel includes: A laser pulse is emitted using a laser rangefinder installed between the infrared thermal imaging camera and the visible light camera; The system receives the echo signal reflected by the photovoltaic panel and calculates the vertical distance based on the flight time.

[0007] Preferably, in the dynamic parallax compensation module, the step of calculating the dynamic parallax compensation angle includes: Obtain the camera installation distance between the two cameras; The arctangent angle is calculated from the ratio of the camera mounting spacing to the vertical distance, and the arctangent angle is used as the dynamic parallax compensation angle.

[0008] Preferably, in the spatial pose analysis module, the step of extracting the current UAV attitude data, position data, time data, and the dynamic parallax compensation angle includes: When an abnormally high temperature region is detected by scanning an infrared image using a threshold segmentation method, the coordinates of its center pixel are extracted as the suspected hot spot. The instantaneous roll angle, pitch angle, and yaw angle are recorded as the attitude data of the UAV, the latitude and longitude are recorded simultaneously as the position data, and the system absolute time is read as the time data.

[0009] Preferably, in the polarization extinction feedforward control module, the step of calculating the polarization direction of the reflected light includes: The true observation direction is calculated based on the initial observation direction of the visible light camera, the attitude data of the UAV, and the rotation state after performing the dynamic parallax compensation angle. The location data and the time data are input into the ephemeris database to construct the solar incidence direction; The polarization direction of the reflected light is calculated by cross product of the vertical direction of the panel and the direction of solar incidence.

[0010] Preferably, in the polarization extinction feedforward control module, the step of obtaining the target extinction angle includes: The polarization direction of the reflected light is projected onto an imaging plane with the actual observation direction as the normal direction; Obtain the horizontal reference direction of the photosensitive element inside the visible light camera; Calculate the angle between the projection direction and the horizontal reference direction, and add 90 degrees to the angle to obtain the target extinction angle.

[0011] Preferably, in the bidirectional feature verification module, the step of determining the actual fault by combining the confidence level of the suspected hotspot includes: The confidence level of the abnormal high temperature output by the infrared thermal imaging camera is used as the confidence level of the suspected hot spot; The visible light image is analyzed using a target detection algorithm to obtain the confidence level of physical damage; When the confidence level of the suspected hot spot is greater than a preset infrared threshold and the confidence level of the physical damage is greater than a preset visible light threshold, the existence of the actual fault is confirmed.

[0012] Preferably, in the bidirectional feature verification module, the step of combining reflection geometric overlap to remove artifact data includes: The specular reflection direction of sunlight is calculated based on the solar incident direction and the perpendicular direction of the panel; Calculate the degree of collinearity between the observation direction of the visible light camera and the reflection direction of the mirror when the camera captures the image, and use the absolute value of this degree of overlap as the reflection geometric overlap.

[0013] Preferably, in the bidirectional feature verification module, the step of removing artifact data includes: When the confidence level of the suspected hot spot is greater than a preset infrared threshold and the confidence level of the physical damage is less than or equal to a preset visible light threshold, and the reflective geometric overlap is greater than a preset overlap threshold; If the current signal is determined to be a specular heating artifact caused by an ambient heat source, this data is discarded as the artifact data.

[0014] Secondly, the present invention provides the following technical solution: a dual-camera intelligent cruise detection method for photovoltaic power station drones, the method comprising: S1, the infrared thermal imaging camera and visible light camera of the multi-source data acquisition module simultaneously acquire images of the photovoltaic panel, and the vertical distance to the photovoltaic panel is obtained using a laser rangefinder; S2, the dynamic parallax compensation module calculates the dynamic parallax compensation angle based on the vertical distance and camera installation spacing, drives the stepper motor to adjust the angle, so that the optical centers of the two cameras intersect; S3, when a suspected hot spot is detected by the spatial pose analysis module, the current UAV attitude data, position data, time data and the dynamic parallax compensation angle are extracted; S4, the polarization extinction feedforward control module calculates the sun's incident direction based on the position data and time data, and calculates the polarization direction of the reflected light by combining the UAV attitude data, dynamic parallax compensation angle and the vertical direction of the panel to obtain the target extinction angle, and drives the rotatable circular polarizing mirror to rotate to that angle to capture a visible light image; S5, the physical damage of the visible light image is identified by the bidirectional feature verification module, the real fault is determined by the confidence of suspected hot spots, and artifact data is eliminated by the reflection geometric overlap.

[0015] The present invention has the following beneficial effects: 1. In this invention, a laser rangefinder is used to obtain the vertical distance in real time, and the dynamic parallax compensation angle is calculated in combination with the camera installation spacing. This drives a stepper motor to fine-tune the angle of the visible light camera, so that the optical centers of the infrared and visible light cameras can intersect on the photovoltaic panel at any flight altitude. This effectively eliminates the dynamic parallax caused by the physical baseline and ensures the correspondence between the suspected infrared hot spot and the close-up visible light image in physical space.

