Wing frost thickness detection method and system in all-weather ground environment

Abstract

The invention relates to a wing frost thickness detection method and system in an all-weather ground environment, and the method comprises the steps: collecting a multispectral wing surface image of a simulation wing, and the multispectral wing surface image comprises a multiband image; extracting a multidimensional feature vector based on the multispectral wing surface image; based on the multi-dimensional feature vector, a machine learning classification algorithm is adopted, the wing frost types are recognized, and the wing frost types comprise a transparent ice layer, a white frost layer and a frost-free layer; identifying a frosted area boundary based on the wing frost type to obtain the frosted area boundary; and on the basis of the frosted area boundary, the frost thickness is obtained by combining calibration and geometric calculation. According to the method, the reflection and absorption characteristics of different wave bands of the wing surface can be obtained by collecting the multispectral image, the identification accuracy and reliability can be improved for distinguishing different frost types, and the accurate measurement of the frost thickness is realized by utilizing a double-branch fusion strategy edge detection technology.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of aviation flight safety, and particularly relates to a wing ice and frost thickness detection method and system under all-weather ground environment. BACKGROUND

[0002] During parking or taxiing of an aircraft, especially under adverse weather conditions such as low temperature and high humidity, icing or frosting phenomenon is prone to occur on the surface of the wing, which will seriously affect the aerodynamic performance of the aircraft and even cause flight accidents. Therefore, timely and accurate detection of wing surface ice and frost is a key link to ensure aviation safety.

[0003] Currently, manual visual inspection or handheld tools are commonly used to detect wing ice and frost at airports. This method relies on the experience and subjective judgment of the operator, and is easily affected by factors such as light, angle of view, and personnel fatigue, resulting in large uncertainty in the detection results and difficulty in standardization and quantification. Traditional contact measurement tools such as calipers and thickness gauges are difficult to adapt to the detection requirements of large-area, complex curved surfaces, and micron-level ice and frost on the wing. Especially at night, under low light or adverse weather conditions, manual detection is difficult to carry out, affecting the continuity and reliability of the detection.

[0004] In related technologies, some research attempts to use image recognition, electrical method, and structured light to detect wing ice and frost, but there are still deficiencies, which are difficult to meet the real-time and intelligent frosting detection requirements of civil aviation on the ground. SUMMARY

[0005] In view of the above-mentioned defects or deficiencies in the prior art, the present application provides a wing ice and frost thickness detection method and system under all-weather ground environment.

[0006] In a first aspect, a wing ice and frost thickness detection method under all-weather ground environment is provided, comprising: collecting a multispectral wing surface image of a simulated wing, the multispectral wing surface image comprising a multi-band image; extracting a multi-dimensional feature vector based on the multispectral wing surface image; identifying a wing ice and frost type based on the multi-dimensional feature vector using a machine learning classification algorithm, the wing ice and frost type including a transparent ice layer, a white frost layer, and no frosting; performing edge detection on the multispectral wing surface image to obtain a frosting area boundary; and obtaining the ice and frost thickness based on the frosting area boundary in combination with calibration and geometric calculation.

[0007] In an optional embodiment, the multispectral wing surface image is collected by dual-band synchronous imaging and an LED light supplement array, and the dual-band includes a visible light band and a near-infrared band.

[0008] In an alternative embodiment, a multi-dimensional feature vector is extracted based on the multispectral wing surface image, specifically comprising: calculating reflectance ratio features between different wave bands based on the multispectral wing surface image: wherein R NIR is near-infrared reflectance, R VIS is visible light reflectance; analyzing geometric features of a spectral curve based on the multispectral wing surface image to obtain spectral curve slope features: wherein R(λ) is reflectance at wavelength λ, λ1 and λ2 are characteristic wavelength points, respectively; analyzing variation features of polarization state and object surface interaction based on the multispectral wing surface image to obtain polarization characteristic parameters, the polarization characteristic parameters comprising: degree of polarization: polarization angle: ellipticity: wherein I is total light intensity, Q, U and V are Stokes parameters, respectively; wherein the multi-dimensional feature vector comprises the reflectance ratio features, the spectral curve slope features and the polarization characteristic parameters.

