A searchlight control system for a drone

Abstract

The present application relates to the technical field of unmanned aerial vehicle searchlight control, and specifically relates to a searchlight control system of unmanned aerial vehicle, comprising a multi-source data acquisition module for obtaining water surface image and unmanned aerial vehicle flight data; a water surface type analysis module for determining calm or wave water surface according to image gray variance and edge density; a searchlight output sub-control module for selecting light control strategy according to water surface type, combining multi-feature recognition high confidence target and determining lamp wick power distribution parameters; a searchlight output execution module for driving lamp wick light according to parameters; a target real-time tracking module for calculating searchlight projection angle adjustment amount to aim at the target; and an image verification feedback module for calculating image definition every certain frame number and triggering system adjustment, which can solve the problem that the existing searchlight system cannot dynamically adapt to water surface optical characteristics, suppress reflection interference and accurately illuminate in the night water complex environment, and significantly improves the accuracy and efficiency of target recognition.

Description

Technical Field

[0001] This invention relates to the field of drone searchlight control technology, and more specifically to a searchlight control system for drones. Background Technology

[0002] With the increasing application of drone technology in inspection and rescue, nighttime water patrol and search and rescue missions have placed higher demands on the intelligence level of airborne searchlight systems. As a key mission payload, the searchlights carried by drones directly affect the accuracy of target identification and mission execution efficiency. However, the adaptability of existing drone searchlight control systems in complex water environments is still significantly insufficient, especially in scenarios with multiple interference factors such as strong reflection, dynamic waves, and low illumination, making it difficult to achieve stable and accurate lighting control.

[0003] Among these, searchlight control technology primarily focuses on basic angle adjustment and fixed power output, lacking the ability to dynamically perceive and respond to the optical characteristics of the water surface. Current mainstream solutions mostly adopt a uniform brightness illumination strategy, failing to distinguish the reflection differences between calm and wavy water surfaces. This makes the image acquisition process susceptible to specular reflection or random wave highlights, resulting in overexposure or underexposure of the target area and severely weakening the recognition ability of the visual system. While some improved solutions introduce infrared or laser-assisted positioning, their core objective is obstacle warning or equipment self-localization, failing to address the specific analysis of the contrast between the target and the water surface reflection in water rescue scenarios.

[0004] Existing technologies for searchlight control suffer from several limitations. First, light intensity adjustment relies on preset levels or simple photosensitivity feedback, failing to dynamically adjust the output power of each lamp element based on real-time water surface conditions. Second, there is a lack of effective mechanisms for differentiating water surface types, failing to adopt differentiated control strategies for the uniform reflection of calm water surfaces and the unsteady scattering of wavy water surfaces. Third, target recognition and lighting control are disconnected; the searchlight illumination area and potential target locations do not coordinate, resulting in wasted effective lighting resources in non-critical areas. Finally, in environments with strong reflections, the lack of ambient light compensation and multi-feature fusion target verification mechanisms makes it easy to misjudge waves as targets awaiting rescue, leading to incorrect lighting focus. These problems make existing systems unsuitable for high-precision, high-reliability scenarios such as nighttime water rescue, necessitating an intelligent searchlight control system that integrates water surface state perception, reliable target recognition, and refined searchlight control. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a searchlight control system for unmanned aerial vehicles (UAVs). This system can solve the problem that existing UAV searchlight systems cannot dynamically adapt to changes in the optical characteristics of the water surface, effectively suppress reflection interference, and achieve adaptive and precise illumination of the target area in complex scenarios such as nighttime water patrol and rescue. The present invention achieves coordinated control of the searchlight array through multi-source data perception and intelligent decision-making, thereby improving the accuracy and efficiency of target identification in complex optical media environments.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A searchlight control system for an unmanned aerial vehicle (UAV) includes:

[0008] The multi-source data acquisition module acquires water surface images captured by the visual camera mounted on the UAV, and also acquires UAV flight data information.

[0009] The water surface type analysis module determines the water surface type based on the grayscale variance and edge density of the water surface image. The water surface type includes calm water surface and wavy water surface.

[0010] The searchlight output control module selects a lighting control strategy based on the water surface type. The lighting control strategy includes analyzing the reflectivity of the water surface image and identifying high-confidence targets through multi-feature fusion in conjunction with UAV flight data. Based on the location of the high-confidence targets and the collected parameters, the power allocation parameters of each lamp core are determined.

[0011] The searchlight output execution module drives multiple lamp cores of the searchlight to provide illumination according to the power allocation parameters.