[0016] 2. In this invention, by extracting the instantaneous spatial pose, time and position data of the UAV, the polarization direction of the incident sunlight and the reflected light from the surface is calculated in a feedforward manner, the target extinction angle is directly obtained and the rotatable circular polarizing mirror is driven in one step. This eliminates the traditional time-consuming blind scanning mechanism based on image feedback, and achieves zero-delay filtering of glare on photovoltaic glass mirrors, ensuring that a clear and non-reflective underlying physical image can be obtained instantly even in strong light environments.

[0017] 3. In this invention, a two-way feature verification mechanism is introduced, which logically cross-verifies the confidence of infrared hot spots, the physical damage identification results of visible light images, and the geometric overlap of camera reflection. This can identify and automatically eliminate false hot spots formed by environmental heat sources such as the sky or clouds reflecting off the glass surface, overcoming the defect that relying solely on infrared temperature measurement is susceptible to interference, and improving the accuracy and reliability of defect diagnosis. Attached Figure Description

[0018] Figure 1 This is a system framework diagram of a dual-camera intelligent cruise detection device for photovoltaic power plants proposed in this invention; Figure 2 This is a flowchart of a method for intelligent cruise detection of photovoltaic power plants using dual cameras by drones, as proposed in this invention. Detailed Implementation

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

[0020] Example 1 In a first embodiment of the present invention, the present invention provides a dual-camera intelligent cruise detection device for photovoltaic power plants using unmanned aerial vehicles, such as... Figure 1 As shown, it includes the following steps: The multi-source data acquisition module simultaneously acquires images of the photovoltaic panel through an infrared thermal imaging camera and a visible light camera, and uses a laser rangefinder to obtain the vertical distance to the photovoltaic panel. The visible light camera is equipped with a stepper motor and a rotatable circular polarizing mirror. In the multi-source data acquisition module, the steps for obtaining the vertical distance to the photovoltaic panel using a laser rangefinder include: A laser pulse is emitted using a laser rangefinder installed between an infrared thermal imaging camera and a visible light camera; It receives the echo signal reflected by the photovoltaic panel and calculates the vertical distance based on the flight time; Specifically, the multi-source data acquisition module is mounted on the airborne pod base plate of the UAV, and its hardware structure mainly includes an infrared thermal imaging camera, a visible light camera, and a laser rangefinder.

[0021] An infrared thermal imaging camera and a visible light camera are rigidly mounted on the pod substrate at a fixed baseline distance. They are used to simultaneously acquire image data of the photovoltaic panel as the UAV flies along a set route. The infrared thermal imaging camera continuously outputs an image of the infrared thermal radiation distribution on the surface of the photovoltaic panel, while the visible light camera simultaneously outputs a visible light image of the underlying physical texture of the photovoltaic panel.

[0022] A stepper motor is mechanically connected to the bottom of the visible light camera, which is controlled to perform micro-adjustments of multi-axis angles. A rotatable circular polarizer is physically mounted at the front end of the optical lens of the visible light camera. This polarizer is driven by an independent micro-gear mechanism to perform rotation based on the central optical axis, thereby changing the direction of the transmission axis to filter reflected light at a specific polarization angle.

[0023] The laser rangefinder is mounted at the midpoint of the physical baseline formed by the line connecting the infrared thermal imaging camera and the visible light camera. This laser rangefinder is electrically connected to the edge computing processing unit of the UAV via a data bus to obtain the vertical distance between the detection device and the surface of the photovoltaic panel directly below it in real time.

[0024] In the multi-source data acquisition module, the step of obtaining the vertical distance to the photovoltaic panel using a laser rangefinder is performed according to the following procedure.

[0025] The first step involves the multi-source data acquisition module sending a trigger command to the laser rangefinder's transmitter. The laser rangefinder then emits intensity-modulated laser pulses towards the surface of the photovoltaic panel directly below. A high-precision timer within the system synchronously records the start timestamp of the laser pulse emission.

[0026] The second step involves the laser pulse reflecting off the glass surface of the photovoltaic panel, generating an echo signal. The receiver of the laser rangefinder continuously listens for this echo signal, and when it is detected, a timer records the end timestamp of the echo reception.

[0027] The third step involves the control unit reading the start and end timestamps, calculating the time difference between them, and using this difference as the flight time of the laser pulse. Subsequently, the system calculates the vertical distance from the laser rangefinder to the photovoltaic panel based on this flight time, using the following formula: ; In this formula, Indicates the vertical distance from the device to the surface of the photovoltaic panel; This represents the constant velocity of a laser pulse in the current air medium. This represents the calculated time of flight of the laser pulse.

[0028] The system repeats the above transmission, reception, and calculation steps at a set sampling frequency, and transmits the output vertical distance data sequence to the subsequent data processing nodes in real time, providing a spatial scale reference for parallax compensation and spatial pose calculation of visible light cameras.

