Method for judging image blurring degree based on airfield pavement FOD monitoring system

Abstract

The invention relates to the technical field of image processing, in particular to a method for judging the image blurring degree based on an airfield pavement FOD monitoring system. Comprising the following steps: selecting a reference image; edge detection and energy value calculation; collecting and marking image data; constructing a deep learning model; training a deep learning model; and evaluating and predicting. According to the design, before FOD detection is carried out, the fuzzy degree of the detected image is detected, and the image which does not affect foreign matter detection is screened out for subsequent operation; by understanding the application of the image blurring degree in FOD detection, the efficiency, accuracy and stability of a detection system can be remarkably improved, manual intervention is reduced, the automation level is improved, resource utilization is optimized, and a solid guarantee is provided for airport safety and operation.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method for judging the blur degree of an image based on an airport pavement FOD monitoring system. Background Art

[0002] Image processing technology plays a vital role in judging the blur degree of an image. It not only includes common grayscale, edge detection, filtering, and feature extraction technologies, but also involves image enhancement, restoration, compression, and other aspects. The comprehensive application of these technologies can help the model understand the image content more accurately and extract effective feature information, thereby improving the accuracy of image blur judgment.

[0003] Airport FOD (Foreign Object Debris) refers to foreign matter that may damage aircraft, such as metal objects such as screws and nails, as well as concrete and asphalt fragments, which pose a serious threat to airport safety. When these foreign objects are present on the airport runway, they are easily sucked into the engine, causing engine failure; at the same time, the fragments will also accumulate in the mechanical devices, affecting the normal operation of equipment such as landing gear and wings, which will not only damage the aircraft, but may also endanger life safety.

[0004] At present, the main measure to reduce the impact of FOD is manual inspection, that is, manual inspection by inspectors and then closing the runway. However, this method has the problems of low efficiency and poor reliability, and takes up valuable runway use time. With the development of science and technology, automated FOD monitoring systems have gradually attracted attention. In such a system, it is crucial to detect the blurriness of the detection image. By screening out clear images that do not affect foreign object detection, the efficiency and accuracy of subsequent operations can be improved.

[0005] Understanding the application of image blur in FOD detection is of great significance to improving the performance of the detection system. It can significantly improve the efficiency, accuracy and stability of the detection system, reduce manual intervention, improve the level of automation, optimize resource utilization, and provide solid protection for airport safety and operation.

[0006] In addition, in recent years, deep learning technology has made significant progress in the field of image processing. By constructing a deep learning model, the features of the image can be automatically learned, thereby achieving a more accurate judgment of the degree of image blur. At the same time, combined with big data and advanced computing equipment, it is possible to quickly process and analyze massive amounts of image data, providing more reliable support for the airport pavement FOD monitoring system. In summary, image processing technology has important application value in the airport pavement FOD monitoring system, and accurate judgment of the degree of image blur is one of the key factors to improve system performance. In view of this, we propose a method for judging the degree of image blur based on the airport pavement FOD monitoring system. Summary of the invention

[0007] The purpose of the present invention is to provide a method for determining the degree of image blur based on an airport pavement FOD monitoring system, so as to solve the problems raised in the above-mentioned background technology.

[0008] In order to solve the above technical problems, one of the purposes of the present invention is to provide a method for determining the blur degree of an image based on an airport pavement FOD monitoring system, comprising the following steps:

[0009] S1. Select reference images: Select high-quality and clear images that fully cover specific application scenarios, typical objects and backgrounds from the image database based on the FOD monitoring system;

[0010] S2, edge detection and energy value calculation: perform edge detection on the reference image and calculate its energy value;

[0011] S3. Collect and annotate image data: Collect massive amounts of image data and pre-process them as the basis for model training;

[0012] S4. Construct deep learning model: Design a deep learning model and perform detailed configuration;

[0013] S5. Training deep learning model: Continuously adjust and optimize the model through training to obtain a high-performance optimized model;

[0014] S6. Evaluation and prediction: Use the optimized model to predict the blur level of the image to be predicted.

[0015] As a further improvement of the technical solution, in step S1, the specific steps of selecting the reference image are as follows:

[0016] S1.1. Determine the application scenario of the image, clarify the specific scenario and field in which the image whose blur degree is to be evaluated is applied, and the selection of the reference image needs to be carried out for the specific application scenario; collect candidate images, and collect a batch of representative high-quality images as candidate reference images from the image database based on the FOD monitoring system, covering various typical objects and backgrounds in the application scenario;

[0017] S1.2, image clarity screening, preliminary screening of candidate images, eliminating images that are obviously unclear or contain noise, artifacts, etc. that affect visual quality, and retaining images with high clarity and no obvious distortion;

[0018] S1.3. Use the image processing technology Laplace operator to perform edge detection on the candidate image and calculate the quantitative index of the edge strength of the image to reflect the clarity and detail of the image. At the same time, combine it with manual screening to obtain the benchmark image.

