An unmanned aerial vehicle inspection visual enhancement method and system for highways

Abstract

The application belongs to the technical field of computer vision, and provides a UAV inspection vision enhancement method and system for expressways, which comprises configuring multiple sensors for a UAV, and acquiring inspection images of each sensor; performing feature extraction on the inspection images of each sensor to obtain feature values of the inspection images of each sensor, thereby generating a suspected disease feature point distribution map of the inspection images; performing image fusion on all suspected disease feature point distribution maps of the inspection images to generate a fused image suspected disease feature point distribution map; establishing a suspected disease feature analysis model according to the fused image suspected disease feature point distribution map and environmental feature data of an inspection area, and generating a disease existence probability; and generating a UAV inspection vision enhancement image according to the disease existence probability; the application can intuitively present disease risk distribution, and significantly improves the detection accuracy, anti-interference capability and auxiliary decision effect of UAV inspection.

Description

Technical Field

[0001] This invention belongs to the field of computer vision technology, and in particular relates to a visual enhancement method and system for unmanned aerial vehicle (UAV) inspection of highways. Background Technology

[0002] With the rapid increase in the operational mileage of highways, traditional manual inspection methods face severe challenges in terms of efficiency, coverage, and safety. Drone inspection, with its mobility, flexibility, and comprehensive field of view, has become an important auxiliary means for detecting highway defects. However, in the complex highway inspection scenarios, single-sensor imaging is easily affected by environmental factors such as lighting and weather, has limited ability to identify subtle defects, and struggles to effectively distinguish between real defects and interference signals, thus hindering further improvements in inspection effectiveness.

[0003] Currently, drone-based highway inspections mainly rely on images from single sensors such as visible light or infrared, using image processing and machine learning for defect identification. Existing technologies largely focus on optimizing or classifying single images. Some studies attempt to fuse multi-sensor data, but the fusion methods are relatively simple, lacking refined collaborative anomaly analysis of the features of each sensor, and failing to consider environmental characteristics during inspections as a basis for judgment.

[0004] However, existing technologies have significant shortcomings: First, the depth of multi-sensor data fusion is insufficient, failing to quantitatively extract and fuse abnormal information from multi-dimensional features to generate detailed distribution maps of suspected diseases, resulting in a low detection rate of early minor diseases; second, the disease identification model does not integrate environmental feature data (such as temperature, humidity, and light) of the inspection area, making it impossible to assess the impact of the environment on the reliability of specific sensor detection, leading to a high misjudgment rate; third, the output results lack probabilistic and visualized disease risk representation, which is not conducive to the inspection personnel's hierarchical assessment and accurate decision-making for suspicious areas. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a visual enhancement method and system for unmanned aerial vehicle (UAV) inspection of highways, thus solving the aforementioned problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a visual enhancement method for unmanned aerial vehicle (UAV) inspection of highways, the method specifically comprising:

[0007] Equip the drone with multiple sensors and acquire inspection images from each sensor;

[0008] Feature extraction is performed on the inspection images of each sensor to obtain the feature values ​​of the inspection images of each sensor;

[0009] Based on the feature values ​​of the inspection images from each sensor, a distribution map of suspected disease feature points in the inspection images is generated;

[0010] All inspection images of suspected disease feature point distribution maps are fused to generate a fused image of suspected disease feature point distribution map;

[0011] Acquire environmental feature data of the inspection area, and establish a suspected disease feature analysis model based on the distribution map of suspected disease feature points in the fused image and the environmental feature data of the inspection area to generate the probability of disease existence.

[0012] Based on the probability of disease presence, visually enhanced images of drone inspections are generated; whereby, the visually enhanced images of drone inspections refer to inspection images that display the probability of disease presence.

[0013] Based on the above technical solutions, the present invention also provides the following optional technical solutions:

[0014] Further technical solution: The method for generating the distribution map of suspected disease feature points in the inspection image specifically includes:

[0015] Based on the feature values ​​of the inspection images from each sensor, a feature anomaly index is generated;

[0016] Based on the characteristic anomaly index, a distribution map of suspected disease feature points in the inspection images is generated.

[0017] Further technical solution: The specific method for generating the feature anomaly index includes:

[0018] Through the formula: ;

[0019] Generate Feature Anomaly Index ;

[0020] In the formula, the characteristic anomaly index This refers to the position in the inspection image of the k-th sensor. Characteristic abnormality index, This represents the position in the inspection image of the k-th sensor. The i-th eigenvalue, This represents the position in the inspection image of the k-th sensor. The reference value of the i-th feature, This represents the position in the inspection image of the k-th sensor. The difference threshold of the i-th feature This is a constant term used to prevent the denominator from being zero. This represents the position in the inspection image of the k-th sensor. The weight coefficient of the i-th feature, where n represents the position in the inspection image of the k-th sensor. The number of features.

[0021] Further technical solution: The method for generating the probability of disease existence specifically includes:

[0022] Obtain environmental characteristic data of the inspection area, and generate the credibility of suspected disease feature points based on the environmental characteristic data of the inspection area;

[0023] Obtain the distribution location of suspected disease feature points in the distribution map of suspected disease feature points in the fused image, and generate the basic probability of disease existence based on the distribution location of suspected disease feature points;

[0024] A suspected disease feature analysis model is established based on the basic probability of disease existence and the credibility of suspected disease feature points to generate the probability of disease existence.

[0025] Further technical solution: The method for generating the credibility of the suspected disease feature points specifically includes:

[0026] Through the formula: ;

[0027] Credibility of generated suspected disease feature points ;

[0028] In the formula, the confidence level of suspected disease feature points This refers to the location in the fused image. The credibility of suspected disease feature points This represents the j-th environmental characteristic data of the inspection area. This represents the near endpoint of the j-th environmental feature reference range in the inspection area. This represents the difference threshold of the j-th environmental feature in the inspection area. This indicates the location in the fused image. The weighting coefficients for suspected disease feature points, where m represents the number of environmental feature data points in the inspection area. This represents the sensitivity coefficient to deviation of the ring feature.

[0029] Further technical solution: Confidence of the suspected disease feature points The calculation formula is constrained by the following condition: the j-th environmental feature data of the inspection area. The j-th environmental feature is not within the reference range of the inspection area; if the j-th environmental feature data of the inspection area... Within the j-th environmental feature reference range of the inspection area, the corresponding The value of the item is 1.

[0030] Further technical solutions: The specific methods for generating the basic probability of the disease's existence include:

[0031] Through the formula: ;

[0032] There is a basic probability of disease generation. ;

[0033] In the formula, the disease has a basic probability. This refers to the location in the fused image. The disease has a basic probability. This indicates the location in the fused image. The total number of suspected disease feature points within a radius of R. This represents a reference value for the number of suspected disease feature points within a radius of R. This represents the growth rate coefficient, which is used to control the rate at which the basic probability of disease presence increases with the number of feature points.