[0029] The dynamic parallax compensation module calculates the dynamic parallax compensation angle based on the vertical distance and camera mounting spacing, and drives the stepper motor to adjust the angle so that the optical centers of the two cameras intersect. In the dynamic parallax compensation module, the steps for calculating the dynamic parallax compensation angle include: Obtain the camera mounting distance between the two cameras; The arctangent angle is calculated by the ratio of the camera mounting spacing to the vertical distance, and the arctangent angle is used as the dynamic parallax compensation angle. Specifically, the dynamic parallax compensation module establishes a communication connection with the multi-source data acquisition module and the stepper motor via an electrical control bus. Because the infrared thermal imaging camera and the visible light camera are rigidly fixed on the same airborne pod base plate, a fixed physical distance exists between the optical centers of the two cameras, causing spatial parallax when observing the same target due to the initially parallel optical axes of the two cameras. The dynamic parallax compensation module receives real-time vertical distance data output from the multi-source data acquisition module, calculates and outputs the corresponding motor drive signal to solve the technical problem of field-of-view deviation in dual-light images.

[0030] In the dynamic parallax compensation module, the steps for calculating the dynamic parallax compensation angle are performed as follows.

[0031] The first step involves the control unit within the dynamic parallax compensation module reading the camera mounting distance between the two cameras from memory. This camera mounting distance is the known straight-line physical distance between the perpendicular lines connecting the main optical centers of the infrared thermal imaging camera and the main optical centers of the visible light camera.

[0032] The second step involves the control unit continuously acquiring vertical distance data transmitted from the multi-source data acquisition module. The system establishes a spatial right-angled triangle geometric solution model, using the read camera installation spacing as the opposite side parameter of the right-angled triangle and the acquired vertical distance as the adjacent side parameter, and calculates the arithmetic ratio of these two parameters.

[0033] Third, the system performs an arctangent mathematical operation on the obtained ratio to obtain the corresponding spatial deflection angle value, and sets this arctangent angle as the dynamic disparity compensation angle for the current control cycle. The specific calculation formula for this step is as follows: ; In this formula, This represents the dynamic disparity compensation angle calculated by the system. This indicates the camera mounting distance between the infrared thermal imaging camera and the visible light camera as read by the system. This indicates the real-time vertical distance from the device, which is input from the multi-source data acquisition module, to the photovoltaic panel.

[0034] The fourth step is for the dynamic parallax compensation module to calculate the dynamic parallax compensation angle. This is converted into a drive pulse signal corresponding to the number of steps. The control unit then sends this drive pulse signal through electrical circuitry to the stepper motor mounted at the bottom of the visible light camera.

[0035] The stepper motor drives the mechanical axis of the visible light camera to rotate at a corresponding angle based on the received drive pulse signal. After the rotation is completed, the observation optical axis of the visible light camera tilts, causing the extended optical center line of the infrared thermal imaging camera to overlap with the extended optical center line of the visible light camera at a distance of [missing information]. The two cameras converge at the surface of the photovoltaic panel. This electromechanical linkage ensures that the center points of the fields of view of the images output by the two cameras at the current flight altitude remain spatially aligned, thus achieving a physical alignment effect of the image data.

[0036] The spatial pose analysis module extracts the current UAV attitude data, position data, time data, and dynamic parallax compensation angle when a suspected hot spot is detected. In the spatial pose analysis module, the steps for extracting the current UAV attitude data, position data, time data, and dynamic parallax compensation angle include: When an abnormally high temperature region is detected by scanning an infrared image using a threshold segmentation method, the coordinates of its center pixel are extracted as a suspected hot spot. Record and extract the instantaneous roll angle, pitch angle, and yaw angle as UAV attitude data, simultaneously record latitude and longitude as position data, and read the system absolute time as time data; Specifically, following the multi-source data acquisition module and the dynamic parallax compensation module, a spatial pose analysis module is also included. This spatial pose analysis module establishes data communication connections with the infrared thermal imaging camera, the UAV's onboard inertial navigation system, the satellite positioning system, the system clock circuit, and the aforementioned dynamic parallax compensation module. The spatial pose analysis module is used to lock onto the moment a physical defect is detected in the infrared field of view, simultaneously locking in the multi-dimensional spatial and temporal state parameters of the UAV platform, providing accurate spatiotemporal reference data for subsequent calculations of the geometric relationship of solar reflection.

[0037] In the spatial pose analysis module, the steps for extracting the current UAV attitude data, position data, time data, and dynamic parallax compensation angle are performed according to the following process.

[0038] The first step involves the spatial pose resolution module's computational unit continuously receiving real-time infrared image frame sequences output by the infrared thermal imaging camera and performing a global pixel grayscale scan on each frame of the infrared image using an image thresholding method. The system presets a grayscale threshold corresponding to a fixed temperature in its internal registers and compares the pixel grayscale values ​​in the infrared image one by one.

[0039] The second step involves identifying a closed, connected region in the infrared image where the pixel grayscale value exceeds the preset grayscale threshold as a localized abnormal high-temperature region. Subsequently, the calculation unit extracts the center pixel coordinates of this abnormal high-temperature region within the two-dimensional pixel coordinate system of the infrared image and sets these center pixel coordinates as the suspected hot spot. The analytical calculation formula for these center pixel coordinates is as follows: ; ; In the above formula, The x-coordinate represents the center pixel of the extracted suspected hot spot; The vertical coordinate represents the center pixel of the extracted suspected hot spot; This indicates the total number of pixels contained within the abnormally high temperature region; Indicates the first in this region The x-coordinate of each pixel; Indicates the first in this region The ordinate of each pixel.