[0009] In an alternative embodiment, the machine learning classification algorithm is an improved multilayer perceptron, and based on the multi-dimensional feature vector, a machine learning classification algorithm is used to identify the wing ice and frost type, specifically comprising: assigning different weights to the multi-dimensional feature vector according to its contribution to ice and frost type identification, and fusing to form a weighted feature vector: F polarization = 0.6 x DOP + 0.25 x a norm + 0.15 x x norm ; F fusion = 0.4 x S slope + 0.4 x F polarization + 0.2 x R ratio ; wherein F polarization is a polarization characteristic comprehensive feature, DOP is the degree of polarization, a norm and x norm are normalized polarization angle and ellipticity, respectively, F fusion is the most weighted feature S slope is the curve slope feature, and R ratio is the ratio of near-infrared and visible light reflectance; inputting the weighted feature vector into a multilayer perceptron model to obtain an ice and frost type identification model; identifying the wing ice and frost type based on the ice and frost type identification model.

[0010] In an alternative embodiment, a frost area boundary is identified based on the wing ice and frost type to obtain the frost area boundary, specifically comprising: defining a double-branch fusion strategy: E final = a E canny + b E morph; wherein, a, β are fusion weights respectively, E canny E is an edge result of traditional edge detection, morph E is an edge result of morphological edge detection; based on the wing ice and frost type, a double-branch fusion strategy is used for frost area boundary identification, to obtain the frost area boundary, the double-branch fusion strategy being a detection strategy combining traditional edge detection and morphological edge detection.

[0011] In an optional implementation, based on the frost area boundary, calibration and geometric calculation are combined to obtain the ice and frost thickness, specifically including: converting the pixel distance measured in the multispectral wing surface image into a proportional coefficient between actual physical thickness and actual length: wherein, L real is the actual physical length of the calibration ruler, L pixel is the pixel length of the calibration ruler in the image; the wing reference surface edge line and the frost top edge line are identified through edge detection; the vertical pixel distance between the wing reference surface edge line and the frost top edge line is measured to obtain the ice and frost thickness: wherein, h is the ice and frost thickness, D pixel is the frost layer pixel distance, and θ is the camera observation angle.

[0012] In a second aspect, the present application further provides a wing ice and frost thickness detection system in an all-weather ground environment, including: a multispectral image acquisition module, adapted to acquire a multispectral wing surface image of a wing, the multispectral wing surface image including a multi-band image; an ice and frost identification module, in communication connection with the multispectral image acquisition module, extracting a multi-dimensional feature vector according to the acquired multispectral wing surface image, and identifying a wing ice and frost type based on the multi-dimensional feature vector using a machine learning classification algorithm, the wing ice and frost type including transparent ice layer, white frost layer and no frost; an edge detection and thickness measurement module, in communication connection with the ice and frost identification module, to identify a frost area boundary based on the wing ice and frost type, and obtain an ice and frost thickness based on the frost area boundary in combination with calibration and geometric calculation.

[0013] In an optional implementation, the wing ice and frost thickness detection system in an all-weather ground environment further includes: a support frame; a wing platform, rotatably arranged on the support frame; a simulated wing, arranged on the wing platform, capable of simulating the surface state of a real aircraft wing; a driving mechanism, connected with the wing platform, to drive the spatial attitude of the wing platform; wherein the multispectral image acquisition module is arranged on the wing platform to acquire the multispectral wing surface image of the simulated wing.

[0014] In an alternative embodiment, the wing ice and rime thickness detection system under all-weather ground environment further comprises a light-shielding baffle, which is arranged around the imaging area.

[0015] In an alternative embodiment, the wing ice and rime thickness detection system under all-weather ground environment further comprises a data processing control box, which is arranged on the support frame and integrated with the ice and rime identification module and the edge detection and thickness measurement module.

[0016] The wing ice and rime thickness detection system under all-weather ground environment provided by the present application has the following advantages: by collecting multispectral images, the reflection and absorption characteristics of the wing surface in different wave bands can be obtained, the feature vectors extracted from the multispectral images contain the spectral characteristics of the wing surface, which is helpful to improve the accuracy and reliability of identification in distinguishing different types of ice and rime, and by classifying the feature vectors through a machine learning algorithm, the type of ice and rime on the wing can be automatically identified, and by combining the boundary information of the icing area and the calibration and geometric calculation, the thickness of the ice and rime can be accurately measured. BRIEF DESCRIPTION OF DRAWINGS

[0017] Other features, objects, and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments thereof, which description should be taken in conjunction with the accompanying drawings:

[0018] Figure 1 A flowchart of the wing ice and rime thickness detection method under all-weather ground environment provided by an embodiment of the present application is shown in the figure;

[0019] Figure 2 A flowchart of the machine learning classification algorithm of the wing ice and rime thickness detection method under all-weather ground environment provided by an embodiment of the present application is shown in the figure;

[0020] Figure 3 A block diagram of the wing ice and rime thickness detection system under all-weather ground environment provided by an embodiment of the present application is shown in the figure;

[0021] Figure 4 A structural diagram of the wing ice and rime thickness detection system under all-weather ground environment provided by an embodiment of the present application is shown in the figure;

[0022] Figure 5 A block diagram of the data storage and management module of the wing ice and rime thickness detection system under all-weather ground environment provided by an embodiment of the present application is shown in the figure.