[0012] The real-time target tracking module calculates the adjustment amount of the projection angle of the searchlight in real time based on the position of the high-confidence target or the center point of the water surface image so that the center point of the water surface image always coincides with the high-confidence target.

[0013] The image verification feedback module calculates the image sharpness index of the water surface image at specific frame intervals, and triggers the system to re-evaluate and adjust the image sharpness index based on the image sharpness index.

[0014] Furthermore, the lighting control strategy includes a theoretical sub-strategy and a control sub-strategy. The theoretical sub-strategy is matched to a calm water surface type and includes determining the total power of the searchlight according to a preset mapping table and distributing it evenly to each lamp core. The control sub-strategy is matched to a wavy water surface type and includes determining the individual output power of each lamp core of the searchlight according to a pre-trained neural network model.

[0015] Furthermore, the theoretical sub-strategy includes a preset mapping table, which contains a one-to-one correspondence of total power, altitude, and ambient light intensity. The UAV flight data information includes the current altitude of the searchlight and the current ambient light intensity. The theoretical total power is indexed in the mapping table based on the current altitude of the searchlight and the current ambient light intensity. If there is no completely matching index entry in the mapping table, the theoretical total power is calculated by multidimensional linear interpolation.

[0016] Furthermore, the control sub-strategy includes converting the pixel coordinates of high-confidence targets into target azimuth and pitch deviations, analyzing the current height of the water waves based on the water surface image, and then inputting the current height of the searchlight, the current height of the water waves, the target azimuth and pitch deviations, and the current ambient light intensity into a pre-trained neural network model for forward propagation calculation to output the power allocation parameters for each lamp core.

[0017] Furthermore, the lighting control strategy includes a reflectance analysis step, which includes retrieving the preset standard ambient light intensity and calculating the ambient light compensation coefficient with the current ambient light intensity. Then, based on the ambient light compensation coefficient, the reflectance of the water surface image is calculated pixel by pixel to obtain the water surface reflectance distribution map after environmental compensation.

[0018] Furthermore, the lighting control strategy also includes a reflectivity gradient determination step, a texture feature-assisted filtering step, and a shape feature secondary filtering step.

[0019] The reflectivity gradient determination step involves using the Sobel operator to calculate the gradients in the horizontal and vertical directions of the water surface reflectivity distribution map after environmental compensation, and synthesizing the gradient magnitude. If the gradient magnitude of a pixel is greater than a preset gradient threshold, it is initially marked as a candidate anomaly.

[0020] The texture feature-assisted screening step calculates the local binary pattern texture features of the local neighborhood where the candidate anomaly point is located, and matches them with the templates in the preset target texture template library. If the matching degree is greater than or equal to the preset threshold, the candidate anomaly point is retained and upgraded to a suspected anomaly point.

[0021] The secondary screening step of the shape features involves obtaining the circumscribed contour of the suspected abnormal point through a contour extraction algorithm, and calculating the area and aspect ratio of the minimum bounding rectangle of each contour. If the area is less than a preset area threshold, the suspected abnormal point is determined to be an interference point and is removed. The remaining points are confirmed as the final abnormal points.

[0022] Furthermore, the lighting control strategy also includes an anomaly block clustering step and a target confidence assessment step;

[0023] The anomaly block clustering step involves setting a neighborhood radius and a minimum number of points for the final anomaly points to cluster spatially connected anomaly points into clusters, generating one or more anomaly blocks. Each anomaly block is considered a suspected target to be tracked.

[0024] The target confidence assessment step involves extracting the center coordinates, area value, and mean reflectance of each anomalous block, constructing a confidence function based on the texture matching degree, area value, and mean reflectance to calculate the target confidence, and determining whether the anomalous block is a high-confidence target based on the target confidence.

[0025] Furthermore, the real-time target tracking module includes a searchlight tracking strategy, which includes comparing the center coordinates of the high-confidence target with the center coordinates of the visual image to calculate the pixel deviation value, and calculating the target azimuth adjustment amount and pitch adjustment amount of the searchlight based on the visual camera settings parameters and the current height of the searchlight through geometric relationships.

[0026] Furthermore, the image verification feedback module includes a verification strategy, which includes performing frame-by-frame processing on the water surface image after the detection and control, analyzing the image clarity index of the current frame, comparing the image clarity index with the minimum clarity threshold preset during initialization, and if the clarity index calculated three times consecutively is lower than the minimum clarity threshold, the system is triggered to re-evaluate and adjust the command to optimize the strategy in each module.