[0019] As a further improvement of the technical solution, in step S1.3, the expression of the Laplace operator is:

[0020]

[0021] According to the application scenario or image characteristics, weights are introduced to adjust the importance of the derivatives of pixels in different directions in the image. The optimized expression is:

[0022]

[0023] Among them, f represents the gray value of the image, x and y represent the coordinates in the image, and α and β are adjustable weight coefficients;

[0024] At the same time, the edge strength quantitative indicators include at least:

[0025] Gradient amplitude: used to reflect the intensity of the change of the pixel in a certain direction;

[0026] Gradient direction: used to indicate the direction of the edge;

[0027] Local contrast: used to measure the contrast difference between a pixel and its surrounding neighborhood;

[0028] Edge continuity: used to evaluate the degree of continuity of edges in an image;

[0029] High-frequency energy: After converting the image to the frequency domain, the energy of the high-frequency components is used to reflect the strength of the edge.

[0030] As a further improvement of the technical solution, in step S2, the specific steps of edge detection and energy value calculation are as follows:

[0031] S2.1, converting the original image into a grayscale image;

[0032] S2.2, the grayscale image is subjected to denoising using a median filter, the purpose of which is to effectively remove salt and pepper noise (i.e., impulse noise) while maintaining the edge information of the image;

[0033] S2.3, using the Laplace edge detection operator to perform edge detection on the image, and calculating the energy value of the image after edge detection;

[0034] The energy value calculation formula can be recorded as:

[0035]

[0036] Among them, X i is the pixel value of the image after edge detection, N is the total number of pixels, N = W × H, W is the width of the image, and H is the height of the image.

[0037] As a further improvement of the technical solution, in step S2.2, the specific algorithm expression of using median filtering for denoising is:

[0038] Assume that the image is I, the size is M×N, and the window size of the median filter is k×k (usually an odd number);

[0039] For each pixel (x,y) in the image:

[0040] Determine the window range centered on the pixel point. The coordinate range of the window is:

[0041] Extract the pixel values ​​within the window and get a set of pixel values:

[0042] Sort the pixel values ​​in the set P;

[0043] Take the sorted middle value as the filter output value of the pixel;

[0044] Repeat the above steps until all pixels in the image are traversed.

[0045] As a further improvement of the present technical solution, in step S3, the specific steps of collecting and annotating image data are as follows:

[0046] S3.1. Collect image data through public image databases, web crawlers to capture images that meet the requirements, or camera images, etc., to ensure the diversity of image data and cover different object lighting conditions;

[0047] S3.2. Define blur levels according to the degree of blur of the image, using a 0-1 scoring system, where 0 represents completely clear and 1 represents extremely blurry, and label according to the predefined blur criteria;

[0048] S3.3, Data preprocessing: Data cleaning, removing low-quality images and images with incorrect annotations, dealing with noise, distortion and other issues in images, and performing image data enhancement processing, such as rotation, scaling, cropping, flipping, etc., to improve the generalization ability of the model;

[0049] S3.4. Divide the data into training set, validation set and test set to ensure that the image blur degree in each data set is evenly distributed.

[0050] As a further improvement of the technical solution, in step S3.3, data cleaning at least includes:

[0051] Missing value handling:

[0052] (1) Delete rows or columns containing missing values: For missing values ​​with a small proportion, directly delete the rows or columns containing missing values;

[0053] (2) Interpolation filling: Use interpolation methods to fill missing values; where:

[0054] Linear interpolation algorithm expression: For two given data points (x1, y1) and (x2, y2), to interpolate and fill the missing value y at x, you can use the linear interpolation formula:

[0055] The nearest neighbor interpolation algorithm is expressed as: find the non-missing data point closest to the missing value and assign the value of the point to the missing value;

[0056] (3) Mean, median or mode filling: use the mean, median or mode of the data to fill in missing values;

[0057] Outlier handling:

[0058] (1) Statistical method: Identify outliers by calculating the mean and standard deviation of the data;

[0059] (2) Cluster analysis: cluster the data. Outliers are usually far away from the clusters where other data points are located.

[0060] (3) Box plot method: Use box plot to identify outliers. Data points outside the upper and lower edges of the box plot are considered outliers.

[0061] Duplicate data processing:

[0062] (1) Data deduplication: Identify and delete duplicate data rows;

[0063] (2) Data merging: merging or integrating similar data rows;

[0064] Noise data processing:

[0065] (1) Filtering: Use filtering algorithms to remove noise;

[0066] (2) Smoothing: using smoothing algorithms to reduce data fluctuations;

[0067] Data format conversion:

[0068] (1) Type conversion: convert the data type into the appropriate type;

[0069] (2) Unit conversion: unify the units of data to ensure data consistency;

[0070] Data standardization or normalization:

[0071] (1) Standardization: transform the data into a distribution with zero mean and unit variance;

[0072] (2) Normalization: Scaling the data to a specific range.