[0040] Third, at the instant the calculation unit obtains the coordinates of the center pixel, the spatial pose analysis module generates a hardware synchronization interrupt signal. In response to this interrupt signal, the system immediately reads the spatial three-axis attitude angles currently output by the airborne inertial navigation system, specifically recording the roll, pitch, and yaw angle values ​​of the UAV under the current system, and packaging these three parameters as the UAV's attitude data at the current moment.

[0041] The fourth step involves the spatial pose analysis module reading the geographic location coordinates output by the satellite positioning system within the same hardware synchronization interrupt cycle. Specifically, it records the longitude and latitude values ​​of the current location of the UAV and packages them as the location data at the current moment.

[0042] Fifth, the spatial pose resolution module accesses the system clock circuit, reads the current system absolute timestamp, and uses it as the current time data; simultaneously, it reads the dynamic parallax compensation angle output by the dynamic parallax compensation module within the current control cycle via the internal data bus. The system combines and encapsulates the above-mentioned UAV attitude data, position data, time data, and dynamic parallax compensation angle, and outputs them to the next-level polarization extinction feedforward control module.

[0043] The polarization extinction feedforward control module calculates the solar incident direction based on position and time data, and calculates the polarization direction of reflected light by combining UAV attitude data, dynamic parallax compensation angle and vertical direction of panel to obtain the target extinction angle, and drives the rotatable circular polarizer to rotate to that angle to capture visible light image. In the polarization extinction feedforward control module, the steps for calculating the polarization direction of the reflected light include: The true observation direction is calculated based on the initial observation direction of the visible light camera, the attitude data of the UAV, and the rotation state after performing dynamic parallax compensation. Input location and time data into the ephemeris database to construct the direction of solar incidence; The polarization direction of the reflected light is calculated by cross product of the vertical direction of the panel and the direction of solar incidence. In the polarization extinction feedforward control module, the steps for obtaining the target extinction angle include: The polarization direction of the reflected light is projected onto an imaging plane with the actual observation direction as the normal direction; Obtain the horizontal reference direction of the internal photosensitive element of the visible light camera; Calculate the angle between the projection direction and the horizontal reference direction, and add 90 degrees to the angle to obtain the target extinction angle; Specifically, a polarization extinction feedforward control module is cascaded after the spatial pose analysis module. This module's hardware includes an embedded microprocessor, internal memory, and a control interface. Its data input is electrically connected to the spatial pose analysis module via a data bus, and its control output is electrically connected to a micro-gear mechanism mounted in front of a visible light camera. The polarization extinction feedforward control module addresses the technical problem of specular reflection from the photovoltaic panel glass surface masking underlying physical damage in strong light environments. It directly outputs an extinction drive signal by calculating spatial geometric relationships.

[0044] In the polarization extinction feedforward control module, the operation steps for calculating the polarization direction of the reflected light are executed according to the following process.

[0045] The first step involves the microprocessor reading the initial observation direction vector of the visible light camera embedded within the system, and simultaneously reading the UAV attitude data (roll, pitch, and yaw angles) and dynamic parallax compensation angle from the spatial pose analysis module. The system uses the UAV attitude data to construct a three-dimensional Euler rotation matrix that transforms the system coordinate system to the global geographic coordinate system. The physical rotation state after the stepper motor performs dynamic parallax compensation is then superimposed onto this rotation matrix to calculate the true observation direction of the visible light camera in the global geographic coordinate system, which is defined as the true observation direction vector. .

[0046] The second step involves the microprocessor extracting the location data (longitude and latitude) and time data (system absolute time) from the spatial pose resolution module. The system then inputs this location and time data into the solar ephemeris database built into the memory. The ephemeris calculation program outputs the solar altitude angle and solar azimuth angle corresponding to the current geographic coordinates. Based on this, the system constructs the spatial direction of sunlight in the global geographic coordinate system, defining it as the solar incidence direction vector. .

[0047] The third step involves the microprocessor calling the pre-defined vertical direction of the photovoltaic panel surface (i.e., the normal direction of the photovoltaic panel array) and defining it as the vertical direction vector of the panel. According to the laws of optical polarization physics, after natural light is reflected from a dielectric surface, its principal polarization direction is perpendicular to the incident plane formed by the incident light and the surface normal. The system's perpendicular vector to the panel... and the vector of the direction of solar incidence Performing a cross product yields the polarization direction vector of the reflected light. The specific calculation formula for this step is as follows: ; In this formula, This represents a three-dimensional spatial vector indicating the polarization direction of the reflected light, obtained from the calculation. This represents a three-dimensional spatial vector of the solar incidence direction constructed based on location and time data. This represents a three-dimensional spatial vector perpendicular to the surface of the photovoltaic panel. This represents the cross product operation of vectors.

[0048] In the polarization extinction feedforward control module, the operation steps for obtaining the target extinction angle are executed according to the following process.

[0049] The first step is to establish a two-dimensional imaging plane coordinate system based on the physical plane containing the photosensitive element inside the visible light camera. The microprocessor then converts the actual observation direction vector... This serves as the normal vector to the two-dimensional imaging plane. The system applies a three-dimensional to two-dimensional orthogonal projection algorithm to calculate the polarization direction vector of the reflected light. Projected onto the camera's imaging plane, this yields a two-dimensional projection direction vector. .