[0027] Furthermore, the water surface type analysis module includes extracting an analysis region from the water surface image, calculating its gray-level variance, and obtaining the edge density by analyzing the number of edge pixels in the analysis region through edge detection. The gray-level variance and edge density are compared with preset gray-level variance thresholds and edge density thresholds, respectively. If the gray-level variance is less than or equal to the gray-level variance threshold and the edge density is less than or equal to the edge density threshold, the current water surface is determined to be a calm water surface; otherwise, the current water surface is determined to be a wavy water surface.

[0028] The beneficial effects of this invention are as follows: By setting up water surface type analysis, the environmental adaptability of the illumination system can be improved; by setting up an illumination output sub-control module, the accuracy of target recognition can be improved and fine-grained illumination control can be achieved; and by setting up an image verification feedback module, the image capture can always maintain consistent clarity. Specifically, the water surface type analysis module accurately distinguishes between calm and wavy water surfaces based on the image grayscale variance and edge density, and matches theoretical sub-strategies and control sub-strategies accordingly, avoiding overexposure and underexposure problems caused by uniform illumination, and adapting to complex scenes such as strong water reflection and dynamic waves at night; the lighting control strategy integrates reflectivity analysis and multiple features. The screening and confidence assessment effectively eliminates interference such as waves, accurately identifies high-confidence targets, and solves the problem of target misjudgment in existing systems. The searchlight output control module determines the power allocation parameters of each lamp core for different water surface types through mapping table interpolation or pre-trained neural networks. Combined with the real-time target tracking module, it dynamically adjusts the searchlight angle to avoid wasting lighting resources and ensure accurate illumination of the target area. The image verification feedback module regularly checks the image clarity. If it fails to meet the standard continuously, it triggers the system to re-evaluate and adjust, ensuring that the lighting strategy is always adapted to the environment and target status, and improving the efficiency and reliability of nighttime water patrol and rescue missions. Attached Figure Description

[0029] Figure 1 This is the overall flowchart of the present invention;

[0030] Figure 2 This is a flowchart of the control sub-strategy when the water surface type is a wave surface in this invention;

[0031] Figure 3 This is a flowchart of the water surface type determination process in this invention;

[0032] Figure 4 This is a flowchart of the detection and image analysis process when the water surface type is calm water surface in this invention. Detailed Implementation

[0033] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Identical components are denoted by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0034] Currently, the searchlights on drones used for inspection and rescue operations can only control the angle of the light source, not the light source for specific environments, such as the waterborne inspection and rescue mentioned in this application. During waterborne inspection and rescue, image acquisition is hampered by water surface reflections and wave undulations, making it difficult to clearly extract and track the target. Therefore, this invention designs a multi-source perception and intelligent decision-making drone searchlight control system. Specifically, the structure includes the drone itself and a visual camera and searchlight mounted on it. The searchlight consists of an array integrating several independently controllable LED chips. The calibration of the visual camera involves establishing a precise conversion model from image grayscale values ​​to actual water surface reflectivity by photographing a series of standard grayscale plates under controlled lighting conditions, ensuring the accuracy of subsequent reflectivity calculations. This conversion model is typically a nonlinear function that can be obtained through polynomial fitting.

[0035] System components, such as Figure 1 As shown, the system includes a multi-source data acquisition module, a water surface type analysis module, a searchlight output control module, a searchlight output execution module, a target real-time tracking module, and an image verification feedback module. All modules are integrated within the searchlight controller. Furthermore, the system incorporates two prerequisite steps: a core threshold and a model preset. Through extensive laboratory simulations and actual field tests, a theoretically optimal reflectivity threshold for target visibility in nighttime water surface images was determined. Additionally, by collecting a large sample dataset containing images of calm and wavy water surfaces, grayscale variance and edge density were calculated and statistically analyzed to determine grayscale variance and edge density thresholds that effectively distinguish between calm and wavy water surfaces. Finally, a pre-trained power control model is implanted into the system. This model can be based on a backpropagation neural network, with a network structure including an input layer, at least one hidden layer, and an output layer, or it can be a multivariate linear regression model. The input-output relationship of this model has been verified through extensive simulation and experimental data. For example, under specific input parameter combinations, the model can output a precise power allocation scheme for each lamp core.

[0036] After takeoff, the drone enters a continuous data acquisition mode. The multi-source data acquisition module is responsible for simultaneously acquiring water surface images captured by the visual camera and drone flight data. The drone flight data includes its own flight status parameters and environmental parameters. The flight status parameters include the drone's flight altitude and attitude. The drone's flight altitude can be used to deduce the height of the searchlight's output port from the water surface. The drone's flight altitude can be measured in real time by the millimeter-wave radar on its own body, and it can also be estimated from water surface images captured by the visual camera. Similarly, the two can be combined to obtain an accurate and reliable altitude.