[0034] Further technical solution: The expression of the suspected disease feature analysis model is specifically as follows: ;

[0035] In the expression, This indicates the location in the fused image. The probability of the disease existing. This indicates the location in the fused image. The disease has a basic probability. This indicates the location in the fused image. The total number of suspected disease feature points within a radius of R. This represents the position in the q-th fused image. The reliability of suspected disease feature points within a radius of R.

[0036] A visual enhancement system for unmanned aerial vehicle (UAV) inspection of highways, the system being used to execute the aforementioned visual enhancement method for UAV inspection of highways, specifically including:

[0037] The image acquisition unit is used to equip the drone with multiple sensors and acquire inspection images from each sensor.

[0038] The feature extraction unit is used to extract features from the inspection images of each sensor to obtain the feature values ​​of the inspection images of each sensor.

[0039] The image analysis unit is used to generate a distribution map of suspected disease feature points in the inspection images based on the feature values ​​of the inspection images from each sensor.

[0040] The image fusion unit is used to fuse the distribution maps of suspected disease feature points in all inspection images to generate a fused image distribution map of suspected disease feature points.

[0041] The disease existence probability analysis unit is used to acquire environmental feature data of the inspection area, establish a suspected disease feature analysis model based on the distribution map of suspected disease feature points in the fused image and the environmental feature data of the inspection area, and generate the disease existence probability.

[0042] The visual enhancement unit is used to generate visually enhanced images for UAV inspection based on the probability of disease presence; wherein, the visually enhanced images for UAV inspection refer to inspection images that show the probability of disease presence.

[0043] Further technical solution: The image analysis unit specifically includes:

[0044] The anomaly analysis module is used to generate a characteristic anomaly index based on the feature values ​​of the inspection images of each sensor.

[0045] The distribution map output module is used to generate a distribution map of suspected disease feature points in the inspection image based on the feature anomaly index.

[0046] The disease existence probability analysis unit specifically includes:

[0047] The credibility analysis module is used to acquire environmental characteristic data of the inspection area and generate credibility of suspected disease feature points based on the environmental characteristic data of the inspection area.

[0048] The basic probability analysis module is used to obtain the distribution location of suspected disease feature points in the distribution map of suspected disease feature points in the fused image, and generate the basic probability of disease existence based on the distribution location of suspected disease feature points;

[0049] The final probability output module is used to establish a suspected disease feature analysis model based on the basic probability of disease existence and the credibility of suspected disease feature points, and generate the probability of disease existence.

[0050] This invention provides a visual enhancement method and system for unmanned aerial vehicle (UAV) inspection of highways, which has the following advantages compared with the prior art:

[0051] This invention achieves collaborative and refined detection of suspected disease points by calculating the anomaly index of multi-sensor features and fusing distribution maps. By introducing environmental features of the inspection process to assess the credibility of each suspected point and combining its spatial density to generate the probability of disease presence, the interpretability and environmental adaptability of the judgment results are improved. Finally, a probabilistic visually enhanced image is generated and output, which intuitively presents the disease risk distribution, significantly improving the detection accuracy, anti-interference ability and decision support effect of UAV inspection. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating a visual enhancement method for unmanned aerial vehicle (UAV) inspection of highways provided by the present invention.

[0053] Figure 2 This is a flowchart illustrating step S30 of the present invention.

[0054] Figure 3 This is a flowchart illustrating step S50 of the present invention.

[0055] Figure 4 This is a schematic diagram of the structure of a visual enhancement system for unmanned aerial vehicle (UAV) inspection of highways provided by the present invention.

[0056] Figure 5 This is a schematic diagram of the image analysis unit provided by the present invention.

[0057] Figure 6 This is a schematic diagram of the structure of the disease existence probability analysis unit provided by the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0059] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0060] Please see Figure 1 The present invention provides a visual enhancement method for unmanned aerial vehicle (UAV) inspection of highways, comprising the following steps:

[0061] Step S10: Configure the drone with multiple sensors and acquire inspection images from each sensor;

[0062] Step S20: Extract features from the inspection images of each sensor to obtain the feature values ​​of the inspection images of each sensor;

[0063] Step S30: Generate a distribution map of suspected disease feature points in the inspection images based on the feature values ​​of each sensor's inspection images;

[0064] Step S40: Perform image fusion on the distribution maps of suspected disease feature points in all inspection images to generate a fused image distribution map of suspected disease feature points;

[0065] Step S50: Obtain environmental feature data of the inspection area, establish a suspected disease feature analysis model based on the distribution map of suspected disease feature points in the fused image and the environmental feature data of the inspection area, and generate the probability of disease existence;

[0066] Step S60: Generate a visually enhanced image for UAV inspection based on the probability of disease presence; wherein, the visually enhanced image for UAV inspection refers to an inspection image that displays the probability of disease presence.

[0067] Among them, the UAV inspection visual enhancement method refers to a technical solution that uses UAVs equipped with multiple sensors to inspect highways, and through in-depth analysis and fusion of multi-source image data and environmental data, finally generates visually enhanced images that can intuitively show the probability of the existence of defects; the method aims to improve the accuracy, reliability and visualization level of highway defect detection.

[0068] Sensors refer to various detection devices carried on drones to acquire information about highway surfaces and the surrounding environment; these sensors may include, but are not limited to, visible light cameras, infrared thermal imagers, lidar, multispectral cameras, etc. They can capture image data from different dimensions to cope with complex and ever-changing environmental conditions and different types of disease characteristics.

[0069] Inspection images refer to raw image data collected in real time by various sensors on drones during the inspection process; these images carry visual information about the road surface condition, structural features, and potential defects of highways.

[0070] Feature extraction refers to the process of identifying and quantifying representative information from raw inspection images; these features can be texture, color, shape, edge, temperature distribution, etc., and they can reflect potential abnormal areas or signs of disease in the image;

[0071] The distribution map of suspected defects in inspection images refers to a graphic representation of the distribution of areas or points in an image that may contain defects, based on the feature value analysis results of a single sensor image; this distribution map can preliminarily indicate potential abnormal areas.

[0072] Image fusion refers to the process of integrating image information from different sensors or different processing stages to generate a single image containing more comprehensive and richer information. Image fusion can compensate for the deficiencies of information from a single sensor, thereby improving the ability to identify diseases.

[0073] The fused image suspected disease feature point distribution map refers to the comprehensive distribution map generated by integrating the suspected disease feature point distribution maps of inspection images from all single sensors; this map gathers information from multiple sources and provides a more comprehensive view of suspected disease areas;

[0074] Environmental characteristic data of the inspection area refers to environmental parameter information related to the inspection area during the drone inspection process; these data may include, but are not limited to, temperature, humidity, light intensity, wind speed, rainfall, etc., which have an impact on sensor imaging quality and the manifestation characteristics of defects;

[0075] The visually enhanced image of drone inspection refers to the final image generated by overlaying or otherwise displaying the probability information of the presence of diseases on the basis of the original inspection image; the image helps inspection personnel to quickly locate and understand the disease risk area through intuitive visual cues.

