Pavement crack detection method and device based on hyperspectral image, terminal and medium

Abstract

The invention discloses a pavement crack detection method and device based on a hyperspectral image, a terminal and a medium. The method comprises the following steps: acquiring the hyperspectral image of a target pavement based on unmanned intelligent equipment and a hyperspectral camera; performing correction and feature extraction on the hyperspectral image to determine a spectral feature matrix; and determining an image category according to a linear classifier and a dictionary obtained by learning the spectral feature matrix and the low-rank dictionary. According to the invention, through the unmanned intelligent device and the hyperspectral camera, hyperspectral image acquisition of a large-area or complex scene can be completed in a short time, errors and labor intensity of manual operation are reduced, and working efficiency and data quality are improved; meanwhile, in consideration of different spectral reflection values of different factors in the hyperspectral image, spectral features in the hyperspectral image are extracted for image classification, and the problem that false alarms may be triggered and false cracks are generated due to the influence of irrelevant factor interference when detection is carried out based on pictures collected by a common camera in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of computer vision, and in particular to a road surface crack detection method, device, terminal and medium based on hyperspectral images. Background Art

[0002] Pavement cracks are the initial manifestation of pavement diseases and the most common pavement diseases, and they play a major role in pavement disease detection. Traditional pavement crack detection relies on on-site manual visual inspection, which is inefficient, has great safety hazards, and is difficult to quickly and accurately assess the health of the pavement. Recently, with the development of computer technology and imaging technology, the use of machine vision methods instead of manual inspection can greatly improve the detection efficiency.

[0003] However, the scenes on the road surface are usually very complex. When machine vision methods detect images collected by ordinary cameras, they are affected by many irrelevant factors such as parking lines, oil stains, water stains, vegetation, etc., which may trigger false alarms and generate pseudo cracks, reducing the accuracy of road crack detection.

[0004] Therefore, the existing technology still needs to be improved and developed. Summary of the invention

[0005] The technical problem to be solved by the present invention is that, in view of the above-mentioned defects of the prior art, a pavement crack detection method, device, terminal and medium based on hyperspectral images are provided, aiming to solve the problem in the prior art that detection based on pictures collected by the camera is affected by interference from irrelevant factors, which may trigger false alarms and generate pseudo cracks, thereby reducing the accuracy of pavement crack detection.

[0006] The technical solution adopted by the present invention to solve the problem is as follows: In a first aspect, an embodiment of the present invention provides a method for detecting pavement cracks based on a hyperspectral image, wherein the method comprises: Based on the unmanned driving intelligent device and the hyperspectral camera, an image of the target road surface is collected to obtain a hyperspectral image corresponding to the target road surface; Correcting and extracting features from the hyperspectral image to determine a spectral feature matrix; A linear classifier and a dictionary obtained based on low-rank dictionary learning are obtained, and an image category is determined according to the linear classifier, the dictionary, and the spectral feature matrix.

[0007] In one implementation method, the hyperspectral image is corrected and features are extracted to determine a spectral feature matrix, including: Acquire a full black calibration image and a full white calibration image, and determine a hyperspectral correction image according to the full black calibration image, the full white calibration image, and the hyperspectral image; The spectral reflectance values ​​in the hyperspectral corrected image are extracted to determine the spectral feature matrix.

[0008] In one implementation method, a linear classifier and a dictionary are obtained based on low-rank dictionary learning, including: Construct low-rank dictionary learning optimization function; Acquire a training set, iteratively solve the low-rank dictionary learning optimization function according to the training set, and determine the dictionary and low-rank expression matrix corresponding to the training set, wherein the training set includes a number of training samples and category label matrices corresponding to the training samples; The linear classifier is determined according to the low-rank expression matrix corresponding to the training set and the category label matrix.

[0009] In one implementation method, the low-rank dictionary learning optimization function includes:

[0010] Subject to , in, is the training set, is the initial low-rank expression matrix The nuclear norm of Indicates noise of norm, is the initial dictionary, represents the square of the Frobenius norm, and as a parameter.