[0037] Environmental parameters include wave height and ambient light intensity. Wave height can be obtained through a combination of visual and laser sensors. A sequence of continuous water surface images captured by a visual camera is analyzed using optical flow or background subtraction methods to determine the vertical displacement and wave frequency of water surface pixels, thereby estimating the relative wave height. Optical flow mainly calculates the motion vectors of pixels between adjacent frames to identify the vertical displacement of water surface texture. A laser wave height sensor mounted on a UAV directly acquires the distance change of the water surface relative to the UAV, and after data processing, accurate wave height data is obtained. This invention prioritizes the use of data from direct measurement methods, with visual estimation results used as auxiliary verification. In addition, ambient light intensity can be acquired in real time by a high-precision photosensitive sensor built into the visual camera, providing a benchmark for the dynamic adjustment of the searchlight to adapt to different nighttime environments such as moonlight, no moonlight, and urban light pollution.

[0038] After the drone collects various data, the system automatically enters the water surface type analysis, such as... Figure 3 As shown, the water surface type analysis module performs real-time analysis on the water surface image to determine whether the current water surface is calm or wavy. Specifically, for the water surface image after preprocessing steps such as noise reduction and distortion correction, a region located in the center of the image is selected as the analysis region. The analysis region can be all areas within the illumination range or a partial square region. Within this analysis region, the variance of the gray values ​​of all pixels is calculated. The gray value variance reflects the dispersion of the pixel gray values. Calm water surfaces have a concentrated distribution of pixel gray values ​​and a small variance due to uniform light reflection. Wavy water surfaces, on the other hand, have larger fluctuations in pixel gray values ​​and a larger variance due to light scattering and shadow formation.

[0039] Simultaneously, the Canny edge detection algorithm is applied to the analysis area to count the number of edge pixels detected within that area. This number is then divided by the total number of pixels in the analysis area to obtain the edge density. Calm water surfaces typically have fewer edges and lower edge density; wavy water surfaces, due to their complex wave and ripple textures, have more edges and higher edge density. In terms of type determination logic, the calculated grayscale variance and edge density are combined with a preset threshold to determine the water surface type. If the calculated grayscale variance is less than or equal to the preset calm water surface grayscale variance threshold, and the edge density is less than or equal to the preset edge density threshold... If the calculated grayscale variance is greater than the preset grayscale variance threshold for calm water surfaces, or if the edge density is greater than the preset edge density threshold, the system determines that the current water surface is a wave surface and triggers the subsequent control sub-strategy lighting mode. To avoid misjudgments caused by noise in a single frame image, brief water surface disturbances, or occasional sensor errors, the system introduces a three-frame confirmation mechanism. Only when three consecutive frames of images meet the same type of judgment condition will the system finally confirm the water surface type and switch the lighting control strategy accordingly.

[0040] Once the current water surface type is successfully determined, such as Figure 4 As shown, the searchlight output control module in the system makes decisions to control the output of the searchlights. Specifically, when the water surface is determined to be calm, since the water surface reflectivity distribution is relatively uniform and stable, no complex dynamic adjustments are needed. The searchlight output control module will trigger a theoretical sub-strategy, simply outputting the total power of the searchlights based on a preset theoretical model and evenly distributing it to each lamp core to ensure that the water surface reflectivity reaches the optimal threshold. Specifically, this is based on the real-time collected data on the current height of the searchlights. and current ambient light intensity The system queries a preset mapping table, which contains a one-to-one correspondence of power, height, and light intensity. This table, established through experimental calibration, records the total power value of the searchlight that enables the water surface reflectivity to reach the system's preset optimal threshold under different searchlight heights and ambient light intensities. If a match cannot be found in the mapping table... and For index entries with perfectly matching values, the searchlight output control module uses a multidimensional linear interpolation algorithm to calculate the theoretical total power. This method ensures a smooth transition and accurate estimation between discrete data points in the mapping table. Since calm water surfaces have uniform reflection characteristics, the searchlight output control module adopts a uniform distribution strategy to distribute the calculated theoretical total power evenly to all the lamp cores of the searchlight.