[0050] The second step involves the microprocessor retrieving the horizontal reference direction of the internal photosensitive element of the visible light camera from the hardware registers. This horizontal reference direction corresponds to the horizontal axis of the pixel array of the photosensitive element and is defined here as the reference direction vector. .

[0051] Third step: The system calculates the projection direction vector. With the reference direction vector The angle between them in the imaging plane Since a circular polarizer reaches its maximum extinction rate when its transmission axis is orthogonal to the polarization direction of the reflected light, the system at this angle... Add ninety degrees to the numerical value. (radians), to calculate the target extinction angle The specific mathematical expression for this calculation step is as follows: ; ; In the above formula, This represents the plane angle between the projection direction vector and the horizontal reference direction vector; This represents the projection direction vector of the polarization direction of the reflected light onto the camera's imaging plane. This represents the horizontal reference direction vector of the photosensitive element; Represents the dot product operation of vectors; Represents the magnitude of the vector; This represents the final calculated target extinction angle.

[0052] The fourth step involves the polarization extinction feedforward control module calculating the target extinction angle. This signal is converted into a corresponding motor control electrical signal and sent to a miniature gear at the front end of the visible light camera. Upon receiving the signal, the miniature gear rotates, directly driving a rotatable circular polarizer to rotate once around its optical central axis to the target extinction angle. The physical location.

[0053] Fifth, after the hardware feedback that the rotatable circular polarizer has reached the target extinction angle position is received, the polarization extinction feedforward control module sends a shutter trigger command to the visible light camera. The visible light camera performs image exposure under this polarization filtering state, captures a visible light image with surface reflections removed, and outputs this image data to the next processing module.

[0054] The two-way feature verification module identifies physical damage in visible light images, determines the real fault by combining the confidence level of suspected hot spots, and eliminates artifact data by combining the reflection geometric overlap. In the bidirectional feature verification module, the steps for determining the actual fault based on the confidence level of suspected hot spots include: The confidence level of abnormal high temperature output from the infrared thermal imaging camera is used as the confidence level of suspected hot spots; Analyze visible light images using target detection algorithms to obtain confidence levels of physical damage; When the confidence level of the suspected hot spot is greater than the preset infrared threshold and the confidence level of the physical damage is greater than the preset visible light threshold, a real fault is confirmed. In the bidirectional feature verification module, the steps for removing artifact data by combining reflection geometric overlap include: The direction of specular reflection of sunlight is calculated based on the direction of sunlight incident and the perpendicular direction of the panel; Calculate the degree of collinearity between the observation direction of the visible light camera and the reflection direction of the mirror when the camera captures the image, and take the absolute value of this degree of overlap as the reflection geometric overlap. In the bidirectional feature verification module, the steps for removing artifact data include: When the confidence level of the suspected hot spot is greater than the preset infrared threshold and the confidence level of the physical damage is less than or equal to the preset visible light threshold, and the reflectance geometric overlap is greater than the preset overlap threshold; The current signal is determined to be a specular heating artifact caused by an ambient heat source, and this data is discarded as artifact data. Specifically, a bidirectional feature verification module is installed after the polarization extinction feedforward control module. This module serves as the logic judgment terminal of the detection device. Its input terminal receives feature data output from the infrared thermal imaging camera, the visible light image frame after polarization extinction, and the geometric parameter vector calculated by the polarization extinction feedforward control module via a data bus. The bidirectional feature verification module is used to solve the problem of false alarms caused by environmental heat source radiation interference during inspection through cross-verification of multi-source features, thereby improving the accuracy of defect diagnosis.

[0055] In the bidirectional feature verification module, the operation steps for determining the actual fault by combining the confidence level of suspected hot spots are executed according to the following process.

[0056] The first step involves the module acquiring the abnormal high temperature confidence level synchronously output by the infrared thermal imaging camera when a suspected hot spot is detected, and using this as the suspected hot spot confidence level. This suspected hot spot confidence level is calculated by the internal processing unit of the infrared camera based on the temperature rise rate of the abnormal area and the deviation between the local maximum temperature and the background average temperature.

[0057] The second step involves the module calling its internally stored target detection algorithm to analyze the polarized extinction visible light image. This target detection algorithm uses a pre-set deep convolutional neural network to extract edge texture, color contrast, and geometric contour features from the image, identify whether there is physical damage to the photovoltaic panel surface in the visible light image, and output the corresponding physical damage confidence value.

[0058] Third, the calculation unit reads the system's preset infrared and visible light thresholds. The system performs a logical judgment operation. When the confidence level of the suspected hot spot is greater than the preset infrared threshold and the confidence level of the physical damage is greater than the preset visible light threshold, the system confirms that the suspected hot spot is caused by a fault in the underlying physical structure of the photovoltaic panel, and determines that the current detection target has the actual fault.

[0059] In the bidirectional feature verification module, the steps for removing artifact data by combining reflection geometric overlap are executed according to the following process.