[0076] Specifically, step S10, configuring the UAV with multiple sensors and acquiring inspection images from each sensor, is a fundamental step in this method. To comprehensively capture highway damage information, the UAV can be equipped with various types of sensors. For example, a high-resolution visible light camera can be configured to acquire images of macroscopic damage such as road surface cracks and potholes, while an infrared thermal imager can be configured to detect temperature anomalies such as cavities and delamination beneath the road surface. Furthermore, a multispectral camera can be configured to acquire spectral information in different bands to identify material anomalies or early-stage damage. In practice, these sensors can work independently or collaboratively, synchronously acquiring their respective image data during UAV flight. The acquired image data can be a sequence of static images or a continuous video stream, which is then transmitted to the processing unit for further analysis.

[0077] In step S20, feature extraction is performed on the inspection images of each sensor to obtain the feature values ​​of each sensor's inspection images, which is a crucial step in identifying potential defects. For visible light images, edge detection algorithms (such as the Canny operator and the Sobel operator) can be used to extract the edge features of cracks, or texture analysis methods (such as gray-level co-occurrence matrix and local binary mode) can be used to quantify the anomalies in road surface texture. For infrared thermal images, features such as temperature gradients, the area and intensity of hot spot regions can be extracted. For multispectral images, spectral features such as the reflectance ratio of different bands or vegetation indices can be calculated. These feature extraction methods aim to transform the raw image data into quantitative feature values ​​that can reflect defect information, facilitating subsequent automated analysis.

[0078] In step S30, a distribution map of suspected disease feature points in the inspection images is generated based on the feature values ​​of each sensor, aiming to initially locate potential disease areas. After obtaining the feature values ​​of each sensor, abnormal areas can be identified by setting thresholds. For example, for visible light images, if the edge density or texture roughness of a certain area exceeds a preset threshold, that area is marked as a suspected disease feature point. For infrared thermal images, if the temperature of a certain area is significantly higher or lower than the surrounding environment, that area is also marked. These marked suspected disease feature points can be represented in the form of a binary image, where white pixels represent suspected disease points and black pixels represent normal areas, thus forming a distribution map of suspected disease feature points for that sensor.

[0079] In step S40, all the suspected disease feature point distribution maps from the inspection images are fused to generate a fused image suspected disease feature point distribution map. This aims to integrate multi-source information and improve the comprehensiveness of disease identification. After acquiring multiple suspected disease feature point distribution maps from different sensors, a logical "OR" operation can be used for fusion. For example, if any sensor's distribution map marks a suspected disease point at a certain location, the fused distribution map will also mark it as a suspected disease point at that location. Alternatively, a weighted average method can be used, assigning different weights based on the reliability of different sensors or their sensitivity to specific diseases, and superimposing multiple distribution maps to generate a comprehensive fused image suspected disease feature point distribution map. This fused map can gather abnormal information detected by different sensors in different dimensions.

[0080] In step S50, environmental feature data of the inspection area is acquired. A suspected disease feature analysis model is established based on the distribution map of suspected disease feature points in the fused image and the environmental feature data of the inspection area to generate the probability of disease presence. This is one of the core innovations of this method. Environmental feature data can be obtained through environmental sensors (such as thermometers, hygrometers, and light sensors) mounted on the UAV, or through data from external weather stations. When establishing the suspected disease feature analysis model, a rule-based expert system can be used. For example, if the fused image shows a suspected crack in a certain area, and environmental data shows that the area has recently experienced drastic temperature changes, the probability that the crack is a real disease can be increased according to preset rules. Alternatively, a machine learning model, such as a support vector machine or neural network, can be used, taking the feature point distribution information of the fused image and the environmental feature data as input, to train the model and output the probability of disease presence. This model, by comprehensively considering image anomalies and environmental influences, provides a more accurate quantitative assessment of diseases.

[0081] In step S60, a visually enhanced image for drone inspection is generated based on the probability of disease presence. This enhanced image, which displays the probability of disease presence, is the final output of this method. After calculating the probability of disease presence in each suspected area, this probability information can be visually overlaid onto the original inspection image. For example, a heatmap with different colors or transparency can be used on the original image to represent the probability of disease presence; the higher the probability, the darker or more vibrant the color. Alternatively, a border can be drawn around the suspected disease area, with specific probability values ​​labeled next to the border. This visually enhanced image can intuitively show inspectors the potential risks of disease, helping them quickly identify high-risk areas and conduct targeted further inspections.

[0082] For preferred options, please refer to [link / reference]. Figure 2 The present invention further proposes a method for generating the distribution map of suspected disease feature points in the inspection image, specifically including:

[0083] Step S31: Generate a feature anomaly index based on the feature values ​​of the inspection images of each sensor;

[0084] Step S32: Generate a distribution map of suspected disease feature points in the inspection image based on the feature anomaly index;

[0085] In step S31, the generation of the feature anomaly index aims to quantify the degree of deviation of features at each location in the inspected image from the normal or reference state. This index transforms the original, multi-dimensional feature values ​​into a unified numerical value that can intuitively reflect the degree of anomaly. For example, statistical methods can be used to generate the anomaly index by calculating the deviation of each feature value from a preset normal range (such as the mean plus or minus the standard deviation); alternatively, anomaly detection algorithms in machine learning, such as One-Class Support Vector Machine (One-Class SVM) or Isolation Forest, can be used to train and evaluate the feature values, thereby outputting the anomaly score for each location as the feature anomaly index.

[0086] In step S32, the purpose of generating a distribution map of suspected disease feature points in the inspection image is to spatially map and visualize the quantified anomaly information from step S31, thereby clearly indicating which regions or points in the image are considered potential disease feature points. For example, a threshold for a feature anomaly index can be set, and all pixels or regions with an anomaly index exceeding the threshold can be marked as suspected disease feature points and highlighted on the image with a specific color, brightness, or symbol to form an intuitive distribution map; alternatively, clustering algorithms or morphological operations in image processing can be used to group and refine continuous regions with high anomaly indices to identify more accurate and representative suspected disease feature point regions.

[0087] The proposed solution introduces a feature anomaly index as an intermediate step, first transforming the original inspection image feature values ​​into quantified anomaly indicators. This transformation allows for the identification of disease features to move beyond simply relying on the absolute magnitude or comparison of feature values, instead basing it on the degree of deviation from the normal state. Subsequently, a distribution map of suspected disease feature points is generated based on these quantified anomaly indices, effectively filtering out normal variations and noise interference in the image and highlighting truly abnormal areas. This two-stage processing approach makes it possible to extract meaningful disease indication information from complex and diverse sensor data, thus providing more accurate and reliable input for subsequent disease analysis.

[0088] By employing the aforementioned technical solution, the feature values ​​of the inspection images are transformed into feature anomaly indices. Then, a distribution map of suspected disease feature points is generated based on these indices, enabling more refined and accurate identification of potential disease areas within the images. This method effectively avoids potential misjudgments or omissions that might result from directly judging diseases based on raw feature values. It improves the sensitivity and specificity of identifying suspected disease feature points, thus laying a more solid foundation for subsequent disease probability analysis. This allows the entire UAV inspection visual enhancement method to provide more reliable disease early warning information.