[0041] When the current water surface is determined to be a wave surface, such as Figure 2As shown, the water surface reflection characteristics are complex and dynamic. The searchlight output control module will trigger a sub-strategy to reliably identify targets through reflectivity analysis and multi-feature fusion. A pre-trained neural network model is used to achieve independent and adaptive power allocation for each lamp element of the searchlight. First, reflectivity analysis and reliable anomaly identification are performed. Traditional reflectivity analysis is easily affected by ambient light fluctuations and random wave reflections, leading to target misjudgment. This stage significantly improves target recognition accuracy through strategies such as environmental compensation, multi-feature fusion, and confidence screening. Based on the grayscale value and reflectivity conversion model calibrated during initialization, a real-time ambient light compensation coefficient is introduced to eliminate the potential impact of ambient light intensity fluctuations on reflectivity calculation. First, the current ambient light intensity is acquired in real-time through the photosensor built into the vision camera. At the same time, the preset standard ambient light intensity is retrieved during initialization. Calculate the ambient light compensation coefficient , If the actual ambient light intensity is higher than the standard value, then If the grayscale value is too bright, the reflectance needs to be reduced to correct the overly bright environment; conversely, it should be increased. For water surface images that have undergone preprocessing such as noise reduction and distortion correction, the reflectance is calculated pixel by pixel. ,in, For pixel coordinates, This is the grayscale value of the pixel. The grayscale values ​​and reflectance fundamental functions obtained during initialization using a standard grayscale plate are used to output the final environmentally compensated water surface reflectance distribution map. This distribution map ensures that the reflectance calculation results of the same physical object are consistent and comparable under different ambient lighting conditions.

[0042] To avoid misidentifying waves as targets by relying solely on the continuity of reflectance or a single threshold, this stage combines reflectance gradient, texture features, and shape features to determine anomalies. The Sobel operator is used to calculate horizontal and vertical gradients on the environmentally compensated water surface reflectance distribution map. The Sobel operator approximates the image gradient by calculating the gray-level difference between a pixel and its neighboring pixels, thus synthesizing the horizontal gradient. and vertical gradient Gradient magnitude: Gradient magnitude This reflects the steepness of the reflectivity change; if the gradient magnitude of a certain pixel... Greater than the preset gradient threshold If the reflectance of a pixel changes abruptly, it is considered to be a candidate outlier and is initially marked as such.

[0043] Next, texture feature-assisted screening is performed. For the local neighborhood of each candidate anomaly, its local binary pattern texture feature is calculated. This feature describes the relative grayscale relationship between the pixels in the neighborhood and the center pixel, and can effectively capture local texture patterns. The local texture features of water splashes are usually randomly distributed and have no fixed texture pattern. Their local texture histograms are relatively flat. However, the local texture features of rescue targets such as human clothing and lifebuoys are characterized by local texture consistency. For example, the fabric texture of clothing and the specific stripe texture of lifebuoys have specific peaks or patterns in their local texture histograms. The local texture feature histograms of candidate anomalies are matched with templates in the preset target texture template library. The target texture template library contains local texture feature texture samples of common rescue targets such as human bodies, lifebuoys, and life jackets. It is constructed through real data collection and annotation. If the matching degree is greater than or equal to the preset threshold, the candidate anomaly is retained and upgraded to a suspected anomaly.

[0044] Furthermore, a secondary screening based on shape features is performed. For all suspected anomalies, their bounding contours are obtained using a contour extraction algorithm. For each extracted contour, the area and aspect ratio (ratio of the longer side to the shorter side) of its minimum bounding rectangle are calculated. If the area of ​​the bounding rectangle is less than a preset area threshold... If the aspect ratio is greater than the preset aspect ratio threshold, the suspected abnormal point is identified as an interference point and removed. The remaining points are confirmed as the final abnormal points. These points have three characteristics: abrupt change in reflectivity, target texture pattern, and reasonable target shape.

[0045] After identifying the final outliers, the control strategy continues with outlier clustering and target confidence assessment. The final outliers are clustered using the Density-Based Clustering (DBSCAN) algorithm, replacing the traditional K-means algorithm. This is because DBSCAN does not require pre-setting the number of clusters and is better suited to complex scenarios with no target, a single target, or multiple targets. In the DBSCAN algorithm, a neighborhood radius is set. and minimum points After clustering, the system generates one or more anomalous blocks, each of which is considered a suspected target requiring rescue. The system also calculates the core parameters of each anomalous block, including its center coordinates (i.e., the coordinates of all pixels within the block). coordinates and Average of coordinates , indicating the center position of the suspected target in the pixel coordinate system; area size, i.e., the total number of pixels contained in the anomaly block. Mean reflectance, which is the average reflectance of all pixels within the abnormal block. Based on this, target confidence is calculated, and a confidence function is constructed based on the area of ​​the anomalous block, the mean reflectance, and the texture matching degree: ,in, The preset maximum target area, For the theoretically optimal target reflectivity, For texture matching, this function comprehensively evaluates the confidence level of the target by weighted combination of multiple features, retaining only the confidence score. Larger anomalous blocks are considered high-confidence targets, and subsequent searchlight control model calculations are performed only on these high-confidence targets, further reducing the probability of misidentifying non-target interference objects as targets.