[0073] As a further improvement of the technical solution, in step S4, the specific steps of constructing the deep learning model are as follows:

[0074] S4.1. Model design: determine the number of layers, number of neurons in each layer, convolution kernel size, pooling layer configuration, etc. The model includes an input layer, several convolution layers, pooling layers, fully connected layers, and an output layer.

[0075] S4.2. Configure model hyperparameters, select learning rate, optimizer, loss function of cross entropy loss, etc., set parameters such as batch size and number of training rounds.

[0076] As a further improvement of the technical solution, in step S5, the specific steps of training the deep learning model are as follows:

[0077] S5.1. Initialize the model. Use the deep learning framework, PyTorch, to build and initialize the model.

[0078] S5.2. Train the model using the training set data, continuously adjust the model parameters to minimize the loss function, traverse the entire training set once per epoch, and evaluate the model performance on the validation set after each epoch;

[0079] S5.3. Model tuning: Use the test set data to evaluate the final model, calculate accuracy, precision, recall, F1 value and other indicators, comprehensively evaluate the model performance, and tune the model according to the evaluation results, including adjusting the network structure, changing the loss function, adjusting the optimizer, etc.; Among them:

[0080] The accuracy formula can be written as:

[0081]

[0082] The recall formula can be written as:

[0083]

[0084] The accuracy formula can be written as:

[0085]

[0086] In the above formula, precision is accuracy, recall is recall rate, ACC is accuracy rate, TP is true positive example, which is predicted as positive example and is actually positive example; FP is false positive example, which is predicted as positive example but is actually negative example; FN is false negative example, which is predicted as negative example but is actually positive example; TN is true negative example, which is predicted as negative example and is actually negative example.

[0087] As a further improvement of the technical solution, in step S6, the specific steps of evaluation and prediction are as follows:

[0088] S6.1, obtaining image data to be predicted from the video stream of the FOD monitoring system, and performing the same data preprocessing operation as that during training on the image to be predicted;

[0089] S6.2. Load the model file. Use the deep learning framework to load the previously trained model file and ensure that the model parameters are correctly loaded into the memory.

[0090] S6.3, image blur degree prediction, input the image data to be predicted into the loaded model, call the inference function of the model, predict the blur degree of the input image, and the model will output a prediction result of the blur degree;

[0091] S6.4. Analyze the prediction results according to the output format of the model.

[0092] The second object of the present invention is to provide an image blur detection platform device, including a processor, a memory, and a computer program stored in the memory and running on the processor, and the processor is used to implement the steps of the above-mentioned method for determining the image blur degree based on the airport pavement FOD monitoring system when executing the computer program.

[0093] The third object of the present invention is to provide a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the steps of the above-mentioned method for determining the degree of image blur based on the airport pavement FOD monitoring system.

[0094] Compared with the prior art, the present invention has the following beneficial effects:

[0095] 1. In the method for judging the image blur degree based on the airport pavement FOD monitoring system, the image blur degree is detected before FOD detection, and images that do not affect foreign object detection are screened out, thereby avoiding invalid processing of blurred images, thereby improving the efficiency of the entire FOD detection system for subsequent operations;

[0096] 2. The method for judging the blur degree of an image based on the airport pavement FOD monitoring system can accurately judge the blur degree of an image, which helps to more accurately identify FOD on the airport pavement, reduce misjudgments and missed judgments, and improve the accuracy of detection; the automated image blur degree judgment method reduces the reliance on manual inspections, reduces labor costs, and also improves the objectivity and consistency of detection;

[0097] 3. The method for judging the blur degree of an image based on the airport pavement FOD monitoring system can stably process and analyze the image without being affected by human factors, thereby improving the stability of the entire FOD monitoring system; avoiding wasting computing resources and time on blurry images, so that system resources can be more effectively allocated to key detection tasks;

[0098] 4. The method for judging the degree of image blur based on the airport pavement FOD monitoring system combines deep learning models and big data, which can process massive amounts of image data and adapt to the airport pavement FOD monitoring system's demand for large-scale data processing; timely and accurate detection and removal of FOD will help improve airport safety and ensure the normal operation of aircraft and the safety of passengers' lives. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Figure 1 This is an exemplary overall method flow diagram of the present invention;

[0100] Figure 2 It is a structural diagram of an exemplary electronic computer platform device in the present invention. DETAILED DESCRIPTION

[0101] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0102] Example 1

[0103] like Figure 1 As shown, this embodiment provides a method for determining the blur degree of an image based on an airport pavement FOD monitoring system, comprising the following steps:

[0104] S1. Select reference images: Select high-quality and clear images that fully cover specific application scenarios, typical objects and backgrounds from the image database based on the FOD monitoring system;

[0105] S2, edge detection and energy value calculation: perform edge detection on the reference image and calculate its energy value;

[0106] S3. Collect and annotate image data: Collect massive amounts of image data and pre-process them as the basis for model training;

[0107] S4. Construct deep learning model: Design a deep learning model and perform detailed configuration;

[0108] S5. Training deep learning model: Continuously adjust and optimize the model through training to obtain a high-performance optimized model;

[0109] S6. Evaluation and prediction: Use the optimized model to predict the blur level of the image to be predicted.