[0089] Preferably, the present invention further proposes a method for generating the feature anomaly index, specifically including:

[0090] Through the formula: ;

[0091] Generate Feature Anomaly Index ;

[0092] In the formula, the characteristic anomaly index This refers to the position in the inspection image of the k-th sensor. Characteristic abnormality index, This represents the position in the inspection image of the k-th sensor. The i-th eigenvalue, This represents the position in the inspection image of the k-th sensor. The reference value of the i-th feature, This represents the position in the inspection image of the k-th sensor. The difference threshold of the i-th feature This is a constant term used to prevent the denominator from being zero. This represents the position in the inspection image of the k-th sensor. The weight coefficient of the i-th feature, where n represents the position in the inspection image of the k-th sensor. The number of features;

[0093] The feature anomaly index quantifies the degree to which features extracted at a specific location (x, y) in the inspection image of the k-th sensor deviate from the normal or reference state. A higher value indicates a greater degree of feature anomaly at that location, thus suggesting a higher probability of disease presence in that area. The generation of this index provides a quantitative basis for subsequent identification of suspected disease feature points.

[0094] The i-th feature value at position (x, y) in the inspection image acquired by the k-th sensor is a quantized representation of the i-th specific feature extracted at position (x, y) in the inspection image acquired by the k-th sensor. For example, for visible light images, this feature value can represent pixel brightness, color components (such as R, G, B values), texture features (such as the energy and contrast of the gray-level co-occurrence matrix), edge density, etc.; for thermal infrared images, this feature value can represent the temperature value or temperature gradient at that location; for lidar point cloud data, this feature value can represent geometric features such as height, slope, or roughness. These feature values ​​are extracted from the original image data using image processing algorithms or deep learning models.

[0095] The reference value of the i-th feature at position (x, y) in the inspection image of the k-th sensor represents the expected or baseline value of the i-th feature at position (x, y) under normal, defect-free conditions. This reference value can be obtained in various ways. For example, it can be calculated based on statistical data of historical healthy road sections, taking the average or median; it can also be obtained by sampling and statistically analyzing the features of known healthy areas in the current inspection area; or, an empirical threshold conforming to the normal state of a highway can be preset.

[0096] The difference threshold of the i-th feature at position (x, y) in the inspection image of the k-th sensor is used to measure the allowable fluctuation range or degree of variation of the i-th feature at position (x, y) under normal conditions. Its function is to normalize the deviation of the feature value, enabling effective comparison of features with different dimensions or fluctuation ranges. This threshold can be obtained based on historical data or statistical analysis; for example, it can be the standard deviation of the feature value, the interquartile range, or a reasonable tolerance range set based on expert experience.

[0097] The constant term is a very small positive number. Its main function is to ensure that the denominator in the formula does not have a zero value, thereby avoiding calculation errors or numerical instability.

[0098] The weight coefficient of the i-th feature at position (x, y) in the inspection image of the k-th sensor reflects the relative importance or contribution of the i-th feature in judging the damage. Different features may have different indicative effects on different types of damage; for example, color features may be more sensitive to oil stains on the road surface, while texture features may be more sensitive to cracks. This weight coefficient can be determined through expert experience, machine learning methods (such as feature selection algorithms), or statistical analysis based on historical damage data to highlight features that are more critical for damage identification.

[0099] This application's scheme extracts features from inspection images from each sensor and calculates the absolute deviation between each feature value at each location and a preset reference value. To eliminate the influence of different feature dimensions and fluctuation ranges, this absolute deviation is normalized by a difference threshold (with a constant term added to avoid the denominator being zero), thus obtaining the standardized deviation degree of each feature. Subsequently, these standardized deviation degrees are multiplied by their respective weighting coefficients to reflect the relative importance of different features in disease identification. Finally, all weighted standardized deviation degrees are summed to generate a feature anomaly index for that location. This index comprehensively reflects the anomaly degree of all relevant features at that location. In this way, even when faced with multiple features from different sensors with different physical meanings and numerical ranges, they can be uniformly quantified into a comparable anomaly index, thus providing an accurate and robust quantitative basis for subsequently generating a distribution map of suspected disease feature points in the inspection images. This method effectively solves the problem of insufficient accuracy and robustness that may result from directly comparing raw feature values, making the identification of potential disease areas more accurate.

[0100] Through the above technical solution, this application provides a method for quantifying and comprehensively evaluating the degree of multi-source feature anomalies. This method normalizes the deviation of each feature and introduces weighting coefficients to reflect the indicative importance of different features to road surface defects. This allows features from different sensors, with different physical meanings and numerical ranges, to be uniformly converted into comparable feature anomaly indices. This effectively solves the problem of insufficient defect identification accuracy caused by inconsistent feature dimensions and sensitivities in traditional methods, making early and accurate identification of road surface defects on highways possible. By generating accurate feature anomaly indices, suspected defect areas can be located and identified more accurately, significantly improving the generation quality and reliability of suspected defect feature point distribution maps in inspection images. This lays a solid foundation for subsequent defect probability analysis and the generation of visually enhanced images from UAV inspections, improving the intelligence level and efficiency of highway inspections.

[0101] For preferred options, please refer to [link / reference]. Figure 3 The present invention further proposes a method for generating the probability of the existence of the aforementioned disease, specifically including:

[0102] Step S51: Obtain environmental characteristic data of the inspection area, and generate the credibility of suspected disease feature points based on the environmental characteristic data of the inspection area;

[0103] Step S52: Obtain the distribution location of suspected disease feature points in the distribution map of suspected disease feature points in the fused image, and generate the basic probability of disease existence based on the distribution location of suspected disease feature points;

[0104] Step S53: Based on the basic probability of disease existence and the credibility of suspected disease feature points, establish a suspected disease feature analysis model to generate the probability of disease existence;

[0105] Acquiring environmental characteristic data of the inspection area refers to collecting external environmental information closely related to the condition of the highway pavement, structural integrity, and the occurrence and development of defects. This data may include, but is not limited to, meteorological data (such as temperature, humidity, rainfall, and solar radiation intensity), geographical data (such as altitude, slope, and soil type), traffic data (such as traffic flow and the proportion of heavy-duty vehicles), and pavement material characteristics. Environmental characteristic data can be obtained through various means, including environmental sensors mounted on drones, ground weather stations, traffic monitoring systems, geographic information system (GIS) databases, or historical records.

[0106] The credibility of generating suspected disease feature points refers to assessing, based on acquired environmental feature data, the likelihood or reliability that a suspected disease feature point detected in an image is indeed a disease under current environmental conditions. For example, in a humid and rainy environment, the propagation rate of road cracks may accelerate; in this case, crack feature points detected in the image should be assigned a higher credibility.