[0023] Reference signs: 1, solar cell panel; 2, support rod; 3, simulated wing; 4, wing platform; 5, microscopic camera; 6, rotating support arm; 7, rotatable mechanism; 8, data processing control box; 9, bottom plate; 10, light-shielding baffle. DETAILED DESCRIPTION

[0024] The application will be further described below with reference to the accompanying drawings Figure 1 and examples. It is to be understood that the specific examples described herein are merely illustrative of the application and are not intended to limit the scope of the application. It is further noted that, for the sake of brevity, conventional elements associated with the application (e.g., sensors, controllers, etc.) are not described in detail or are illustrated in block diagram form.

[0025] In the related art, some researches on aircraft icing prediction methods based on image recognition mainly rely on visible light imaging, which is difficult to obtain high-quality images in night or low-illumination environment, and the detection accuracy is limited. Some propose a three-dimensional imaging method for icing based on a piezoelectric array. The method directly contacts the wing surface through a piezoelectric sensor array to detect icing. Although this method can obtain relatively accurate thickness information, as a contact detection method, it has obvious defects: the sensor array needs to be directly attached to the wing surface, which may damage the ice layer structure and affect the authenticity of the detection results; the contact process may leave marks or cause slight damage on the wing surface, which poses a safety hazard; the arrangement and disassembly process of the piezoelectric array is complex, the detection efficiency is low, and it is difficult to meet the operation demand of airport rapid turnaround.

[0026] The existing systems lack data integration capability with airport management systems, and the information island phenomenon is serious. The system integration of these methods is low, and the automation and intelligent level is limited, which is difficult to meet the demand of all-weather, real-time and intelligent frost detection of civil aviation ground. Therefore, an intelligent detection system capable of identifying and measuring the thickness of ice and frost with high precision in all-weather environment is urgently needed.

[0027] It should be noted that the embodiments and features in the embodiments in the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0028] Reference will be made to Figure 1 A flowchart of a wing ice and frost thickness detection method in an all-weather ground environment is provided for the embodiments of the present application. The wing ice and frost thickness detection method in an all-weather ground environment comprises the following steps:

[0029] Step S101: Collecting a multispectral wing surface image of a simulated wing, the multispectral wing surface image comprising a multi-band image.

[0030] Step S103: Extracting a multi-dimensional feature vector based on the multispectral wing surface image.

[0031] Step S105: Identifying the wing ice and frost type based on the multi-dimensional feature vector using a machine learning classification algorithm, the wing ice and frost type comprising a transparent ice layer, a rime layer and no frost.

[0032] Step S107: Performing edge detection on the multispectral wing surface image to obtain a frost area boundary.

[0033] Step S109: Based on the frost area boundary, combined with calibration and geometric calculation, the ice and frost thickness is obtained.

[0034] Registration, denoising, enhancement, and other operations on multispectral images can improve the signal-to-noise ratio and ensure spatial alignment of images of different wavebands. By collecting multispectral images, the reflection and absorption characteristics of the wing surface at different wavebands can be obtained. The feature vectors extracted from the multispectral images contain the spectral characteristics of the wing surface, which helps to distinguish different types of ice and frost, and improves the accuracy and reliability of identification. Through machine learning algorithms, the feature vectors can be classified to automatically identify the type of ice and frost on the wing. The beneficial effect of this step is that it can quickly and accurately identify different types of ice and frost. Using adaptive edge detection algorithms such as the improved Canny algorithm and deep learning-based edge detection networks, the precise boundary of the frost area can be extracted. Identifying the boundary of the frost area helps to determine the coverage of the ice and frost. By combining the boundary information of the frost area with calibration and geometric calculation, the thickness of the ice and frost can be accurately measured. Automation can reduce human intervention and improve detection efficiency, making it suitable for airborne or ground-based detection systems.