[0110] In this embodiment, in step S1, the specific steps of selecting the reference image are as follows:

[0111] S1.1. Determine the application scenario of the image, clarify the specific scenario and field in which the image whose blur degree is to be evaluated is applied, and the selection of the reference image needs to be carried out for the specific application scenario; collect candidate images, and collect a batch of representative high-quality images as candidate reference images from the image database based on the FOD monitoring system, covering various typical objects and backgrounds in the application scenario;

[0112] S1.2, image clarity screening, preliminary screening of candidate images, eliminating images that are obviously unclear or contain noise, artifacts, etc. that affect visual quality, and retaining images with high clarity and no obvious distortion;

[0113] S1.3. Use the image processing technology Laplace operator to perform edge detection on the candidate image and calculate the quantitative index of the edge strength of the image to reflect the clarity and detail of the image. At the same time, combine it with manual screening to obtain the benchmark image.

[0114] The expression of the Laplace operator is:

[0115]

[0116] According to the application scenario or image characteristics, weights are introduced to adjust the importance of the derivatives of pixels in different directions in the image. The optimized expression is:

[0117]

[0118] Among them, f represents the gray value of the image, x and y represent the coordinates in the image, and α and β are adjustable weight coefficients.

[0119] Furthermore, edge strength quantitative indicators include:

[0120] (i) Gradient amplitude: reflects the degree of change of a pixel in a certain direction. The gradient amplitude can be calculated by calculating the derivatives of the pixel in the horizontal and vertical directions, and then using the Euclidean distance or the square root of the sum of squares. The specific algorithm includes: for an image f(x,y), the horizontal derivative G at the point (x,y) x and the vertical derivative G y It can be obtained by difference calculation, for example using the Sobel operator or the Prewitt operator. Then, the gradient magnitude M can be calculated as

[0121] (ii) Gradient direction: indicates the direction of the edge. The gradient direction can be obtained by calculating the ratio of the horizontal and vertical derivatives. The gradient direction can be calculated as θ = arctan(G y / G x ).

[0122] (III) Local contrast: measures the contrast difference between a pixel and its surrounding neighborhood. It can be expressed by calculating the variance of the local area, contrast ratio and other indicators. For example, the variance of the pixel value in a small window centered on the pixel can be calculated as an indicator of local contrast. Specifically, it includes:

[0123] Assume that the image is I, and the size of the small window centered at the pixel point (x, y) is k×k (usually an odd number);

[0124] (1) Determine the window range:

[0125] The coordinate range of the window is:

[0126] (2) Extract the pixel value within the window:

[0127] Get the pixel value set within the window:

[0128] (3) Calculate the mean of the pixel values ​​within the window:

[0129] (4) Calculate the variance:

[0130] Among them, Var is the variance of the pixel value in a small window centered on the pixel point (x, y), which can be used as an indicator of local contrast.

[0131] (IV) Edge continuity: Evaluate the degree of continuity of the edge in the image. This can be expressed by calculating the edge length, continuity score and other indicators. An edge tracking algorithm can be used to determine the edge length and give a corresponding score based on the edge continuity. Specifically:

[0132] A1. Initialization: Select a starting edge point as the starting point of tracking; set the current point as the starting edge point; initialize the edge length to 0;

[0133] A2. Edge tracking: Starting from the current point, check the adjacent pixels according to the preset direction order (such as clockwise or counterclockwise); if the gray value or feature of the adjacent pixel meets the definition of edge (such as a sudden change in gray value), mark the point as an edge point and update the current point to this point; repeat step A2 until you return to the starting edge point or no adjacent point that meets the edge definition can be found;

[0134] A3. Calculate edge length: During the tracking process, record the number of edge points; the edge length is equal to the number of edge points minus 1 (because the starting point is calculated repeatedly).

[0135] The specific implementation method can be adjusted according to the image representation and the definition of the edge. For example, if it is a binary image, the edge point can be determined based on whether the pixel value is 1; if it is a grayscale image, the edge can be determined based on the change in grayscale value.

[0136] (V) High-frequency energy: After converting the image to the frequency domain, the energy of the high-frequency component can reflect the strength of the edge. The image can be converted to the frequency domain using Fourier transform, and then the energy of the high-frequency component can be calculated. Specifically:

[0137] (1) Fourier transform: Perform a two-dimensional discrete Fourier transform (DFT) on the image. The fast Fourier transform (FFT) algorithm can be used to improve computational efficiency.