[0046] Once a high-confidence target is identified, the searchlight output control module will filter and extract key input parameters for the searchlight control model from the real-time acquired data, including the current height of the searchlight. Current height of the water waves , coordinates of the center of the anomaly block and the current ambient light intensity The center coordinates of the high-confidence targets obtained in the outlier block clustering step in the pixel coordinate system are... Using camera internal and external parameters, as well as the current altitude of the searchlight. Converted to target azimuth deviation in searchlight coordinate system (Yaw angle deviation) and pitch angle deviation For example, if the target is 100 pixels to the right of the center of the image, this pixel deviation will be converted into a specific yaw angle deviation value relative to the direction of the drone's nose, and the pitch angle deviation will be calculated in the same way.

[0047] After the aforementioned parameters are extracted, the searchlight output control module inputs the standardized input parameters into a preset searchlight control model for power allocation calculation. The searchlight control model is a pre-trained backpropagation neural network model. The training objective of this model is to output power allocation parameters for each lamp core based on five key parameters: current searchlight height, current wave height, target azimuth deviation, pitch deviation, and current ambient light intensity. This ensures that the output power allocation can stably maintain the optimal target recognition reflectivity range for the water surface in high-confidence target areas. Its training data package... The data includes actual nighttime water scene test and simulation data, comprising n datasets. Each dataset includes the aforementioned five key parameters and the optimal output power of the lamp core. It consists of labeled data, covering scenes on calm water under different light intensities and scenes on wavy water under different light intensities. The labeled data can be achieved through actual measurement calibration or simulation optimization. Specifically, by conducting actual measurements using one dataset and manually adjusting or simulating fine-tuning the lamp core's output power to achieve the best visual image acquisition effect for the water surface reflectivity, the backpropagation neural network model's input layer contains five neurons, each corresponding to the standardized current height of the searchlight. Current height of the water waves Target azimuth deviation Pitch angle deviation and the current ambient light intensity The neural network model has at least one hidden layer, which uses the ReLU activation function to introduce non-linearity. The output layer contains neurons corresponding to the number of searchlight wicks. That is, the output data of the searchlight control model is the target power value for each wick output by each neuron. The model calculates the target power value for each wick through forward propagation. The neural network model is designed for the orientation of high-confidence targets in the searchlight coordinate system. , The model increases the power of the lamps covering the target area to enhance illumination, ensuring the target reflectivity reaches the optimal threshold and improving contrast. Simultaneously, for non-target areas, especially areas with detected strong reflections from waves, the model reduces the output power of the corresponding lamps to minimize water reflection interference and prevent image overexposure. Furthermore, the model also adjusts the power based on wave height. Adjusting the overall power compensation coefficient: The greater the wave height, the stronger the water surface scattering and absorption effect. The model will appropriately increase the overall power to offset the light attenuation caused by the wave height, ensuring effective illumination of the target area. After each frame image analysis is completed, the above steps of reflectivity analysis and reliable identification of anomalies, acquisition of input parameters for the searchlight control model, and model calculation and lamp core power output will be repeated to achieve real-time dynamic adjustment of the searchlight power. The adjustment frequency is consistent with the frame rate of the vision camera to ensure that the system can quickly respond to dynamic changes in the water surface and target movement.

[0048] Regardless of whether the water surface is determined to be calm or wavy, the real-time target tracking module must calculate and adjust the projection angle of the searchlight in real time based on the visual image to ensure that the center of the searchlight's aperture is always accurately aligned with the suspected target or area of ​​interest. In terms of target center positioning, when the water surface is calm, if the searchlight output control module does not identify any abnormal points, i.e., there is no target to be rescued in the current field of view, the system defaults to using the center point of the visual image as the temporary target center to ensure that the area illuminated by the searchlight remains in the center of the image. If the system identifies a high-confidence target, such as a lifebuoy or buoy floating on calm water, the center of the abnormal block of the high-confidence target in the pixel coordinate system is used as the target center.