[0011] In one implementation method, determining the linear classifier according to the low-rank expression matrix corresponding to the training set and the category label matrix includes: Construct a linear classifier optimization function; The linear classifier optimization function is iteratively solved according to the low-rank expression matrix corresponding to the training set and the category label matrix to determine the linear classifier.

[0012] In one implementation, the linear classifier optimization function includes: , in, is the low-rank expression matrix corresponding to the training set, is the category label matrix, is the initial linear classifier, is the linear classifier, As parameters, represents the square of the Frobenius norm.

[0013] In one implementation method, determining the image category according to the linear classifier, the dictionary, and the spectral feature matrix includes: Performing a low-rank expression on the spectral feature matrix according to the dictionary to determine a low-rank expression matrix corresponding to the spectral feature matrix; The image category is determined according to the linear classifier and a low-rank expression matrix corresponding to the spectral feature matrix.

[0014] In a second aspect, an embodiment of the present invention further provides a road crack detection device based on a hyperspectral image, wherein the road crack detection device based on a hyperspectral image comprises an unmanned driving intelligent device and a hyperspectral camera mounted on the unmanned driving intelligent device, wherein the unmanned driving intelligent device carries the hyperspectral camera to collect a hyperspectral image of a target road surface, and processes and analyzes the collected hyperspectral image, and the unmanned driving intelligent device comprises: An image acquisition module, used to acquire a hyperspectral image of the target road surface collected by the hyperspectral camera; A feature extraction module, used for correcting and extracting features from the hyperspectral image, and determining a spectral feature matrix; The category determination module is used to obtain a linear classifier and a dictionary obtained based on low-rank dictionary learning, and determine the image category according to the linear classifier, the dictionary and the spectral feature matrix.

[0015] In a third aspect, an embodiment of the present invention further provides a terminal, comprising a memory and one or more processors; the memory stores one or more programs; the program comprises instructions for executing any of the above-described methods for detecting pavement cracks based on hyperspectral images; and the processor is used to execute the program.

[0016] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium on which a plurality of instructions are stored, wherein the instructions are suitable for being loaded and executed by a processor to implement any of the above-mentioned pavement crack detection methods based on hyperspectral images.

[0017] Beneficial effects of the present invention: The embodiments of the present invention collect hyperspectral images of the target road surface based on unmanned driving intelligent equipment and hyperspectral cameras; perform correction and feature extraction on the hyperspectral images to determine the spectral feature matrix; and determine the image category based on the linear classifier and dictionary obtained by learning the spectral feature matrix and low-rank dictionary. The present invention can complete the hyperspectral image collection of large areas or complex scenes in a short time through unmanned driving intelligent equipment and hyperspectral cameras, reduce errors and labor intensity of manual operations, and improve work efficiency and data quality; at the same time, considering that the spectral reflectance values ​​of different factors in the hyperspectral images are different, the present invention extracts spectral features in the hyperspectral images for image classification, which solves the problem that the existing technology is affected by interference from irrelevant factors when performing detection based on pictures collected by ordinary cameras, which may trigger false alarms and generate pseudo cracks. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a flow chart of a road crack detection method based on hyperspectral images provided in an embodiment of the present invention.

[0020] Figure 2 It is a schematic diagram of the implementation principle of the pavement crack detection method based on hyperspectral images provided in an embodiment of the present invention.

[0021] Figure 3 Spectral reflectance curves of different categories provided by the embodiments of the present invention.

[0022] Figure 4 It is a schematic diagram of the internal modules of the pavement crack detection device based on hyperspectral image provided by an embodiment of the present invention.

[0023] Figure 5 It is a principle block diagram of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The present invention discloses a road crack detection method, device, terminal and medium based on hyperspectral image. In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0025] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0026] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as herein.