[0107] Obtaining the distribution location of suspected disease feature points in the fused image distribution map refers to accurately identifying and extracting the spatial coordinate information of each suspected disease point from the distribution map obtained after multi-sensor image fusion processing. This location information can be pixel coordinates, geographic coordinates, or relative positions with respect to a reference point. This process can be achieved through image processing techniques, such as using threshold segmentation, connected component analysis, or object detection algorithms to locate high-response regions or marker points in the distribution map and record the positions of their center points or bounding boxes.

[0108] The generation of the baseline probability of disease presence refers to the preliminary assessment of the likelihood of disease presence in a region based solely on the spatial distribution density, clustering degree, or morphological characteristics of suspected disease feature points in the fused image, without considering environmental factors. For example, in a local area, if the number of suspected disease feature points is dense and exhibits a specific pattern (such as linear cracks or reticular cracks), the baseline probability of disease presence in that area will increase accordingly. The baseline probability can be generated using kernel density estimation, calculating the number of suspected disease points within a certain radius around each pixel and normalizing them to a probability value between 0 and 1; alternatively, it can be determined by consulting a pre-defined disease density-probability lookup table based on the type and local density of suspected disease points.

[0109] Establishing a suspected disease feature analysis model refers to constructing a mathematical or logical framework to comprehensively consider the basic probability of disease presence and the credibility of suspected disease feature points, in order to obtain a more comprehensive and accurate probability of disease presence. This model aims to integrate dual evidence from image spatial distribution information and environmental context information.

[0110] The proposed solution refines the generation method of the disease existence probability. First, in step S51, the system acquires environmental feature data of the inspection area and generates the credibility of suspected disease feature points based on this data. This process aims to introduce an assessment of the impact of external environmental factors on the authenticity of the disease. For example, under specific climatic or geographical conditions, certain image features may be more likely to represent the actual disease, thus providing important contextual information for subsequent disease judgment. Simultaneously, in step S52, the system extracts the spatial distribution information of suspected disease feature points from the distribution map of suspected disease feature points in the fused image and generates the basic probability of disease existence accordingly. This step focuses on initially judging the likelihood of disease based purely on the clustering degree and spatial pattern of image visual features, reflecting the intuitive representation of disease in the image. Subsequently, in step S53, the credibility of the suspected disease feature points generated in step S51 is combined with the basic probability of disease existence generated in step S52 to establish a suspected disease feature analysis model. This model comprehensively assesses the likelihood of disease presence by organically integrating corrective information from the environmental context and raw evidence from the spatial distribution of the image, thereby generating a more accurate and reliable disease presence probability. In this way, the proposed solution overcomes the assessment bias that may result from a single information source, making the calculation of disease presence probability more comprehensive and precise, and providing a solid foundation for subsequent visual enhancement.

[0111] Through the above technical solution, this application can effectively integrate environmental feature data of the inspection area with the spatial distribution information of suspected disease feature points in the fused image. By independently assessing the impact of environmental factors on the authenticity of suspected diseases (credibility of suspected disease feature points) and the probability of diseases based purely on the spatial distribution of image features (basic probability of disease existence), and establishing a suspected disease feature analysis model on this basis, the probability of disease existence can be calculated more comprehensively and precisely. This fusion of multi-dimensional information significantly improves the accuracy and reliability of disease existence probability assessment, avoiding misjudgments or omissions that may be caused by a single information source. Therefore, the generated UAV inspection visual enhancement images can more accurately highlight real disease areas, effectively guide the attention of inspection personnel, and improve the efficiency and quality of highway inspections.

[0112] Preferably, the present invention further proposes a method for generating the credibility of the suspected disease feature points, specifically including:

[0113] Through the formula: ;

[0114] Credibility of generated suspected disease feature points ;

[0115] In the formula, the confidence level of suspected disease feature points This refers to the location in the fused image. The credibility of suspected disease feature points This represents the j-th environmental characteristic data of the inspection area. This represents the near endpoint of the j-th environmental feature reference range in the inspection area. This represents the difference threshold of the j-th environmental feature in the inspection area. This indicates the location in the fused image. The weighting coefficients for suspected disease feature points, where m represents the number of environmental feature data points in the inspection area. This represents the sensitivity coefficient for ring feature deviation;

[0116] The formula for calculating the reliability of suspected disease feature points comprehensively considers multiple environmental feature data and their deviation from the reference range, assigning different weights to different environmental features to calculate the reliability of suspected disease feature points at a specific location (x, y). Its purpose is to mathematically model the influence of environmental factors on disease judgment, making the reliability assessment more objective and precise.

[0117] The j-th environmental feature data of the inspection area refers to the quantitative information of environmental factors of the UAV in the highway inspection area. This data is crucial for assessing the authenticity of suspected defect feature points, because certain environmental conditions (such as extreme temperature, humidity, light, wind speed, road surface temperature, rainfall, etc.) may cause changes in image features, thereby affecting the accuracy of defect identification.

[0118] The near endpoint of the j-th environmental feature reference range in the inspection area is a boundary value used to define normal or ideal environmental conditions. When environmental feature data deviates from this reference range, its impact on the reliability of suspected defect feature points gradually increases. This near endpoint can be a preset empirical value or the boundary of a normal fluctuation range obtained from historical data statistics. This near endpoint can be set through expert experience. For example, for road surface temperature, a lower or upper limit of the normal operating temperature range can be set as the near endpoint. If the road surface temperature is below the lower limit of the normal operating temperature range, the near endpoint is the lower limit of the normal operating temperature range; conversely, if the road surface temperature is above the upper limit of the normal operating temperature range, the near endpoint is the upper limit of the normal operating temperature range.

[0119] The difference threshold for the j-th environmental feature in the inspection area is used to measure the sensitivity of the reliability impact when the environmental feature data deviates from the reference range. A smaller difference threshold means that even a slight deviation from the reference range will have a significant impact on reliability; a larger difference threshold means that a greater deviation is required for the environmental feature data to significantly affect reliability.

[0120] The weighting coefficients of suspected disease feature points at location (x, y) in the fused image reflect the relative importance of different environmental features to the reliability of these points. For example, for cracks identified by infrared sensors, temperature changes may be more important than humidity changes, thus the weighting coefficient for temperature would be higher. Furthermore, the weighting coefficients also consider the sensitivity of sensor type to environmental features, ensuring the accuracy of the assessment. These weighting coefficients can be set using expert knowledge and experience; for example, by assigning appropriate weights to each environmental feature based on the characteristics of different sensors (visible light, infrared, lidar, etc.) and their performance under different environmental conditions.

[0121] The environmental feature deviation sensitivity coefficient is a global parameter used to adjust the overall sensitivity of reliability when all environmental features deviate from the reference range. It can act as a moderating factor to uniformly control the stringency of environmental factors in reliability assessment. This coefficient can be determined through system debugging and optimization; for example, in the initial stages of system deployment, the value of the environmental feature deviation sensitivity coefficient can be adjusted through small-scale testing and manual evaluation to achieve the best disease identification results.