[0049] When the water surface type is wavy, the system directly uses the center of the outlier block of the high-confidence target as the target center. In terms of angle deviation calculation, the real-time target tracking module uses the coordinates of the current target center point in the pixel coordinate system. Center coordinates of the visual image A precise comparison is performed to calculate the pixel deviation value. and Based on the visual camera's settings, including the intrinsic parameter matrix: focal length Main point and the current height of the searchlight The target azimuth adjustment amount required for the searchlight is calculated using precise geometric relationships. Pitch angle adjustment amount , ; In this context, pixel size refers to the actual physical size of a single pixel, and angle execution refers to the system sending angle adjustment commands to the servo motor of the searchlight pan-tilt unit, ensuring that the center of the searchlight's aperture can always be accurately and stably aligned with the target.

[0050] In addition, the image verification feedback module is configured to continuously monitor and verify the control effect of the searchlight, and trigger system re-evaluation and adjustment instructions based on the verification results to achieve adaptive optimization of the system strategy. Specifically, this includes image sharpness index calculation and verification strategies and optimization instructions. The image verification feedback module calculates the image sharpness index for the water surface image after the searchlight is controlled every specific number of frames. The image sharpness index includes edge intensity value, image entropy value, and contrast index. The edge intensity value uses the Tenengrad gradient operator to evaluate the edge richness and sharpness of the image. The higher the edge intensity value, the richer the edge information of the image and the higher the visual sharpness. The image entropy value reflects the richness of image information and the complexity of texture. The larger the entropy value, the more detailed information the image contains and the higher the sharpness. The contrast index is used to evaluate the brightness difference between the target and the background. A unified comprehensive image sharpness index is obtained through weighted averaging or a specific fusion algorithm.

[0051] Regarding verification strategies and optimization instructions, the image verification feedback module continuously compares the calculated comprehensive image sharpness index of the current frame with the preset minimum sharpness threshold during initialization. If the system detects that the comprehensive sharpness index calculated three times consecutively is lower than the minimum sharpness threshold, it triggers a system re-evaluation and adjustment instruction. This indicates that the current searchlight illumination strategy may no longer be suitable for the current environment or target state and needs optimization. Upon receiving the re-evaluation instruction, the system will initiate a comprehensive adaptive adjustment process. The water surface type analysis module will quickly re-determine the water surface type to confirm whether the environment has changed significantly. The searchlight output sub-control module will adjust the output based on the latest multi-level... Based on the source data and water surface type, reflectivity analysis, multi-feature fusion identification, anomaly clustering, and target confidence assessment are re-executed to ensure accurate identification of high-confidence targets. For wavy water surfaces, the searchlight output control module recalculates the power allocation parameters of each lamp core using a pre-trained neural network model. For calm water surfaces, the mapping table is re-queried to calculate the total power and distribute it evenly. The real-time target tracking module recalculates the projection angle adjustment of the searchlight based on the latest identified target position and drives the gimbal to adjust accordingly. This ensures that the searchlight control system can continuously optimize the lighting effect in various complex nighttime water environments, maximizing the accuracy and efficiency of target identification.

[0052] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A searchlight control system for an unmanned aerial vehicle (UAV), characterized in that: include: The multi-source data acquisition module acquires water surface images captured by the visual camera mounted on the UAV, and also acquires UAV flight data information. The water surface type analysis module determines the water surface type based on the grayscale variance and edge density of the water surface image. The water surface type includes calm water surface and wavy water surface. The searchlight output control module selects a lighting control strategy based on the water surface type. The lighting control strategy includes analyzing the reflectivity of the water surface image and identifying high-confidence targets through multi-feature fusion in conjunction with UAV flight data. Based on the location of the high-confidence targets and the collected parameters, the power allocation parameters of each lamp core are determined. The searchlight output execution module drives multiple lamp cores of the searchlight to provide illumination according to the power allocation parameters. The real-time target tracking module calculates the adjustment amount of the projection angle of the searchlight in real time based on the position of the high-confidence target or the center point of the water surface image so that the center point of the water surface image always coincides with the high-confidence target. The image verification feedback module calculates the image sharpness index of the water surface image at specific frame intervals, and triggers the system to re-evaluate and adjust the image sharpness index based on the image sharpness index.

2. The search and control system for an unmanned aerial vehicle (UAV) according to claim 1, characterized in that: The lighting control strategy includes a theoretical sub-strategy and a control sub-strategy. The theoretical sub-strategy is matched to a calm water surface and includes determining the total power of the searchlight according to a preset mapping table and distributing it evenly to each lamp core. The control sub-strategy is matched to a wavy water surface and includes determining the individual output power of each lamp core of the searchlight according to a pre-trained neural network model.