[0027] Pavement cracks are the initial manifestation of pavement diseases and the most common pavement diseases, and they play a major role in pavement disease detection. Traditional pavement crack detection relies on on-site manual visual inspection, which is inefficient, has great safety hazards, and is difficult to quickly and accurately assess the health of the pavement. Recently, with the development of computer technology and imaging technology, the use of machine vision methods instead of manual inspection can greatly improve the detection efficiency.

[0028] However, the scenes on the road surface are usually very complex. When machine vision methods detect images collected by ordinary cameras, they are affected by many irrelevant factors such as parking lines, oil stains, water stains, vegetation, etc., which may trigger false alarms and generate pseudo cracks, reducing the accuracy of road crack detection.

[0029] In view of the above-mentioned defects of the prior art, the present invention provides a road crack detection method based on hyperspectral images, wherein the method collects hyperspectral images of the target road surface based on unmanned driving intelligent equipment and hyperspectral cameras; corrects and extracts features of the hyperspectral images to determine the spectral feature matrix; and determines the image category based on the linear classifier and dictionary obtained by learning the spectral feature matrix and low-rank dictionary. The present invention can complete the hyperspectral image collection of large areas or complex scenes in a short time through unmanned driving intelligent equipment and hyperspectral cameras, reduce the errors and labor intensity of manual operations, and improve work efficiency and data quality; at the same time, considering that the spectral reflectance values ​​of different factors in the hyperspectral images are different, the present invention extracts spectral features in the hyperspectral images for image classification, which solves the problem that the prior art is affected by interference from irrelevant factors when detecting based on pictures collected by ordinary cameras, which may trigger false alarms and generate pseudo cracks.

[0030] Exemplary methods: like Figure 1 As shown, the method includes: Step S100: collecting images of a target road surface based on an unmanned driving intelligent device and a hyperspectral camera to obtain a hyperspectral image corresponding to the target road surface.

[0031] Unmanned intelligent devices can achieve autonomous navigation, obstacle avoidance, and task execution through sensors, control systems, and artificial intelligence. They can operate autonomously without direct human operation, and may include drones or unmanned vehicles. Figure 2 As shown, in this embodiment, a hyperspectral camera is mounted on a drone or unmanned vehicle to build an automatic hyperspectral image acquisition system. The system can complete the hyperspectral image acquisition of large areas or complex scenes in a short time, reduce the error and labor intensity of manual operation, and thus greatly improve work efficiency and data quality.

[0032] Step S200: Correct and extract features from the hyperspectral image to determine a spectral feature matrix.

[0033] Specifically, a hyperspectral image is a three-dimensional cube, which is composed of two-dimensional image space information and one-dimensional spectral information. In terms of image space dimension, hyperspectral images are similar to traditional images and are used to describe the spatial distribution of photographed objects; in terms of spectral dimension, each pixel of a hyperspectral image can obtain a continuous spectral curve. Since different materials react differently to light, the establishment of a model based on spectral features can realize the recognition and detection of objects and targets. In this embodiment, by obtaining a hyperspectral image corresponding to the target road surface, the hyperspectral image is corrected, and the spectral curve characteristics of cracks and other common background interferences (parking lines, oil stains, water stains, vegetation) are analyzed, and spectral features are extracted from them to determine the spectral feature matrix, so as to realize the detection of cracks in the target road surface by classification based on the spectral feature matrix, and have higher detection, recognition and anti-interference capabilities in complex scenes.

[0034] In one implementation, the hyperspectral image is corrected and features are extracted to determine a spectral feature matrix, including: Step S201, obtaining a full black calibration image and a full white calibration image, and determining a hyperspectral correction image according to the full black calibration image, the full white calibration image and the hyperspectral image; Step S202: extracting spectral reflectance values ​​in the hyperspectral correction image to determine the spectral feature matrix.