[0035] Among them, the multispectral wing surface image is collected by dual-band synchronous imaging and LED light supplement array, and the dual-band includes visible light band and near-infrared band.

[0036] Using dual-band synchronous imaging method, through the spectral information collection of visible light and near-infrared band, combined with intelligent light supplement control technology, high-quality wing images in all-weather environment can be obtained.

[0037] The visible light band is 400nm-700nm, and the near-infrared band is 700nm-1100nm. Dual-band synchronous imaging method is adopted.

[0038] The intelligent light supplement system is configured with an integrated high-power LED lamp bead array, with light intensity adjustment range of 0-1000lux, and is configured with an adjustable light shielding baffle system. The light supplement system has an environmental light adaptive function, which automatically adjusts the output light intensity according to the environmental illumination.

[0039] The multi-dimensional feature vector extraction includes spectral reflectance ratio features, spectral curve shape parameters, and polarization characteristic parameters.

[0040] Further, in step S103, multi-dimensional feature vectors are extracted based on the multispectral wing surface image, which includes the following steps:

[0041] Step S1031: Calculate the reflectance ratio features between different wavebands based on the multispectral wing surface image:

[0042]

[0043] wherein R(λ) is reflectance at wavelength λ, λ1 and λ2 are characteristic wavelength points. NIR is near-infrared reflectance, R VIS is visible light reflectance.

[0044] Step S1033: The spectral curve shape parameter is obtained by analyzing the geometric characteristics of the spectral curve. The geometric characteristics of the spectral curve are analyzed based on the multispectral wing surface image to obtain the spectral curve slope characteristic:

[0045]

[0046] wherein R(λ) is reflectance at wavelength λ, λ1 and λ2 are characteristic wavelength points.

[0047] Step S1035: Based on the multispectral wing surface image, the change characteristics of the interaction between the polarization state of light and the object surface are analyzed to obtain the polarization characteristic parameter. The polarization characteristic parameter includes:

[0048] Degree of polarization:

[0049] Polarization angle:

[0050] Ellipticity:

[0051] wherein I is total light intensity, Q, U, and V are Stokes parameters, respectively.

[0052] The multi-dimensional feature vector includes the reflectance ratio characteristic, the spectral curve slope characteristic, and the polarization characteristic parameter.

[0053] Further, the machine learning classification algorithm is an improved multilayer perceptron. Based on the multi-dimensional feature vector, the machine learning classification algorithm is used to identify the ice and frost type of the wing. Specifically, the following steps are included:

[0054] Step S1051: Different weights are given to the multi-dimensional feature vector according to its contribution to the identification of the ice and frost type, and a weighted feature vector is formed by fusion:

[0055] F polarization = 0.6 × DOP + 0.25 × α norm + 0.15 × χ norm ;

[0056] F fusion = 0.4 × S slope + 0.4 × F polarization + 0.2 × R ratio ;

[0057] wherein F polarizationThe comprehensive characteristics of polarization properties, DOP is the degree of polarization, α norm , χ norm These are the normalized polarization angle and ellipticity, respectively, F fusion For the final weighted feature S slope R represents the slope characteristic of the curve. ratio This represents the ratio of near-infrared to visible light reflectance.

[0058] Step S1053: Input the weighted feature vector into the multilayer perceptron model to obtain the frost type recognition model.

[0059] Step S1055: Identify the frost type on the wing based on the frost type recognition model.

[0060] In this embodiment, the machine learning classification training specifically targets different frost types: transparent thin ice, white frost, and no frost. Multispectral image data is collected, and the main feature parameters for the three types are extracted, including the ratio of near-infrared to visible light reflectance, the slope of the spectral curve, and polarization characteristics. The extracted multidimensional feature vectors and their corresponding frost types are used as the training set to train an improved multilayer perceptron (MLP). Before the features are input into the MLP, a weighted feature fusion strategy is adopted. That is, each main feature parameter is assigned a different weight according to its contribution to frost type recognition, and the fusion forms a weighted feature vector.

[0061] like Figure 2 As shown, it is a structural diagram of a neural network for an improved multilayer perceptron, which includes an input layer, multiple hidden layers and an output layer. Each layer consists of several neurons (nodes), and the neurons are connected by weights.

[0062] The input layer has 5 neurons, each representing an input feature. Input features include: spectral reflectance ratio, spectral curve slope, polarization properties (including degree of polarization and polarization angle), and ellipticity.