[0060] The first step involves the computational unit extracting the solar incident direction vector and the vertical direction vector of the panel, calculated by the polarization extinction feedforward control module. The system then uses the specular reflection optics law to calculate the specular reflection direction of sunlight after reflection from the photovoltaic panel surface at the current moment, defining it as the specular reflection direction vector. The specific calculation formula for this step is as follows: ; In this formula, This represents the three-axis unit vector indicating the calculated direction of sunlight reflection from the specular surface. The unit vector representing the direction of solar incidence; The unit vector representing the vertical direction of the panel; This represents the dot product operation of vectors.

[0061] The second step involves the computing unit extracting the actual observation direction vector during the visible light camera's capture. The system calculates the degree of collinearity between the actual observed direction vector and the specular reflection direction vector. The absolute value of this degree of coincidence is taken as the geometric coincidence degree of reflection. Its specific mathematical expression is as follows: ; In the above formula, This represents the calculated reflection geometric coincidence value, which ranges from 0 to 1. This represents the unit vector indicating the true observation direction of the visible light camera at the moment of capture. This represents the calculated unit vector of the direction of sunlight reflection from the specular surface. This indicates the operation of taking the absolute value.

[0062] In the bidirectional feature verification module, the steps for removing artifact data are performed according to the following process.

[0063] The first step is for the computing unit to execute a comprehensive logic verification program. The system reads the confidence scores for the suspected hot spots, physical damage, and reflection geometric overlap values ​​for the current detection period.

[0064] The second step involves comparing the three parameters mentioned above with preset judgment criteria. When the confidence level of the suspected hot spot is greater than the preset infrared threshold, and the confidence level of the physical damage is less than or equal to the preset visible light threshold, and the geometric overlap of the reflection is greater than the preset overlap threshold, the system determines that the current infrared signal is not caused by a physical fault inside the component, but rather by a false high-temperature signal formed by specular reflection of an ambient heat source on the surface of the photovoltaic glass.

[0065] Third, the system identifies the suspected hot spot as a specular heating artifact caused by an environmental heat source and generates a removal instruction. This data is removed from the current inspection defect record form as artifact data and does not proceed to the subsequent fault warning process. Through the aforementioned two-way verification based on optical reflection geometry and multispectral confidence, the reliability of the device in identifying real faults is improved.

[0066] Example 2: In summer, under strong direct sunlight or with extensive cloud cover, the highly reflective glass material on the surface of photovoltaic panels exhibits severe specular reflection. When drones conduct inspections, radiation from the sun or high-temperature clouds is reflected by the glass and enters the infrared thermal imaging camera, creating false high-temperature areas in the infrared image that are highly similar in geometric features and temperature rise range to real hot spots generated by internal short circuits in the modules. Because conventional inspection systems lack the ability to calculate the reflection geometry in real time, and visible light cameras often suffer from overexposure due to specular glare, they cannot extract physical damage features from the panel surface for verification, resulting in numerous false alarms. To address these issues, this invention provides a dual-camera intelligent cruise detection method for photovoltaic power plants using drones, the structure of which is as follows... Figure 2 As shown. The specific implementation process of this method is as follows: S1, the infrared thermal imaging camera and visible light camera of the multi-source data acquisition module simultaneously acquire images of the photovoltaic panel, and the laser rangefinder is used to obtain the vertical distance to the photovoltaic panel. S2, through the dynamic parallax compensation module, calculates the dynamic parallax compensation angle based on the vertical distance and camera mounting distance, drives the stepper motor to adjust the angle, so that the optical centers of the two cameras intersect; S3, when a suspected hot spot is detected by the spatial pose analysis module, extracts the current UAV attitude data, position data, time data and dynamic parallax compensation angle; S4 calculates the sun's incident direction based on position and time data through the polarization extinction feedforward control module, and calculates the polarization direction of reflected light by combining UAV attitude data, dynamic parallax compensation angle and vertical direction of panel to obtain the target extinction angle, and drives the rotatable circular polarizer to rotate to that angle to capture visible light images. S5 identifies physical damage in visible light images through a two-way feature verification module, determines the true fault by combining the confidence level of suspected hot spots, and eliminates artifact data by combining the reflection geometric overlap.

[0067] Specifically, in step S1, the UAV carrying the detection device flies along a preset route, and the multi-source data acquisition module performs synchronous acquisition. The infrared thermal imaging camera and the visible light camera trigger exposure at the same time reference, acquiring the infrared thermal radiation distribution image and the visible light image of the underlying physical texture of the photovoltaic panel, respectively. A laser rangefinder installed at the midpoint of the baselines of the two cameras emits intensity-modulated laser pulses directly below the photovoltaic panel surface. The system's high-precision timing circuit records the flight time of the laser pulse from emission to receiving the echo signal, and calculates the real-time vertical distance between the device and the photovoltaic panel surface based on this time value and the constant speed of light in the air. The acquired image data and vertical distance data are transmitted in real-time to the back-end processing unit via a data bus.