[0035] Specifically, in this embodiment, the hyperspectral camera first scans a standard polytetrafluoroethylene white calibration plate with a reflectivity close to 99.99% to obtain a full white calibration image; the lens cover is covered to collect a full black calibration image with a reflectivity of almost 0%. The hyperspectral correction image is calculated based on the full black calibration image, the full white calibration image and the hyperspectral image, and the calculation formula is as follows: , in, For all-white calibration graphics, is a completely black calibration image. is the original collected hyperspectral image, Correct the image for hyperspectral.

[0036] Based on the hyperspectral correction image, the spectral curve characteristics of cracks and other common background interferences (parking lines, oil stains, water stains, and vegetation) are extracted respectively. These are used as features for subsequent image classification. For example, based on the hyperspectral image collected on the road surface, five common categories are considered, including: cracks, concrete pavement, parking lines, oil stains, and fallen leaves. Figure 3The spectral reflectance curves of the five categories are shown, and it can be seen that each category has its own unique spectral shape and different reflectance intensity (also called "spectral signature"). Therefore, these differences that are invisible to the human eye can be used as descriptive features for crack detection and other background interference distinction, avoiding false alarms caused by interference during crack detection and the generation of pseudo cracks.

[0037] Step S300: Obtain a linear classifier and a dictionary obtained based on low-rank dictionary learning, and determine the image category according to the linear classifier, the dictionary and the spectral feature matrix.

[0038] In short, this embodiment first obtains a linear classifier and a dictionary that are previously obtained through low-rank dictionary learning. The hyperspectral feature matrix is ​​represented in low rank through the dictionary, and then the hyperspectral feature matrix after low rank representation is classified through the linear classifier to obtain the image category corresponding to the spectral feature matrix. Compared with the existing machine vision algorithm, this embodiment uses a linear classifier and dictionary obtained based on low-rank dictionary learning to classify spectral features to obtain image categories. The method in this embodiment is simple, with small amount of calculation, fast calculation, high real-time performance, and is suitable for edge computing. It can be processed on low-power devices or near the data collection location (such as low-power drones and unmanned vehicles), without the need to transmit the collected data to a centralized data center for model training. Therefore, this embodiment loads the algorithm for determining the image category based on the linear classifier, dictionary, and spectral feature matrix on the unmanned driving intelligent device, and mounts the hyperspectral camera on the unmanned driving intelligent device, thereby forming a complete road crack automatic detection system, replacing traditional manual visual detection, and realizing automatic and efficient detection of road surfaces in complex scenes.

[0039] In one implementation, a linear classifier and a dictionary are obtained based on low-rank dictionary learning, including: Step S301, constructing a low-rank dictionary learning optimization function; Step S302: obtaining a training set, iteratively solving the low-rank dictionary learning optimization function according to the training set, and determining the dictionary and low-rank expression matrix corresponding to the training set, wherein the training set includes a number of training samples and category label matrices corresponding to the training samples; Step S303: determining the linear classifier according to the low-rank expression matrix corresponding to the training set and the category label matrix.

[0040] Specifically, this embodiment first obtains a training set, expresses the training set by initializing the dictionary and initializing the low-rank expression matrix, thereby constructing a low-rank dictionary representation model , is the training set, is the initial dictionary, is the initial low-rank expression matrix. Among them, the training set Include Class training samples, where each column represents a training sample, middle The value range of is: , the values ​​in the training samples are spectral reflectance values, each training sample corresponds to a category label, the category labels between the training samples may be the same or different, and the category labels corresponding to the training samples constitute a category label matrix; Represents the sample feature dimension, which is the spectral band, Indicates the number of training samples.

[0041] The low-rank dictionary representation model can be expressed as the following low-rank dictionary learning optimization function: , Subject to , in, is the training set, is the initial low-rank expression matrix The nuclear norm of is defined as the initial low-rank expression matrix The sum of the singular values ​​of The norm is The best convex optimization of the norm, The nuclear norm is a matrix rank function Convex optimization of Indicates noise of norm, is the initial dictionary, represents the square of the Frobineus norm, The term avoids scale changes during dictionary learning. and is a parameter used to balance the weights of each item.