[0063] The first hidden layer has 32 neurons, and the second hidden layer has 16 neurons. The function of the hidden layers is to extract and combine features by learning complex patterns in the input data. Each neuron in a hidden layer is connected to all neurons in the previous layer via weights, forming a fully connected network. Hidden layers can also include specific operations such as feature transformation, non-linear mapping, and overfitting prevention.

[0064] Through nonlinear transformations in the hidden layers, neural networks can learn complex features of the input data and perform effective feature combinations. By incorporating mechanisms to prevent overfitting (such as Dropout) into the hidden layers, the model's generalization ability can be improved, enabling it to perform better on unseen data.

[0065] The hidden layer can reduce the dimensionality of the input data, extract the most useful features, and reduce the computational complexity. Through deep learning of multiple hidden layers, the neural network can better capture the complex patterns in the data, thereby improving the accuracy of classification. Neural networks can adapt to different types of input data, and have strong robustness and adaptability.

[0066] The output layer has 3 neurons, each representing a classification result. The classification results include: no frost, transparent thin ice layer and white frost layer.

[0067] Step S107, based on the wing ice type, the frost area boundary is identified, and the frost area boundary is obtained, which includes the following steps:

[0068] Step S1071: define a double-branch fusion strategy:

[0069] E final =α·E canny +β·E morph ;

[0070] In the formula, α and β are fusion weights, E canny is the edge result of traditional edge detection, and E morph is the edge result of morphological edge detection.

[0071] Step S1073: based on the wing ice type, the double-branch fusion strategy is used to identify the frost area boundary, and the double-branch fusion strategy is a detection strategy combining traditional edge detection and morphological edge detection.

[0072] The camera geometry calibration is combined to calculate the ice thickness, and the stereo geometric measurement principle based on side imaging is adopted to calculate the actual physical thickness of the ice through the accurately identified boundary information.

[0073] Camera calibration establishes the conversion relationship between pixel coordinates and world coordinates. The relationship formula for converting a point (X w ,Y w ,Z w ) in the world coordinate system to pixel coordinates (u, v) is:

[0074]

[0075] In the formula, (u0, v0) represents the position of the principal point in the pixel coordinate system, usually located near the center of the image; dx and dy represent the width and height of each pixel, respectively; f is the focal length of the camera, R is the 3x3 rotation matrix between the two coordinate systems; T is the 3x1 translation vector between the two coordinate systems.

[0076] The image taken by the camera can have distortion, which is a characteristic inherent to the camera and can cause the straight lines in the image to be offset. The distortion can be corrected in the camera calibration process. If the target image has no obvious distortion, the distortion coefficient of the camera can be ignored.

[0077] The present application adopts edge detection technology to identify the wing surface edge line, the frost bottom edge line and the top edge line, and fuses and reconstructs the complete edge information after detecting the edges on different wave bands. Based on the geometric measurement principle of camera calibration, the microscopic camera observes the airfoil at a side view angle, identifies the wing surface edge line, the frost bottom edge line and the top edge line through the edge detection algorithm, and calculates the frost thickness in combination with the camera internal and external parameter calibration and the triangular geometric relationship. In combination with the camera geometric calibration parameters, the high-precision non-contact measurement of the frost thickness is performed through the stereoscopic geometric principle.

[0078] Further, step S109: according to the camera calibration, the physical quantification of the frost thickness is realized, the frost thickness is obtained based on the frost area boundary and in combination with the calibration and geometric calculation, and specifically includes the following steps:

[0079] Step S1091: the pixel distance measured in the multispectral wing surface image is converted into the proportional coefficient between the actual physical thickness pixel and the actual length:

[0080]

[0081] In the formula, k is the conversion coefficient of the pixel to the physical size, L real is the actual physical length of the known calibration ruler, L pixel is the pixel length of the calibration ruler in the image;

[0082] Step S1093: considering the side view observation angle of the camera, the wing reference surface edge line and the frost top edge line are identified through edge detection.

[0083] Step S1095: the vertical pixel distance between the wing reference surface edge line and the frost top edge line is measured to obtain the frost thickness:

[0084]

[0085] In the formula, h is the frost thickness, D pixel is the frost layer pixel distance, and θ is the camera observation angle.

[0086] Specifically, a sequence database can be used to store the detection data, the data table structure includes fields such as timestamp, device ID, detection location, frost type, thickness value, environmental parameters, and supports multiple communication protocols such as TCP / IP, HTTP REST API, MQTT, the data format adopts JSON standard, and can be integrated with airport AOMS, MMS and other management systems through standard interfaces. Data transmission supports 4G / 5G wireless communication, and uses AES-256 encryption algorithm to ensure data security.