[0068] In step S2, the dynamic parallax compensation module performs real-time calculations and electromechanical adjustments. The system reads the fixed installation distance between the infrared thermal imaging camera and the visible light camera stored in its internal registers. The calculation unit uses this fixed installation distance as the opposite side length of a geometric triangle and the acquired real-time vertical distance as the adjacent side length, calculating the current dynamic parallax compensation angle using the arctangent function. Subsequently, the system converts this angle value into a pulse drive signal with a specific number of steps and sends it to the stepper motor at the bottom of the visible light camera. The stepper motor drives the visible light camera's axis of rotation to deflect, causing the visible light camera's observation optical axis to point towards the center of the infrared thermal imaging camera's field of view. This process achieves the physical convergence of the optical centers of the two cameras on the surface of the target photovoltaic panel, eliminating the dynamic parallax caused by the baseline distance.

[0069] In step S3, the spatial pose analysis module performs hotspot locking and parameter latching. The calculation unit performs a global grayscale scan on the real-time input infrared image and compares the pixels using an image thresholding algorithm. When a closed region with a continuously distributed grayscale value exceeding a preset grayscale threshold appears in the image, the system determines that an abnormally high-temperature region exists. The calculation unit extracts the centroid coordinates of the pixels in this region as suspected hotspots. At the instant the coordinate extraction is completed, the system triggers a hardware synchronization interrupt, immediately reads the current roll angle, pitch angle, and yaw angle from the airborne sensor system as UAV attitude data, synchronously reads the current latitude and longitude coordinates as position data, and accesses the system clock to obtain the current absolute timestamp. The above four data items, together with the current dynamic parallax compensation angle, are encapsulated into a spatiotemporal reference packet and transmitted to the control terminal.

[0070] In step S4, the polarization extinction feedforward control module performs optical intervention. The calculation unit inputs the latched position and time data into the built-in solar ephemeris database, retrieves and constructs the solar incidence direction vector at the current moment. The system combines the UAV attitude data and dynamic parallax compensation angle to calculate the true observation direction of the visible light camera in the global coordinate system. Subsequently, the system performs spatial geometric calculations using the preset photovoltaic panel normal direction vector and the solar incidence direction vector to obtain the polarization direction vector of the reflected light. This vector is orthogonally projected onto the camera's photosensitive element plane, and the plane angle between the projection direction and the horizontal reference axis of the photosensitive element is calculated. Based on this, a 90-degree phase compensation is superimposed to obtain the final target extinction angle. The system drives the micro-gear mechanism at the front end of the rotatable circular polarizer to precisely rotate the transmission axis of the polarizer to this angle, physically filtering out specular reflections from the glass surface. The system then triggers the shutter to capture a clear image of the physical damage to the panel.

[0071] In step S5, the bidirectional feature verification module performs the final logical judgment and fault classification. The system first obtains the confidence level of the suspected hot spot output by the infrared thermal imaging camera, and then uses a deep convolutional neural network to extract physical damage features from the polarized extinction visible light image, outputting the physical damage confidence level. Simultaneously, the system calculates the specular reflection vector of sunlight using the solar incident direction and the panel normal direction, and calculates the degree of collinearity between the actual observation direction and this reflection vector, obtaining the reflection geometric overlap degree.

[0072] When the confidence level of a suspected hot spot is greater than a preset infrared threshold, and the confidence level of physical damage is greater than a preset visible light threshold, the system determines that the target point is a real fault caused by component failure and records the fault coordinates. When the confidence level of a suspected hot spot is greater than a preset infrared threshold, but the confidence level of physical damage is less than or equal to a preset visible light threshold, and the reflection geometric overlap is greater than a preset overlap threshold, the system determines that the signal is a reflection artifact of a high-heat source in the environment on the glass surface. The system executes a data rejection command to remove the artifact data from the defect form, thereby ensuring the validity and accuracy of the inspection results.

[0073] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dual-camera intelligent cruise detection device for photovoltaic power plants using unmanned aerial vehicles (UAVs), characterized in that, include: The multi-source data acquisition module simultaneously acquires images of the photovoltaic panel through an infrared thermal imaging camera and a visible light camera, and uses a laser rangefinder to obtain the vertical distance to the photovoltaic panel. The visible light camera is equipped with a stepper motor and a rotatable circular polarizing filter. The dynamic parallax compensation module calculates the dynamic parallax compensation angle based on the vertical distance and camera mounting distance, and drives the stepper motor to adjust the angle so that the optical centers of the two cameras intersect. The spatial pose analysis module extracts the current UAV attitude data, position data, time data, and the dynamic parallax compensation angle when a suspected hot spot is detected. The polarization extinction feedforward control module calculates the solar incident direction based on the position data and time data, and calculates the polarization direction of the reflected light by combining the UAV attitude data, dynamic parallax compensation angle and the vertical direction of the panel to obtain the target extinction angle, and drives the rotatable circular polarizing mirror to rotate to that angle to capture a visible light image. The bidirectional feature verification module identifies physical damage in the visible light image, determines the actual fault by combining the confidence level of suspected hot spots, and eliminates artifact data by combining the reflection geometric overlap.