[0138] Assuming that the image is f(x,y) and its size is M×M, the frequency domain representation after Fourier transform is F(u,v), where u=0,1,...,M-1 and v=0,1,...,N-1. The calculation formula of FFT is:

[0139]

[0140] (2) Calculate the energy of the high-frequency part: The frequency domain can usually be divided into low-frequency and high-frequency parts; low-frequency and high-frequency parts can be divided by setting a frequency threshold;

[0141] To calculate the energy of the high-frequency part, we can sum the squares of the amplitudes in the high-frequency region. Assuming that the coordinate range of the high-frequency region is (u1, v1) to (u2, v2), the energy of the high-frequency part is:

[0142]

[0143] Where |F(u,v)| represents the amplitude after Fourier transform.

[0144] The specific implementation method can use the corresponding mathematical library or programming language to complete the Fourier transform and energy calculation. It should be noted that the choice of frequency threshold will affect the calculation result of high-frequency energy and needs to be adjusted according to the specific situation. In addition, the frequency domain representation after Fourier transform is usually a complex number, and its real and imaginary parts need to be processed to obtain the amplitude.

[0145] In addition, the above-mentioned edge strength quantitative indicators can be used in combination to more comprehensively evaluate the edge strength of the image. In practical applications, appropriate indicators and algorithms can be selected according to specific needs and image characteristics.

[0146] In this embodiment, in step S2, the specific steps of edge detection and energy value calculation are as follows:

[0147] S2.1, converting the original image into a grayscale image;

[0148] S2.2. Use median filtering to denoise the grayscale image. The purpose is to effectively remove salt and pepper noise (i.e. impulse noise) while maintaining the edge information of the image. Among them, median filtering is a nonlinear filtering technology. It has a good removal effect on salt and pepper noise (impulse noise) and can better retain the edge information of the image. The specific algorithm expression of median filtering denoising is:

[0149] Assume that the image is I, the size is M×N, and the window size of the median filter is k×k (usually an odd number);

[0150] For each pixel (x, y) in the image, determine the window range centered on the pixel. The coordinate range of the window is:

[0151] Extract the pixel values ​​within the window and get a set of pixel values:

[0152] Sort the pixel values ​​in the set P;

[0153] Take the sorted middle value as the filter output value of the pixel;

[0154] Repeat the above steps until all pixels in the image are traversed;

[0155] S2.3, using the Laplace edge detection operator to perform edge detection on the image, and calculating the energy value of the image after edge detection; wherein the energy value calculation formula can be expressed as:

[0156]

[0157] Where, X iis the pixel value of the image after edge detection, N is the total number of pixels, N = W × H, W is the width of the image, and H is the height of the image.

[0158] In this embodiment, in step S3, the specific steps of collecting and annotating image data are as follows:

[0159] S3.1. Collect image data in a variety of ways, such as by using a public image database, web crawlers to capture images that meet the requirements, or by using camera images, to ensure the diversity of image data and cover different lighting conditions of objects. A web crawler is a program or script that automatically obtains web page content. It simulates the behavior of a browser, automatically accesses web pages according to certain rules, and extracts the required data. The main workflow of a web crawler is as follows:

[0160] B1. Determine the crawling target: clarify the scope of websites or web pages that need to be crawled;

[0161] B2. Send request: Send HTTP request to the target website to obtain the source code of the web page;

[0162] B3. Parsing web pages: Parse the acquired web page source code and extract useful information, such as text, pictures, links, etc.

[0163] B4.Store data: store the extracted data in a local database or file;

[0164] B5. Follow the rules: During the crawling process, you need to follow the website's rules and laws and regulations to avoid placing too much burden on the website or infringing on the rights of others;

[0165] It is worth mentioning that web crawlers can be used in various application scenarios, such as data collection, search engines, public opinion monitoring, price comparison, etc. However, when using web crawlers, you need to pay attention to legality and compliance and respect the website's terms of use and privacy policy;

[0166] S3.2. Define blur levels according to the degree of blur of the image, using a 0-1 scoring system, where 0 represents completely clear and 1 represents extremely blurry, and label according to the predefined blur criteria;

[0167] S3.3, Data preprocessing: Data cleaning, removing low-quality images and images with incorrect annotations, dealing with noise, distortion and other issues in images, and performing image data enhancement processing, such as rotation, scaling, cropping, flipping, etc., to improve the generalization ability of the model;

[0168] S3.4. Divide the data into training set, validation set and test set according to a certain ratio (such as 5:3:2, etc.) to ensure that the image blur degree in each data set is evenly distributed.