[0087] On the other hand, as Figure 3 shown, the present application also proposes a wing ice thickness detection system in all-weather ground environment, comprising: a multispectral image acquisition module adapted to acquire multispectral wing surface images of the wing, the multispectral wing surface images comprising multi-band images; a frost recognition module in communication connection with the multispectral image acquisition module, extracting a multi-dimensional feature vector from the acquired multispectral wing surface images, and identifying the wing ice type based on the multi-dimensional feature vector using a machine learning classification algorithm, the wing ice type including transparent ice layer, rime layer and no frost. An edge detection and thickness measurement module in communication connection with the frost recognition module, identifying the frost area boundary based on the wing ice type, obtaining the frost area boundary, and obtaining the ice thickness based on the frost area boundary combined with calibration and geometric calculation.

[0088] Specifically, the wing ice thickness detection system in all-weather ground environment is divided into three main modules: a multispectral image acquisition module, an ice type recognition module and a thickness geometric measurement module, and a data storage and management module. Among them, the multispectral image acquisition module uses visible light and near-infrared band to acquire images, uses LED array to supplement light when light is insufficient to ensure image quality, and acquires clear images of the wing surface. The ice type recognition module performs reflectance feature analysis, analyzes the reflectance features of the image, and performs machine learning classification: uses machine learning algorithm to classify the ice type, and outputs the identification result of the ice type. The thickness geometric measurement module performs image processing edge detection: processes the image, detects the edge of the ice, then calibrates and geometrically calculates the thickness of the ice; finally, the thickness measurement result: outputs the measurement result of the ice thickness.

[0089] The multispectral image acquisition module includes a front-end sensing unit, which can be a microscope camera 5, including a visible light band imaging unit and a near-infrared band imaging unit, for image acquisition of the aircraft wing surface in all-weather ground environment. The microscope camera 5 supports external trigger synchronization mode, exposure time adjustable range 1-1000ms, equipped with anti-vibration mechanical structure, the whole machine protection level reaches IP65, working temperature range-40℃ to +50℃.

[0090] AsFigure 3 and Figure 5 As shown, the method of the present invention also includes a data storage and management module for storing and managing data, including type identification results and thickness measurement results. The frost type identification results and thickness measurement data are fused together, and data storage and transmission are completed. Data fusion and system integration are then implemented to establish a data management platform, ultimately enabling real-time monitoring and information interaction with the ground support system.

[0091] Data storage and management module: Input the measurement and identification results and perform real-time data storage: Store the data in real time for subsequent analysis.

[0092] Exceeding safety threshold: Check if the measurement result exceeds the safety threshold.

[0093] If so, push the message to the maintenance system: push the information to the maintenance system and prompt for maintenance.

[0094] If not, continue monitoring: Continue monitoring to ensure data continuity.

[0095] The measurement results are multispectral wing surface image information. The real-time database is used to store data that is currently being processed or recently acquired for quick access and analysis, while the historical database is used to store past data for long-term analysis, trend tracking, and historical comparison.

[0096] Database systems direct information flow in two main directions:

[0097] Public information platform: used for publishing and sharing data.

[0098] Business databases and other databases: These databases may be used for specific business applications or for data integration with other systems.

[0099] Furthermore, such as Figure 4 As shown, the all-weather ground environment wing frost thickness detection system also includes: a support frame; a wing platform 4, rotatably mounted on the support frame; a simulated wing 3, mounted on the wing platform 4, which can simulate the surface state of a real aircraft wing; a drive mechanism, connected to the wing platform 4, to drive the spatial attitude of the wing platform 4; wherein, a multispectral image acquisition module is mounted on the wing platform 4 to acquire multispectral wing surface images of the simulated wing 3.

[0100] In this embodiment, the support frame includes a bottom plate 9, a support rod 2 and a top frame, one end of the support rod 2 is connected with the bottom plate 9, and the other end of the support rod 2 is provided with the top frame. The solar panel 1 is installed on the top frame to provide independent and sustainable power supply for the whole device, and ensure that the system can also stably run under the condition of no external power supply. The bottom plate 9, the support rod 2 and the top frame constitute a support base system, which can provide stable mechanical support and wind resistance for the system, and adapt to long-term deployment in complex outdoor environments.