2. The photovoltaic power station UAV dual-camera intelligent cruise detection device according to claim 1, characterized in that, In the multi-source data acquisition module, the step of obtaining the vertical distance to the photovoltaic panel using a laser rangefinder includes: A laser pulse is emitted using a laser rangefinder installed between the infrared thermal imaging camera and the visible light camera; The system receives the echo signal reflected by the photovoltaic panel and calculates the vertical distance based on the time of flight.

3. The photovoltaic power station UAV dual-camera intelligent cruise detection device according to claim 1, characterized in that, In the dynamic parallax compensation module, the step of calculating the dynamic parallax compensation angle includes: Obtain the camera installation distance between the two cameras; The arctangent angle is calculated from the ratio of the camera mounting spacing to the vertical distance, and the arctangent angle is used as the dynamic parallax compensation angle.

4. The photovoltaic power station UAV dual-camera intelligent cruise detection device according to claim 1, characterized in that, In the spatial pose analysis module, the steps of extracting the current UAV attitude data, position data, time data, and the dynamic parallax compensation angle include: When an abnormally high temperature region is detected by scanning an infrared image using a threshold segmentation method, the coordinates of its center pixel are extracted as the suspected hot spot. The instantaneous roll angle, pitch angle, and yaw angle are recorded as the attitude data of the UAV, the latitude and longitude are recorded simultaneously as the position data, and the system absolute time is read as the time data.

5. The photovoltaic power station UAV dual-camera intelligent cruise detection device according to claim 1, characterized in that, In the polarization extinction feedforward control module, the step of calculating the polarization direction of the reflected light includes: The true observation direction is calculated based on the initial observation direction of the visible light camera, the attitude data of the UAV, and the rotation state after performing the dynamic parallax compensation angle. The location data and the time data are input into the ephemeris database to construct the solar incidence direction; The polarization direction of the reflected light is calculated by cross product of the vertical direction of the panel and the direction of solar incidence.

6. The photovoltaic power station UAV dual-camera intelligent cruise detection device according to claim 5, characterized in that, In the polarization extinction feedforward control module, the step of obtaining the target extinction angle includes: The polarization direction of the reflected light is projected onto an imaging plane with the actual observation direction as the normal direction; Obtain the horizontal reference direction of the photosensitive element inside the visible light camera; Calculate the angle between the projection direction and the horizontal reference direction, and add 90 degrees to the angle to obtain the target extinction angle.

7. The photovoltaic power station UAV dual-camera intelligent cruise detection device according to claim 1, characterized in that, In the bidirectional feature verification module, the step of determining the actual fault by combining the confidence level of the suspected hot spot includes: The confidence level of the abnormal high temperature output by the infrared thermal imaging camera is used as the confidence level of the suspected hot spot; The visible light image is analyzed using a target detection algorithm to obtain the confidence level of physical damage; When the confidence level of the suspected hot spot is greater than a preset infrared threshold and the confidence level of the physical damage is greater than a preset visible light threshold, the existence of the actual fault is confirmed.

8. The photovoltaic power station UAV dual-camera intelligent cruise detection device according to claim 1, characterized in that, In the bidirectional feature verification module, the step of combining reflection geometric overlap to remove artifact data includes: The specular reflection direction of sunlight is calculated based on the solar incident direction and the perpendicular direction of the panel; Calculate the degree of collinearity between the observation direction of the visible light camera and the reflection direction of the mirror when the camera captures the image, and use the absolute value of this degree of overlap as the reflection geometric overlap.

9. The photovoltaic power station UAV dual-camera intelligent cruise detection device according to claim 8, characterized in that, In the bidirectional feature verification module, the step of removing artifact data includes: When the confidence level of the suspected hot spot is greater than a preset infrared threshold and the confidence level of the physical damage is less than or equal to a preset visible light threshold, and the reflective geometric overlap is greater than a preset overlap threshold; If the current signal is determined to be a specular heating artifact caused by an ambient heat source, this data is discarded as the artifact data.

10. A method for intelligent cruise detection of photovoltaic power stations using a dual-camera drone, characterized in that, The method for a dual-camera intelligent cruise detection device for a photovoltaic power station drone according to any one of claims 1-9 includes: S1, the infrared thermal imaging camera and visible light camera of the multi-source data acquisition module simultaneously acquire images of the photovoltaic panel, and the vertical distance to the photovoltaic panel is obtained using a laser rangefinder; S2, the dynamic parallax compensation module calculates the dynamic parallax compensation angle based on the vertical distance and camera installation spacing, drives the stepper motor to adjust the angle, so that the optical centers of the two cameras intersect; S3, when a suspected hot spot is detected by the spatial pose analysis module, the current UAV attitude data, position data, time data and the dynamic parallax compensation angle are extracted; S4, the polarization extinction feedforward control module calculates the sun's incident direction based on the position data and time data, and calculates the polarization direction of the reflected light by combining the UAV attitude data, dynamic parallax compensation angle and the vertical direction of the panel to obtain the target extinction angle, and drives the rotatable circular polarizing mirror to rotate to that angle to capture a visible light image; S5, the physical damage of the visible light image is identified by the bidirectional feature verification module, the real fault is determined by the confidence of suspected hot spots, and artifact data is eliminated by the reflection geometric overlap.

Images