[0169] Furthermore, data cleaning refers to preprocessing data to remove noise, correct errors, fill missing values, etc., so as to improve the quality and usability of data. The following are some common data cleaning methods and their corresponding algorithm expressions:

[0170] 1. Missing value processing:

[0171] (1) Delete rows or columns containing missing values: If the proportion of missing values ​​is small, you can directly delete the rows or columns containing missing values;

[0172] (2) Interpolation filling: Use interpolation methods to fill missing values, such as linear interpolation, polynomial interpolation, nearest neighbor interpolation, etc.; where:

[0173] a. Linear interpolation algorithm expression: For two given data points (x1, y1) and (x2, y2), to interpolate and fill the missing value y at x, you can use the linear interpolation formula:

[0174] b. Expression of the nearest neighbor interpolation algorithm: find the non-missing data point closest to the missing value and assign the value of the point to the missing value;

[0175] (3) Mean, median or mode filling: Use the mean, median or mode of the data to fill missing values.

[0176] 2. Outlier processing:

[0177] (1) Statistical methods: outliers are identified by calculating the mean, standard deviation and other statistical quantities of the data. For example, the three-standard deviation rule can be used, that is, if the value of a data point exceeds the range of the mean plus or minus three standard deviations, it is considered an outlier.

[0178] (2) Cluster analysis: cluster the data. Outliers are usually far away from the clusters where other data points are located.

[0179] (3) Box plot method: Use box plot to identify outliers. Data points outside the upper and lower edges of the box plot can be considered as outliers.

[0180] 3. Duplicate data processing:

[0181] (1) Data deduplication: Identify and delete duplicate data rows;

[0182] (2) Data merging: Similar data rows can be merged or integrated.

[0183] 4. Noise data processing:

[0184] (1) Filtering: Use filtering algorithms to remove noise, such as median filtering, mean filtering, etc.

[0185] (2) Smoothing: Reduce data fluctuations through smoothing algorithms, such as the moving average method.

[0186] 5. Data format conversion:

[0187] (1) Type conversion: convert the data type to an appropriate type, such as converting a string to a numeric type;

[0188] (2) Unit conversion: unify the units of data to ensure data consistency.

[0189] 6. Data standardization or normalization:

[0190] (1) Standardization: Convert the data to a distribution with zero mean and unit variance, usually using the Z-score standardization formula Where x is the original data, μ is the mean, and σ is the standard deviation;

[0191] (2) Normalization: Scale the data to a specific range, such as [0, 1] or [-1, 1]. You can use the formula where x min and x max are the minimum and maximum values ​​of the data respectively.

[0192] The above are some common methods of data cleaning and the corresponding algorithm expressions. The specific application depends on the characteristics of the data and the cleaning goals. In actual operation, multiple methods are usually combined for data cleaning to ensure the quality and reliability of the data.

[0193] In this embodiment, in step S4, the specific steps of constructing the deep learning model are as follows:

[0194] S4.1. Model design: determine the number of layers, number of neurons in each layer, convolution kernel size, pooling layer configuration, etc. The model includes an input layer, several convolution layers, pooling layers, fully connected layers, and an output layer.

[0195] S4.2. Configure model hyperparameters, select learning rate, optimizer, loss function of cross entropy loss, etc., set parameters such as batch size and number of training rounds.

[0196] In this embodiment, in step S5, the specific steps of training the deep learning model are as follows:

[0197] S5.1. Initialize the model. Use the deep learning framework, PyTorch, to build and initialize the model.

[0198] S5.2. Train the model with the training set data, and continuously adjust the model parameters to minimize the loss function. Each epoch (where 1 epoch is equal to training once with all the samples in the training set) traverses the entire training set once, and evaluates the model performance on the validation set after each epoch.

[0199] S5.3. Model tuning: Use the test set data to evaluate the final model, calculate accuracy, precision, recall, F1 value and other indicators, comprehensively evaluate the model performance, and tune the model according to the evaluation results, including adjusting the network structure, changing the loss function, adjusting the optimizer, etc.; Among them:

[0200] The accuracy formula can be written as:

[0201]

[0202] The recall formula can be written as:

[0203]

[0204] The accuracy formula can be written as:

[0205]

[0206] In the above formula, precision is accuracy, recall is recall rate, ACC is accuracy rate, TP is true positive example, which is predicted as positive example and is actually positive example; FP is false positive example, which is predicted as positive example but is actually negative example; FN is false negative example, which is predicted as negative example but is actually positive example; TN is true negative example, which is predicted as negative example and is actually negative example.

[0207] In this embodiment, in step S6, the specific steps of evaluation and prediction are as follows:

[0208] S6.1, obtaining image data to be predicted from the video stream of the FOD monitoring system, and performing the same data preprocessing operation as that during training on the image to be predicted;

[0209] S6.2. Load the model file. Use the deep learning framework to load the previously trained model file and ensure that the model parameters are correctly loaded into the memory.