[0101] The simulation wing 3 is installed on the rotatable wing platform 4, and the simulation wing 3 can simulate the surface state of the real aircraft wing. The microscopic camera 5 is fixed to the front end of the platform and is used for high-resolution, multi-band imaging acquisition of the wing surface. The rotatable mechanism 7 and the rotating support arm 6 can realize the space attitude adjustment of the simulation wing 3 and the microscopic camera 5, so that the system can flexibly adjust the observation angle and meet different detection requirements.

[0102] The wing ice and frost thickness detection system under all-weather ground environment also includes a light shielding baffle, and the light shielding baffle is arranged around the imaging area. The light shielding baffle 10 is arranged around the imaging area, which can effectively shield external stray light, improve imaging quality and ensure the accuracy of multispectral image acquisition.

[0103] The wing ice and frost thickness detection system under all-weather ground environment also includes a data processing control box 8, and the data processing control box 8 is arranged on the support frame. The data processing control box 8 is integrated with an ice and frost identification module and an edge detection and thickness measurement module. The data processing control box 8 is integrated with core function modules such as image processing, data processing and system control, is responsible for analyzing and processing the collected multispectral images, and realizes data interaction with an external management system.

[0104] The application also provides a storage medium, which stores a computer program, and the computer program is executed by a processor to realize the wing ice and frost thickness detection method under all-weather ground environment.

[0105] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing relevant hardware, and the program can be stored in a non-volatile computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiment methods can be included. Wherein, any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM). Each embodiment in the specification is described in a progressive manner, and the same or similar parts of each embodiment can be referred to each other, and each embodiment mainly explains the difference from other embodiments. Especially, for the device, equipment, non-volatile computer storage medium embodiment, because it is basically similar to the method embodiment, so the description is relatively simple, and the relevant part can be referred to the part of the method embodiment.

[0106] The multispectral imaging technology adopted by the present application combines with the machine learning algorithm, and through the multi-dimensional feature fusion of spectral reflectance, spectral curve slope and polarization characteristics, different types of frosting states such as transparent thin ice layer and rime layer can be accurately identified. Compared with the traditional single spectral detection method, the present application has stronger identification ability and discrimination accuracy, and solves the technical problem that the prior art is difficult to distinguish different ice and frost types.

[0107] The system built by the present application is configured with an intelligent LED light supplementing system and all-weather imaging technology, and can work stably under low illumination, rain and snow and other bad weather conditions, overcoming the shortcomings of the traditional detection method limited by environmental conditions. At the same time, the system realizes the full automation process from image acquisition to result output, greatly improves the detection efficiency, reduces the manual intervention and subjective judgment error, and significantly improves the problems of low efficiency and poor consistency of the traditional manual detection method.

[0108] The application combines the multispectral imaging technology and the machine learning algorithm in the wing icing detection field, innovatively proposes the double-branch edge detection, the weighted feature fusion and other key technologies, and forms a complete technical solution.

[0109] The above description is only the preferred embodiment of the application and the explanation of the applied technical principles. It should be understood by those skilled in the art that the application range involved in the application is not limited to the technical solutions formed by the specific combination of the above technical features, and also covers other technical solutions formed by the combination of the above technical features or equivalent features without departing from the application concept. For example, the technical solutions formed by the mutual replacement of the above features and the technical features disclosed in the application (but not limited to) with similar functions.

Claims

1. A method for detecting ice and frost thickness on a wing in an all-weather ground environment, characterized by, The method comprises the following steps: Collecting a multispectral wing surface image of an aircraft wing, the multispectral wing surface image comprising multi-band images; Extracting a multi-dimensional feature vector based on the multispectral wing surface image; Identifying the wing icing type based on the multi-dimensional feature vector using a machine learning classification algorithm, the wing icing type including transparent ice layer, rime layer and no icing; Edge detection is performed on the multispectral wing surface image to obtain the icing area boundary; Based on the icing area boundary, the icing thickness is obtained by combining calibration and geometric calculation.

2. The method of claim 1, wherein The multispectral wing surface image is collected by dual-band synchronous imaging and LED light supplementing array, and the dual-band includes visible light band and near-infrared band.