[0042] According to the training set, the above low-rank dictionary learning optimization function is optimized and solved iteratively to obtain the dictionary corresponding to the training set. and the low-rank expression matrix . After obtaining the low-rank expression matrix corresponding to the training set After that, we can express the matrix and the class label matrix The relationship between , we get a linear classifier.

[0043] In one implementation, determining the linear classifier according to the low-rank expression matrix corresponding to the training set and the category label matrix includes: Step S3031, constructing a linear classifier optimization function; Step S3032: iteratively solve the linear classifier optimization function according to the low-rank expression matrix corresponding to the training set and the category label matrix to determine the linear classifier.

[0044] This embodiment constructs a linear classifier optimization function to represent the relationship between the non-zero coefficients in the low-rank expression matrix corresponding to the training set and the class label matrix. Specifically, the linear classifier optimization function includes: , in, is the low-rank expression matrix corresponding to the training set, is the category label matrix, is the initial linear classifier, is the linear classifier, As parameters, represents the square of the Frobenius norm.

[0045] According to the low-rank expression matrix and category label matrix corresponding to the training set, the above linear classifier optimization function is optimized and solved iteratively to obtain the closed-form solution of the linear classifier optimization function: , in, is the identity matrix, is the low-rank expression matrix corresponding to the training set The transpose of .

[0046] In one implementation, determining the image category according to the linear classifier, the dictionary, and the spectral feature matrix includes: Step S304: performing low-rank expression on the spectral feature matrix according to the dictionary to determine a low-rank expression matrix corresponding to the spectral feature matrix; Step S305: Determine the image category according to the linear classifier and the low-rank expression matrix corresponding to the spectral feature matrix.

[0047] Specifically, this embodiment uses the dictionary obtained by low-rank dictionary learning Spectral feature matrix To perform low-rank expression, is the spectral feature, middle The value range of is: , and obtain the low-rank expression matrix corresponding to the spectral feature matrix ( Based on the spectral feature matrix The mathematical representation of the obtained linear expression is: , Subject to , in, is the initial low-rank expression matrix The nuclear norm of Indicates noise of norm, For the dictionary, is the spectral feature matrix.

[0048] After obtaining the low-rank expression matrix corresponding to the spectral feature matrix, the categories of each factor in the hyperspectral image can be predicted from the expression coefficients. Since the non-zero coefficients of the low-rank representation are often concentrated in the training samples of the same type as the factors in the hyperspectral image, from the perspective of linear representation, the training samples corresponding to the larger coefficients have a greater contribution to the categories corresponding to the factors in the hyperspectral image. The category to which the spectral feature (sample to be tested) belongs can be determined based on the contribution of the training samples to the expression of the factors in the hyperspectral image. The linear classifier can express the relationship between the non-zero coefficients in the low-rank expression matrix corresponding to the training set and the category label matrix. Therefore, the category label matrix corresponding to the low-rank expression matrix of the spectral feature matrix can be solved according to the linear classifier.

[0049] The mathematical representation of determining the image category according to the low-rank expression matrix corresponding to the linear classifier and the spectral feature matrix is ​​as follows: , in, is the predicted image category, yes The Column, specifically represents the spectral characteristics About Dictionary The expression coefficient of .

[0050] Specifically, the algorithm for classifying hyperspectral images based on low-rank dictionary learning is as follows: enter: The training set corresponding to the class training sample , spectral feature matrix (sample to be tested) ,parameter , ; Output: Spectral signature Image categories: 1. By iteratively solving the low-rank dictionary learning optimization function, the low-rank expression matrix corresponding to the dictionary and the training set is obtained; 2. Perform low-rank expression on the spectral feature matrix through the learned dictionary to obtain the low-rank expression matrix corresponding to the spectral feature matrix; 3. According to the low-rank expression matrix and category label matrix corresponding to the training set, the linear classifier optimization function is iteratively solved to obtain a linear classifier, and the spectral features of the hyperspectral image are classified according to the linear classifier to determine the image category of the hyperspectral image.