[0210] S6.3, image blur degree prediction, input the image data to be predicted into the loaded model, call the inference function of the model, predict the blur degree of the input image, and the model will output a prediction result of the blur degree;

[0211] S6.4. Analyze the prediction results according to the output format of the model.

[0212] like Figure 2As shown, this embodiment also provides an image blur detection platform device, which includes a processor, a memory, and a computer program stored in the memory and running on the processor.

[0213] The processor includes one or more processing cores. The processor is connected to the memory through a bus. The memory is used to store program instructions. When the processor executes the program instructions in the memory, the steps of the method for determining the degree of image blur based on the airport pavement FOD monitoring system are implemented.

[0214] Optionally, the memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0215] In addition, the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the above-mentioned method for determining the degree of image blur based on the airport pavement FOD monitoring system.

[0216] Optionally, the present invention also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the steps of the above-mentioned method for determining the degree of image blur based on an airport pavement FOD monitoring system.

[0217] A person of ordinary skill in the art can understand that the process of implementing all or part of the steps of the above-mentioned embodiments can be completed by hardware, or can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the above-mentioned storage medium can be a read-only memory, a disk or an optical disk, etc.

[0218] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for determining image blur based on an airport pavement FOD monitoring system, characterized in that: The steps include: S1. Select reference images: Select high-quality and clear images that fully cover specific application scenarios, typical objects and backgrounds from the image database based on the FOD monitoring system; S2, edge detection and energy value calculation: perform edge detection on the reference image and calculate its energy value; S3. Collect and annotate image data: Collect massive amounts of image data and pre-process them as the basis for model training; S4. Construct deep learning model: Design a deep learning model and perform detailed configuration; S5. Training deep learning model: Continuously adjust and optimize the model through training to obtain a high-performance optimized model; S6. Evaluation and prediction: Use the optimized model to predict the blur level of the image to be predicted.

2. The method for determining the image blur degree based on the airport pavement FOD monitoring system according to claim 1 is characterized in that: In step S1, the specific steps of selecting the reference image are as follows: S1.

1. Determine the application scenario of the image, clarify the specific scenario and field in which the image whose blur degree is to be evaluated is applied, and the selection of the reference image needs to be carried out for the specific application scenario; collect candidate images, and collect a batch of representative high-quality images as candidate reference images from the image database based on the FOD monitoring system, covering various typical objects and backgrounds in the application scenario; S1.2, image clarity screening, preliminary screening of candidate images, eliminating images that are obviously unclear or contain noise and artifacts that affect visual quality, and retaining images with high clarity and no obvious distortion; S1.

3. Use the image processing technology Laplace operator to perform edge detection on the candidate image and calculate the quantitative index of the edge strength of the image to reflect the clarity and detail of the image. At the same time, combine it with manual screening to obtain the benchmark image.

3. The method for determining the image blur degree based on the airport pavement FOD monitoring system according to claim 2 is characterized in that: In step S1.3, the expression of the Laplace operator is: According to the application scenario or image characteristics, weights are introduced to adjust the importance of the derivatives of pixels in different directions in the image. The optimized expression is: Among them, f represents the gray value of the image, x and y represent the coordinates in the image, and α and β are adjustable weight coefficients; At the same time, the edge strength quantitative indicators include at least: Gradient amplitude: used to reflect the intensity of the change of the pixel in a certain direction; Gradient direction: used to indicate the direction of the edge; Local contrast: used to measure the contrast difference between a pixel and its surrounding neighborhood; Edge continuity: used to evaluate the degree of continuity of edges in an image; High-frequency energy: After converting the image to the frequency domain, the energy of the high-frequency components is used to reflect the strength of the edge.

4. The method for determining the image blur degree based on the airport pavement FOD monitoring system according to claim 3 is characterized in that: In step S2, the specific steps of edge detection and energy value calculation are as follows: S2.1, converting the original image into a grayscale image; S2.2, use median filtering to perform denoising on the grayscale image to remove salt and pepper noise while maintaining the edge information of the image; S2.3, using the Laplace edge detection operator to perform edge detection on the image, and calculating the energy value of the image after edge detection; wherein the energy value calculation formula can be expressed as: Where, X i is the pixel value of the image after edge detection, N is the total number of pixels, N = W × H, W is the width of the image, and H is the height of the image.

5. The method for determining the image blur degree based on the airport pavement FOD monitoring system according to claim 4 is characterized in that: In step S2.2, the specific algorithm for denoising using median filtering is expressed as: Assume that the image is I, the size is M×N, and the window size of the median filter is k×k; For each pixel (x,y) in the image: Determine the window range centered on the pixel point. The coordinate range of the window is: Extract the pixel values ​​within the window and get a set of pixel values: Sort the pixel values ​​in the set P; Take the sorted middle value as the filter output value of the pixel; Repeat the above steps until all pixels in the image are traversed.