[0122] The proposed solution for generating the credibility of suspected disease feature points introduces an exponential decay model based on the degree of deviation of environmental feature data, and combines the sensitivity of different sensor types to environmental features to determine weighting coefficients, thereby achieving refined quantification of the credibility of suspected disease feature points. Specifically, for each environmental feature data in the inspection area, its absolute deviation from the near endpoint of a preset reference range is first calculated and normalized with a difference threshold. Subsequently, an exponential function is used to quantify the impact of the degree of environmental feature deviation on credibility, where the environmental feature deviation sensitivity coefficient is used to adjust the overall sensitivity of this impact. The greater the deviation of the environmental feature data from the reference range, the smaller the value of this term, indicating a lower credibility. Next, this exponential term is multiplied by a weighting coefficient determined according to the sensor type's sensitivity to environmental features to reflect the relative importance of different environmental features and their interaction with sensors to credibility. Finally, the weighted impacts of all m environmental features are summed to obtain the credibility of the suspected disease feature point at position (x, y) in the fused image. This method transforms the assessment of the credibility of suspected disease feature points from a simple binary judgment or a rough linear relationship into a dynamic and quantitative reflection of the comprehensive impact of environmental factors on disease assessment. The credibility of suspected disease feature points generated in this way can more accurately correct the baseline probability of disease presence obtained from the distribution map of suspected disease feature points in the fused image, thus making the final generated probability of disease presence closer to reality. This effectively solves the problem that it is difficult to accurately assess the reliability of suspected disease feature points based solely on environmental feature data.

[0123] Through the above technical solution, this application provides a more accurate and quantitative mechanism for assessing the credibility of suspected defect feature points. By comprehensively considering multiple environmental feature data, their deviation from the reference range, and the sensitivity of different sensor types to environmental features, the system can dynamically adjust the credibility of suspected defect feature points. This avoids the one-sidedness or inaccuracy of reliability assessment of suspected defect feature points when environmental conditions are complex and changeable. Ultimately, this refined credibility assessment can effectively correct the basic probability of defect existence, making the final generated probability of defect existence more accurate and reliable. This significantly improves the accuracy and robustness of defect identification in highway UAV inspections, reduces the risk of false alarms and missed alarms, and thus improves inspection efficiency and decision-making quality.

[0124] Preferably, the present invention further proposes the reliability of the suspected disease feature points. The calculation formula is constrained by the following condition: the j-th environmental feature data of the inspection area. The j-th environmental feature is not within the reference range of the inspection area; if the j-th environmental feature data of the inspection area... Within the reference range of the j-th environmental feature in the inspection area The value of the item is 1;

[0125] In this context, "the j-th environmental characteristic data of the inspection area is outside the reference range of the j-th environmental characteristic of the inspection area" refers to a situation where, during drone inspections of highways, the measured value of a specific environmental parameter (such as temperature, humidity, light intensity, wind speed, road surface condition, etc.) exceeds a pre-set or empirically determined normal, healthy, or acceptable value range. For example, if the reference range for road surface temperature is 20-30 degrees Celsius, but the actual measured value is 35 degrees Celsius, then this environmental characteristic data is considered outside the reference range.

[0126] The j-th environmental characteristic data within the inspection area falling within the j-th environmental characteristic reference range of the inspection area means that, during drone inspection of highways, the measured value of a specific environmental parameter falls within a pre-set or empirically determined normal, healthy, or acceptable value range. For example, if the road surface humidity reference range is 40%-60%, and the actual measured value is 45%, then this environmental characteristic data is considered to be within the reference range. This judgment can also be achieved by directly comparing the measured value with the upper and lower limits of the reference range, or by using a preset threshold. Correspondingly... A value of 1 for this item indicates that when the environmental feature data is within the normal range, the impact of this environmental feature on the credibility of suspected disease feature points is set to the maximum value, meaning the credibility is not reduced. This can be achieved by directly forcing the calculation result of this item to 1, and then using this value in the final summation.

[0127] The proposed solution introduces a mechanism to determine whether environmental feature data falls within a reference range when generating the credibility of suspected disease feature points. Specifically, when the j-th environmental feature data of the inspection area is outside its preset reference range, it indicates that the environmental factor may have an adverse impact on the occurrence or development of the disease. In this case, the contribution of this environmental feature to the credibility of the suspected disease feature point is calculated based on its deviation from the reference range; the greater the deviation, the smaller the value of this contribution, thus reducing the credibility of the suspected disease feature point and reflecting the negative impact of adverse environmental conditions on disease judgment. Conversely, when the j-th environmental feature data of the inspection area is within its preset reference range, it indicates that the environmental factor is in a normal state and has no significant negative impact on the occurrence or development of the disease. In this case, the contribution of this environmental feature to the credibility of the suspected disease feature point is directly set to 1, and then the contribution of all environmental features to the credibility of the suspected disease feature point is weighted and summed. This approach ensures that, under normal environmental conditions, environmental factors will not unreasonably reduce the reliability of suspected disease characteristic points, thus avoiding oversensitivity and misjudgment under normal environmental conditions. In this way, the suspected disease characteristic analysis model can more accurately assess the likelihood of disease presence, making the calculation of disease probability more reasonable and stable. Especially when environmental conditions are relatively stable, it can effectively avoid false alarms or missed alarms caused by minor fluctuations in environmental factors.

[0128] The above technical solution more accurately reflects the impact of environmental factors on disease assessment when generating the credibility of suspected disease feature points. When environmental feature data is within the normal reference range, it avoids unreasonable reductions in credibility due to minor fluctuations in environmental factors, thus improving the stability and accuracy of the credibility calculation for suspected disease feature points. This allows subsequent suspected disease feature analysis models to focus more on the disease information reflected in the image features themselves when assessing the probability of disease presence, reducing interference from normal fluctuations in environmental factors. This, in turn, enhances the reliability and practicality of UAV-based visual enhancement images for inspection, effectively preventing false or false alarms of diseases under normal environmental conditions.

[0129] Preferably, the present invention further proposes a method for generating the basic probability of the existence of the disease, specifically including:

[0130] Through the formula: ;

[0131] There is a basic probability of disease generation. ;

[0132] In the formula, the disease has a basic probability. This refers to the location in the fused image. The disease has a basic probability. This indicates the location in the fused image. The total number of suspected disease feature points within a radius of R. This represents a reference value for the number of suspected disease feature points within a radius of R. This represents the growth rate coefficient, used to control the rate at which the basic probability of disease presence increases with the number of feature points;

[0133] The total number of suspected disease feature points within a radius of (x, y) in the fused image refers to the sum of all suspected disease feature points detected within a circular region centered at coordinates (x, y) and with a radius of R in the fused image. This number is a key indicator for measuring the density of anomalies in a local area.

[0134] The reference value for the number of suspected disease feature points within a radius of R is a preset or dynamically adjusted baseline value used to measure the normal or expected level of the number of suspected disease feature points within a given radius R. This reference value plays a normalization and comparison role when calculating the basic probability of disease presence. This reference value can be statistically analyzed based on historical inspection data; for example, by statistically analyzing inspection data from a large number of disease-free or slightly diseased areas, the average or maximum permissible number of suspected disease feature points can be obtained as a reference value.