[0051] This implementation has the following advantages: 1. By equipping unmanned driving intelligent equipment with a hyperspectral camera to collect images of the target road surface, an automatic hyperspectral image acquisition system is formed. The system can complete the hyperspectral image acquisition of large areas or complex scenes in a short time, reduce the errors and labor intensity of manual operation, and thus greatly improve work efficiency and data quality.

[0052] 2. Considering that different materials react differently to light, image classification is performed based on the spectral features of hyperspectral images. Compared with ordinary images that only generate images from three visible light bands (red, green, and blue), hyperspectral images provide spectral features. This solves the problem that when images collected by ordinary cameras are detected, they are affected by irrelevant factors and may trigger false alarms and produce pseudo cracks. It has higher detection, recognition, and anti-interference capabilities in complex scenarios.

[0053] 3. Classification and detection of hyperspectral images are achieved based on low-rank dictionary learning. Compared with existing machine vision algorithms, algorithms based on low-rank dictionary learning are simple, fast to calculate, and highly efficient. This enables the algorithm to run effectively in edge computing environments, such as on low-power devices, or near data collection locations (such as low-power drones and unmanned vehicles), without the need to transmit the collected data to a centralized data center for model training, and to achieve automatic detection of road cracks in complex scenarios, with advantages in practicality and accuracy.

[0054] Based on the above embodiments, the present invention also provides a road crack detection device based on hyperspectral images, such as Figure 4 As shown, the device includes an unmanned driving intelligent device 01 and the hyperspectral camera 02 mounted on the unmanned driving intelligent device, wherein the unmanned driving intelligent device 01 carries the hyperspectral camera 02 to collect hyperspectral images of the target road surface, and processes and analyzes the collected hyperspectral images. The unmanned driving intelligent device 01 includes: An image acquisition module 21 is used to acquire a hyperspectral image of the target road surface collected by the hyperspectral camera; A feature extraction module 22 is used to correct and extract features from the hyperspectral image and determine a spectral feature matrix; The category determination module 23 is used to obtain a linear classifier and a dictionary obtained based on low-rank dictionary learning, and determine the image category according to the linear classifier, the dictionary and the spectral feature matrix.

[0055] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be shown as follows: Figure 5 As shown. The terminal includes a processor, a memory, a network interface, and a display screen connected through a system bus. Among them, the processor of the terminal is used to provide computing and control capabilities. The memory of the terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the terminal is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a road crack detection method based on hyperspectral images is implemented. The display screen of the terminal can be a liquid crystal display screen or an electronic ink display screen.

[0056] Those skilled in the art will understand that Figure 5 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0057] In one implementation, the memory of the terminal stores one or more programs, and is configured to be executed by one or more processors. The one or more programs include instructions for performing a pavement crack detection method based on hyperspectral images.

[0058] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), 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).

[0059] In summary, the present invention discloses a road crack detection method, device, terminal and medium based on hyperspectral images. The method collects hyperspectral images of target roads based on unmanned driving intelligent equipment and hyperspectral cameras; corrects and extracts features of hyperspectral images to determine spectral feature matrices; and determines image categories based on linear classifiers and dictionaries obtained by learning spectral feature matrices and low-rank dictionaries. The present invention can complete hyperspectral image collection of large areas or complex scenes in a short time through unmanned driving intelligent equipment and hyperspectral cameras, reduce errors and labor intensity of manual operations, and improve work efficiency and data quality; at the same time, considering that the spectral reflectance values ​​of different factors in hyperspectral images are different, the present invention extracts spectral features in hyperspectral images for image classification, which solves the problem that the existing technology is affected by interference from irrelevant factors when detecting pictures collected by ordinary cameras, which may trigger false alarms and generate pseudo cracks.