3. The method of claim 1, wherein The multi-dimensional feature vector is extracted based on the multispectral wing surface image, specifically including: Based on the multispectral wing surface image, the reflectivity ratio features between different bands are calculated: wherein R NIR is near infrared reflectance, R VIS is visible light reflectance; Based on the multispectral wing surface image, the geometric features of the spectral curve are analyzed to obtain the spectral curve slope features: In the formula, R(λ) is the reflectivity at wavelength λ, λ1 and λ2 are characteristic wavelength points, respectively; Based on the multispectral wing surface image, the change features of the polarization state of light and the interaction between the object surface are analyzed to obtain the polarization characteristic parameters, including: Degree of polarization: Polarization angle: ellipticity: In the formula, I is the total light intensity, Q, U and V are Stokes parameters, respectively; The multi-dimensional feature vector includes reflectivity ratio features, spectral curve slope features and polarization characteristic parameters.

4. The method of claim 3, wherein The machine learning classification algorithm is an improved multilayer perceptron. Based on the multi-dimensional feature vector, the machine learning classification algorithm is used to identify the wing icing type, specifically including: Different weights are given to the multi-dimensional feature vector according to its contribution to the identification of the icing type, and a weighted feature vector is formed by fusion: F polarization = 0.6 x DOP + 0.25 x a norm + 0.15 x x norm ; F fusion = 0.4 x S slope + 0.4 x F polarization + 0.2 x R ratio ; In the formula, F polarization is a polarization characteristic comprehensive feature, DOP is a degree of polarization, α norm , χ norm are normalized polarization angle and ellipticity respectively, F fusion is a maximum weight feature S slope is a curve slope feature, R ratio is a near-infrared and visible light reflectance ratio. The weighted feature vector is input into the multilayer perceptron model to obtain an icing type identification model; The wing icing type is identified based on the icing type identification model.

5. The method of claim 1 to 4, wherein Based on the wing icing type, the icing area boundary is identified to obtain the icing area boundary, specifically including: Defining a dual-branch fusion strategy: E final = a · E canny + b · E morph ; In the formula, α, β are fusion weights respectively, E canny is the edge result of traditional edge detection, E morph is the edge result of morphological edge detection; Based on the wing icing type, the dual-branch fusion strategy is used to identify the icing area boundary to obtain the icing area boundary, and the dual-branch fusion strategy is a detection strategy combining traditional edge detection and morphological edge detection.

6. The method of claim 1-4, wherein, Based on the icing area boundary, the icing thickness is obtained by combining calibration and geometric calculation, specifically including: The pixel distance measured in the multispectral wing surface image is converted into a proportional coefficient between actual physical thickness pixels and actual length: where L real is the actual physical length of the known scale, L pixel is the pixel length of the scale in the image; The wing reference surface edge line and the icing top edge line are identified by edge detection; The vertical pixel distance between the wing reference surface edge line and the icing top edge line is measured to obtain the icing thickness: where h is the ice accretion thickness, D pixel is the pixel distance of the frost layer, and θ is the camera viewing angle.

7. A system for detecting frost thickness on aircraft wings in all-weather ground environments, characterized in that, The method comprises the following steps: A multispectral image acquisition module is adapted to acquire a multispectral wing surface image of an aircraft wing, the multispectral wing surface image comprising multi-band images; The ice and frost identification module is in communication connection with the multispectral image acquisition module, extracts a multi-dimensional feature vector according to the acquired multispectral wing surface image, and identifies a wing ice and frost type based on the multi-dimensional feature vector by using a machine learning classification algorithm, wherein the wing ice and frost type includes a transparent ice layer, a rime layer and no frost. The edge detection and thickness measurement module is in communication connection with the ice and frost identification module, performs frost area boundary identification based on the wing ice and frost type to obtain a frost area boundary, and obtains an ice and frost thickness based on the frost area boundary and combined calibration and geometric calculation.

8. The all-weather, ground-based, wing ice and frost thickness detection system of claim 7, wherein, Further comprising: a support frame; a wing platform rotatably arranged on the support frame; a simulation wing arranged on the wing platform, the simulation wing being capable of simulating a surface state of a real aircraft wing; a driving mechanism connected with the wing platform to drive a spatial attitude of the wing platform; wherein the multispectral image acquisition module is arranged on the wing platform to acquire the multispectral wing surface image of the simulation wing.

9. The all-weather, ground-based, wing ice and frost thickness detection system of claim 8, wherein, Further comprising a light shielding baffle arranged around an imaging area.

10. The all-weather, ground-based, wing ice and frost thickness detection system of claim 8, wherein, Further comprising a data processing control box arranged on the support frame, the data processing control box being integrated with the ice and frost identification module and the edge detection and thickness measurement module.

Images