3. The search and control system for an unmanned aerial vehicle (UAV) according to claim 2, characterized in that: The theoretical sub-strategy includes a preset mapping table, which contains a one-to-one correspondence of total power, altitude, and ambient light intensity. The UAV flight data information includes the current altitude of the searchlight and the current ambient light intensity. The theoretical total power is indexed in the mapping table based on the current altitude of the searchlight and the current ambient light intensity. If there is no completely matching index entry in the mapping table, the theoretical total power is calculated by multidimensional linear interpolation.

4. The search and control system for an unmanned aerial vehicle (UAV) according to claim 3, characterized in that: The control sub-strategy includes converting the pixel coordinates of high-confidence targets into target azimuth and pitch deviations, analyzing the current height of water waves based on water surface images, and then inputting the current height of the searchlight, the current height of the water waves, the target azimuth and pitch deviations, and the current ambient light intensity into a pre-trained neural network model for forward propagation calculation, outputting the power allocation parameters for each lamp core.

5. A searchlight control system for an unmanned aerial vehicle according to any one of claims 1-4, characterized in that: The lighting control strategy includes a reflectance analysis step, which includes retrieving the preset standard ambient light intensity and calculating the ambient light compensation coefficient with the current ambient light intensity. Then, based on the ambient light compensation coefficient, the reflectance of the water surface image is calculated pixel by pixel to obtain the water surface reflectance distribution map after environmental compensation.

6. The search and control system for an unmanned aerial vehicle (UAV) according to claim 5, characterized in that: The lighting control strategy also includes a reflectivity gradient determination step, a texture feature-assisted filtering step, and a shape feature secondary filtering step. The reflectivity gradient determination step involves using the Sobel operator to calculate the gradients in the horizontal and vertical directions of the water surface reflectivity distribution map after environmental compensation, and synthesizing the gradient magnitude. If the gradient magnitude of a pixel is greater than a preset gradient threshold, it is initially marked as a candidate anomaly. The texture feature-assisted screening step calculates the local binary pattern texture features of the local neighborhood where the candidate anomaly point is located, and matches them with the templates in the preset target texture template library. If the matching degree is greater than or equal to the preset threshold, the candidate anomaly point is retained and upgraded to a suspected anomaly point. The secondary screening step of the shape features involves obtaining the circumscribed contour of the suspected abnormal point through a contour extraction algorithm, and calculating the area and aspect ratio of the minimum bounding rectangle of each contour. If the area is less than a preset area threshold, the suspected abnormal point is determined to be an interference point and is removed. The remaining points are confirmed as the final abnormal points.

7. The search and control system for an unmanned aerial vehicle (UAV) according to claim 6, characterized in that: The lighting control strategy also includes an anomaly block clustering step and a target confidence assessment step; The anomaly block clustering step involves setting a neighborhood radius and a minimum number of points for the final anomaly points to cluster spatially connected anomaly points into clusters, generating one or more anomaly blocks. Each anomaly block is considered a suspected target to be tracked. The target confidence assessment step involves extracting the center coordinates, area value, and mean reflectance of each anomalous block, constructing a confidence function based on the texture matching degree, area value, and mean reflectance to calculate the target confidence, and determining whether the anomalous block is a high-confidence target based on the target confidence.

8. The search and control system for an unmanned aerial vehicle (UAV) according to claim 7, characterized in that: The real-time target tracking module includes a searchlight tracking strategy, which includes comparing the center coordinates of a high-confidence target with the center coordinates of the visual image to calculate the pixel deviation value, and calculating the target azimuth adjustment amount and pitch adjustment amount of the searchlight based on the visual camera settings and the current height of the searchlight through geometric relationships.

9. The search and control system for an unmanned aerial vehicle (UAV) according to claim 1, characterized in that: The image verification feedback module includes a verification strategy, which includes performing frame processing on the water surface image after the detection and control, analyzing the image clarity index of the current frame, comparing the image clarity index with the minimum clarity threshold preset during initialization, and if the clarity index calculated three times in a row is lower than the minimum clarity threshold, the system is triggered to re-evaluate and adjust the strategy in each module.

10. The search and control system for an unmanned aerial vehicle (UAV) according to claim 2, characterized in that: The water surface type analysis module includes extracting an analysis region from the water surface image, calculating its gray-level variance, and obtaining the edge density by analyzing the number of edge pixels in the analysis region through edge detection. The gray-level variance and edge density are compared with preset gray-level variance thresholds and edge density thresholds, respectively. If the gray-level variance is less than or equal to the gray-level variance threshold and the edge density is less than or equal to the edge density threshold, the current water surface is determined to be a calm water surface; otherwise, the current water surface is determined to be a wavy water surface.

Images