[0135] The growth rate coefficient is a parameter used to adjust how quickly the base probability of a disease increases with the number of suspected disease feature points. This coefficient determines the sensitivity of the probability value to changes when the number of suspected disease feature points deviates from a reference value. This coefficient can be determined through model calibration and optimization. For example, by iteratively optimizing a training dataset containing both known and non-disease-affected areas, the value of the growth rate coefficient can be adjusted to best match the model's output probability to the actual situation. Alternatively, it can be set manually based on the needs of the specific application scenario; for example, if a higher sensitivity to a small number of outliers is desired, the value of the growth rate coefficient can be set larger.

[0136] This application's solution quantifies the spatial density of suspected disease feature points by introducing a formula for calculating the basic probability of disease presence. This formula cleverly utilizes the properties of an exponential function, comparing the total number of suspected disease feature points within a radius R around a location (x, y) in the fused image with a preset reference value. The larger the total number of suspected disease feature points within the range relative to the preset reference value, the denser the suspected disease feature points in that area. The term increases accordingly, leading to an increase in the exponential term. The absolute value of increases, thereby making The value of the term approaches zero, ultimately causing the base probability of disease presence to approach 1. Conversely, the opposite is true. The growth rate coefficient precisely controls the sensitivity of the probability value to changes in the number of feature points, ensuring that the model can respond appropriately to anomaly clusters of different densities according to actual needs. This density distribution-based quantification method effectively solves the problem that it is difficult to accurately determine disease risk based on a single or sparse suspected disease feature point. It provides a solid and refined foundation for subsequent in-depth disease analysis combined with environmental feature data, making the final generated disease presence probability more accurate and reliable.

[0137] Through the above technical solution, this application provides a method for quantitatively assessing the spatial clustering degree of suspected disease feature points. It can generate a more refined and accurate basic probability of disease presence based on the density distribution of suspected disease feature points. This probability calculation method based on quantity and density avoids overreaction to isolated or sparse anomalies and also avoids insufficient risk assessment of dense anomalies. It lays a solid foundation for subsequent comprehensive analysis combined with environmental feature data, enabling the final generated disease presence probability to more realistically reflect the actual risk of diseases. This significantly improves the accuracy and reliability of UAV-based visual enhancement inspections, effectively reduces false alarms and missed alarms, and improves the efficiency and quality of highway disease detection.

[0138] Preferably, the present invention further proposes the following expression for the suspected disease feature analysis model: ;

[0139] In the expression, This indicates the location in the fused image. The probability of the disease existing. This indicates the location in the fused image. The disease has a basic probability. This indicates the location in the fused image. The total number of suspected disease feature points within a radius of R. This represents the position in the q-th fused image. The reliability of suspected disease feature points within a radius of R;

[0140] The expression for the suspected disease feature analysis model is a mathematical formula used to quantitatively calculate the probability of disease presence at a specific location. This expression comprehensively considers the basic probability of disease presence and the credibility of suspected disease feature points to provide an accurate disease risk assessment value. The disease presence probability is the final output of this model, representing the likelihood of disease presence at a specific location (x, y) in the fused image. Its value is typically between 0 and 1, with a higher value indicating a higher probability of disease presence.

[0141] This application's solution provides a clear mathematical expression that enables the suspected disease feature analysis model to accurately calculate the probability of disease presence. This expression combines the basic probability of disease presence with the average confidence level of all suspected disease feature points within a local area. Specifically, firstly, the basic probability of disease presence is determined based on the spatial distribution of suspected disease feature points, reflecting the initial signs of disease on the image. Then, for each suspected disease feature point within a local area centered at the current location (x,y) with a radius of R, its corresponding confidence level is calculated. This confidence level comprehensively considers the influence of environmental feature data of the inspection area on disease occurrence. Finally, the confidence levels of these local suspected disease feature points are summed and averaged to obtain an average confidence level. This average confidence level is then multiplied by the basic probability of disease presence to obtain the final probability of disease presence. This calculation method not only considers the number and distribution of suspected disease feature points but also incorporates the influence of environmental factors on the authenticity of these feature points, making the assessment of the probability of disease presence more comprehensive and accurate. In this way, the model can effectively integrate information from multiple sources, overcoming the evaluation bias that may be caused by relying on a single information source or a fuzzy model.

[0142] Through the aforementioned technical solution, this method provides a clear and quantifiable model for analyzing suspected defects, making the calculation of defect probability transparent and standardized. This model effectively integrates the baseline defect probability based on image features with the credibility of feature points based on environmental data, thereby generating a more accurate and reliable defect probability. This significantly improves the accuracy of UAV-based visual enhancement for inspections, helping maintenance personnel to more accurately identify and locate highway defects, and optimizing inspection efficiency and decision-making quality.

[0143] Please see Figure 4 In another embodiment, the present invention also proposes a visual enhancement system for unmanned aerial vehicle (UAV) inspection of highways. This system is used to execute the aforementioned visual enhancement method for UAV inspection of highways, specifically including:

[0144] Image acquisition unit 10 is used to equip the UAV with multiple sensors and acquire inspection images from each sensor;

[0145] The feature extraction unit 20 is used to extract features from the inspection images of each sensor to obtain the feature values ​​of the inspection images of each sensor.

[0146] The image analysis unit 30 is used to generate a distribution map of suspected disease feature points in the inspection images based on the feature values ​​of the inspection images of each sensor.

[0147] The image fusion unit 40 is used to fuse all the inspection images of suspected disease feature point distribution maps to generate a fused image of suspected disease feature point distribution map.

[0148] The disease existence probability analysis unit 50 is used to acquire environmental feature data of the inspection area, establish a suspected disease feature analysis model based on the distribution map of suspected disease feature points in the fused image and the environmental feature data of the inspection area, and generate the disease existence probability.

[0149] The visual enhancement unit 60 is used to generate a visually enhanced image for drone inspection based on the probability of disease presence; wherein, the visually enhanced image for drone inspection refers to an inspection image that displays the probability of disease presence.

[0150] Please see Figure 5 The present invention further proposes that the image analysis unit 30 specifically includes:

[0151] The anomaly analysis module 31 is used to generate a characteristic anomaly index based on the feature values ​​of the inspection images of each sensor.

[0152] The distribution map output module 32 is used to generate a distribution map of suspected disease feature points in the inspection image based on the feature anomaly index.

[0153] Please see Figure 6 The present invention further proposes that the disease existence probability analysis unit 50 specifically includes:

[0154] The credibility analysis module 51 is used to acquire environmental characteristic data of the inspection area and generate credibility of suspected disease feature points based on the environmental characteristic data of the inspection area.

[0155] The basic probability analysis module 52 is used to obtain the distribution location of suspected disease feature points in the distribution map of suspected disease feature points in the fused image, and generate the basic probability of disease existence based on the distribution location of suspected disease feature points.