6. The method for determining the image blur degree based on the airport pavement FOD monitoring system according to claim 5 is characterized in that: In step S3, the specific steps of collecting and annotating image data are as follows: S3.

1. Collect image data in various ways, such as by using public image databases, web crawlers to capture images that meet the requirements, or by using cameras to capture images, to ensure the diversity of image data; S3.

2. Define blur levels according to the degree of blur of the image, using a 0-1 scoring system, where 0 represents completely clear and 1 represents extremely blurry, and label according to the predefined blur criteria; S3.3, Data preprocessing: Data cleaning, removing low-quality images and images with incorrect annotations, enhancing image data, and improving the generalization ability of the model; S3.

4. Divide the data into training set, validation set and test set to ensure that the image blur degree in each data set is evenly distributed.

7. The method for determining the image blur degree based on the airport pavement FOD monitoring system according to claim 6 is characterized in that: In step S3.3, data cleaning at least includes: Missing value handling: (1) Delete rows or columns containing missing values: For missing values ​​with a small proportion, directly delete the rows or columns containing missing values; (2) Interpolation filling: Use interpolation methods to fill missing values; where: Linear interpolation algorithm expression: For two given data points (x1, y1) and (x2, y2), to interpolate and fill the missing value y at x, you can use the linear interpolation formula: The nearest neighbor interpolation algorithm is expressed as: find the non-missing data point closest to the missing value and assign the value of the point to the missing value; (3) Mean, median or mode filling: use the mean, median or mode of the data to fill in missing values; Outlier handling: (1) Statistical method: Identify outliers by calculating the mean and standard deviation of the data; (2) Cluster analysis: cluster the data. Outliers are usually far away from the clusters where the clustered data points are located. (3) Box plot method: Use box plot to identify outliers. Data points outside the upper and lower edges of the box plot are considered outliers. Duplicate data processing: (1) Data deduplication: Identify and delete duplicate data rows; (2) Data merging: merging or integrating similar data rows; Noise data processing: (1) Filtering: Use filtering algorithms to remove noise; (2) Smoothing: using smoothing algorithms to reduce data fluctuations; Data format conversion: (1) Type conversion: convert the data type into the appropriate type; (2) Unit conversion: unify the units of data to ensure data consistency; Data standardization or normalization: (1) Standardization: transform the data into a distribution with zero mean and unit variance; (2) Normalization: Scaling the data to a specific range.

8. The method for determining the image blur degree based on the airport pavement FOD monitoring system according to claim 7 is characterized in that: In step S4, the specific steps of constructing the deep learning model are as follows: S4.

1. Model design: determine the number of layers, the number of neurons in each layer, the size of the convolution kernel, and the configuration of the pooling layer. The model includes an input layer, several convolutional layers, a pooling layer, a fully connected layer, and an output layer. S4.

2. Configure model hyperparameters, select learning rate, optimizer, loss function of cross entropy loss, and set parameters of batch size and number of training rounds.

9. The method for determining the image blur degree based on the airport pavement FOD monitoring system according to claim 8 is characterized in that: In step S5, the specific steps of training the deep learning model are as follows: S5.

1. Initialize the model. Use the deep learning framework, PyTorch, to build and initialize the model. S5.

2. Train the model using the training set data, continuously adjust the model parameters to minimize the loss function, traverse the entire training set once per epoch, and evaluate the model performance on the validation set after each epoch; S5.

3. Model tuning: Use the test set data to evaluate the final model, calculate the accuracy, precision, recall, and F1 value indicators, comprehensively evaluate the model performance, and tune the model based on the evaluation results, including adjusting the network structure, changing the loss function, and adjusting the optimizer; where: The accuracy formula can be written as: The recall formula can be written as: The accuracy formula can be written as: In the above formula, precision is accuracy, recall is recall rate, ACC is accuracy rate, TP is true positive example, which is predicted as positive example and is actually positive example; FP is false positive example, which is predicted as positive example but is actually negative example; FN is false negative example, which is predicted as negative example but is actually positive example; TN is true negative example, which is predicted as negative example and is actually negative example.

10. The method for determining the image blur degree based on the airport pavement FOD monitoring system according to claim 9 is characterized in that: In step S6, the specific steps of evaluation and prediction are as follows: S6.1, obtaining image data to be predicted from the video stream of the FOD monitoring system, and performing the same data preprocessing operation as that during training on the image to be predicted; S6.

2. Load the model file. Use the deep learning framework to load the previously trained model file and ensure that the model parameters are correctly loaded into the memory. S6.3, image blur degree prediction, input the image data to be predicted into the loaded model, call the inference function of the model, predict the blur degree of the input image, and the model will output a prediction result of the blur degree; S6.

4. Analyze the prediction results according to the output format of the model.

Images