[0156] The final probability output module 53 is used to establish a suspected disease feature analysis model based on the basic probability of disease existence and the credibility of suspected disease feature points, and generate the probability of disease existence.

[0157] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A visual enhancement method for highway inspection by unmanned aerial vehicle, characterized in that, The method specifically includes: Equip the drone with multiple sensors and acquire inspection images from each sensor; Feature extraction is performed on the inspection images of each sensor to obtain the feature values ​​of the inspection images of each sensor; Based on the feature values ​​of the inspection images from each sensor, a distribution map of suspected disease feature points in the inspection images is generated; All inspection images of suspected disease feature point distribution maps are fused to generate a fused image of suspected disease feature point distribution map; Acquire environmental feature data of the inspection area, and establish a suspected disease feature analysis model based on the distribution map of suspected disease feature points in the fused image and the environmental feature data of the inspection area to generate the probability of disease existence. Based on the probability of disease presence, visually enhanced images of drone inspections are generated; whereby, the visually enhanced images of drone inspections refer to inspection images that display the probability of disease presence.

2. The visual enhancement method for highway inspection by UAV as claimed in claim 1, wherein, The specific methods for generating the distribution map of suspected disease feature points in the inspection images include: Based on the feature values ​​of the inspection images from each sensor, a feature anomaly index is generated; Based on the characteristic anomaly index, a distribution map of suspected disease feature points in the inspection images is generated.

3. The visual enhancement method for highway inspection by UAV as claimed in claim 2, wherein, The specific methods for generating the characteristic anomaly index include: By the formula: ; Generating a characteristic anomaly index ; In the formula, the characteristic anomaly index This refers to the position in the inspection image of the k-th sensor. Characteristic abnormality index, This represents the position in the inspection image of the k-th sensor. The i-th eigenvalue, This represents the position in the inspection image of the k-th sensor. The reference value of the i-th feature, This represents the position in the inspection image of the k-th sensor. The difference threshold of the i-th feature This is a constant term, used to prevent the denominator from being zero. This represents the position in the inspection image of the k-th sensor. The weight coefficient of the i-th feature, where n represents the position in the inspection image of the k-th sensor. The number of features.

4. The visual enhancement method for unmanned aerial vehicle (UAV) inspection of highways according to claim 1, characterized in that, The specific methods for generating the probability of the disease's existence include: Obtain environmental characteristic data of the inspection area, and generate the credibility of suspected disease feature points based on the environmental characteristic data of the inspection area; Obtain the distribution location of suspected disease feature points in the distribution map of suspected disease feature points in the fused image, and generate the basic probability of disease existence based on the distribution location of suspected disease feature points; A suspected disease feature analysis model is established based on the basic probability of disease existence and the credibility of suspected disease feature points to generate the probability of disease existence.

5. The visual enhancement method for unmanned aerial vehicle (UAV) inspection of highways according to claim 4, characterized in that, The specific methods for generating the credibility of suspected disease feature points include: Through the formula: ; Credibility of generated suspected disease feature points ; In the formula, the confidence level of suspected disease feature points This refers to the location in the fused image. The credibility of suspected disease feature points This represents the j-th environmental characteristic data of the inspection area. This represents the near endpoint of the j-th environmental feature reference range in the inspection area. This represents the difference threshold of the j-th environmental feature in the inspection area. This indicates the location in the fused image. The weighting coefficients for suspected disease feature points, where m represents the number of environmental feature data points in the inspection area. This represents the sensitivity coefficient to deviation of the ring feature.

6. The visual enhancement method for unmanned aerial vehicle (UAV) inspection of highways according to claim 5, characterized in that, The reliability of the suspected disease feature points The calculation formula is constrained by the following condition: the j-th environmental feature data of the inspection area. The j-th environmental feature is not within the reference range of the inspection area; if the j-th environmental feature data of the inspection area... Within the j-th environmental feature reference range of the inspection area, the corresponding The value of the item is 1.

7. The visual enhancement method for unmanned aerial vehicle (UAV) inspection of highways according to claim 4, characterized in that, The specific methods for generating the basic probability of the disease include: Through the formula: ; There is a basic probability of disease generation. ; In the formula, the disease has a basic probability. This refers to the location in the fused image. The disease has a basic probability of existence. This indicates the location in the fused image. The total number of suspected disease feature points within a radius of R. This represents a reference value for the number of suspected disease feature points within a radius of R. This represents the growth rate coefficient, which is used to control the rate at which the basic probability of disease presence increases with the number of feature points.

8. The visual enhancement method for unmanned aerial vehicle (UAV) inspection of highways according to claim 4, characterized in that, The specific expression of the suspected disease feature analysis model is as follows: ; In the expression, This indicates the location in the fused image. The probability of the disease existing. This indicates the location in the fused image. The disease has a basic probability of existence. This indicates the location in the fused image. The total number of suspected disease feature points within a radius of R. This represents the position in the q-th fused image. The reliability of suspected disease feature points within a radius of R.

9. A visual enhancement system for unmanned aerial vehicle (UAV) inspection of highways, characterized in that, The system is used to perform the visual enhancement method for unmanned aerial vehicle (UAV) inspection of highways as described in any one of claims 1-8, specifically including: The image acquisition unit is used to equip the drone with multiple sensors and acquire inspection images from each sensor. The feature extraction unit is used to extract features from the inspection images of each sensor to obtain the feature values ​​of the inspection images of each sensor. The image analysis unit is used to generate a distribution map of suspected disease feature points in the inspection images based on the feature values ​​of the inspection images from each sensor. The image fusion unit is used to fuse the distribution maps of suspected disease feature points in all inspection images to generate a fused image distribution map of suspected disease feature points. The disease existence probability analysis unit is used to acquire environmental feature data of the inspection area, establish a suspected disease feature analysis model based on the distribution map of suspected disease feature points in the fused image and the environmental feature data of the inspection area, and generate the disease existence probability. The visual enhancement unit is used to generate visually enhanced images for UAV inspection based on the probability of disease presence; wherein, the visually enhanced images for UAV inspection refer to inspection images that show the probability of disease presence.

10. The visual enhancement system for unmanned aerial vehicle (UAV) inspection of highways according to claim 9, characterized in that, The image analysis unit specifically includes: The anomaly analysis module is used to generate a characteristic anomaly index based on the feature values ​​of the inspection images of each sensor. The distribution map output module is used to generate a distribution map of suspected disease feature points in the inspection image based on the feature anomaly index. The disease existence probability analysis unit specifically includes: The credibility analysis module is used to acquire environmental characteristic data of the inspection area and generate credibility of suspected disease feature points based on the environmental characteristic data of the inspection area. The basic probability analysis module is used to obtain the distribution location of suspected disease feature points in the distribution map of suspected disease feature points in the fused image, and generate the basic probability of disease existence based on the distribution location of suspected disease feature points; The final probability output module is used to establish a suspected disease feature analysis model based on the basic probability of disease existence and the credibility of suspected disease feature points, and generate the probability of disease existence.

Images