[0060] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A road crack detection method based on hyperspectral images, characterized in that: The method comprises: Based on the unmanned driving intelligent device and the hyperspectral camera, an image of the target road surface is collected to obtain a hyperspectral image corresponding to the target road surface; Correcting and extracting features from the hyperspectral image to determine a spectral feature matrix; A linear classifier and a dictionary obtained based on low-rank dictionary learning are obtained, and an image category is determined according to the linear classifier, the dictionary, and the spectral feature matrix.

2. The method for detecting pavement cracks based on hyperspectral images according to claim 1, characterized in that: Correcting and extracting features from the hyperspectral image to determine a spectral feature matrix includes: Acquire a full black calibration image and a full white calibration image, and determine a hyperspectral correction image according to the full black calibration image, the full white calibration image, and the hyperspectral image; The spectral reflectance values ​​in the hyperspectral correction image are extracted to determine the spectral feature matrix.

3. The method for detecting pavement cracks based on hyperspectral images according to claim 1, characterized in that: Based on low-rank dictionary learning, linear classifiers and dictionaries are obtained, including: Construct low-rank dictionary learning optimization function; Acquire a training set, iteratively solve the low-rank dictionary learning optimization function according to the training set, and determine the dictionary and low-rank expression matrix corresponding to the training set, wherein the training set includes a number of training samples and category label matrices corresponding to the training samples; The linear classifier is determined according to the low-rank expression matrix corresponding to the training set and the category label matrix.

4. The method for detecting pavement cracks based on hyperspectral images according to claim 3 is characterized in that: The low-rank dictionary learning optimization function includes: ， Subject to , in, is the training set, is the initial low-rank expression matrix The nuclear norm of Indicates noise of norm, is the initial dictionary, represents the square of the Frobenius norm, and as a parameter.

5. The method for detecting pavement cracks based on hyperspectral images according to claim 3 is characterized in that: Determining the linear classifier according to the low-rank expression matrix corresponding to the training set and the category label matrix includes: Construct a linear classifier optimization function; The linear classifier optimization function is iteratively solved according to the low-rank expression matrix corresponding to the training set and the category label matrix to determine the linear classifier.

6. The method for detecting pavement cracks based on hyperspectral images according to claim 5, characterized in that: The linear classifier optimization function includes: ， in, is the low-rank expression matrix corresponding to the training set, is the category label matrix, is the initial linear classifier, is the linear classifier, As parameters, represents the square of the Frobenius norm.

7. The method for detecting pavement cracks based on hyperspectral images according to claim 1, characterized in that: Determining the image category according to the linear classifier, the dictionary, and the spectral feature matrix includes: Performing a low-rank expression on the spectral feature matrix according to the dictionary to determine a low-rank expression matrix corresponding to the spectral feature matrix; The image category is determined according to the linear classifier and a low-rank expression matrix corresponding to the spectral feature matrix.

8. A road crack detection device based on hyperspectral images, characterized in that: The device includes an unmanned driving intelligent device and a hyperspectral camera mounted on the unmanned driving intelligent device, wherein the unmanned driving intelligent device carries the hyperspectral camera to collect hyperspectral images of a target road surface, and processes and analyzes the collected hyperspectral images, and the unmanned driving intelligent device includes: An image acquisition module, used to acquire a hyperspectral image of the target road surface collected by the hyperspectral camera; A feature extraction module, used for correcting and extracting features from the hyperspectral image, and determining a spectral feature matrix; The category determination module is used to obtain a linear classifier and a dictionary obtained based on low-rank dictionary learning, and determine the image category according to the linear classifier, the dictionary and the spectral feature matrix.

9. A terminal, characterized in that: The terminal includes a memory and one or more processors; the memory stores one or more programs; the program contains instructions for executing the pavement crack detection method based on hyperspectral images as described in any one of claims 1-7; and the processor is used to execute the program.

10. A computer-readable storage medium having a plurality of instructions stored thereon, characterized in that: The instructions are suitable for being loaded and executed by a processor to implement the steps of the pavement crack detection method based on hyperspectral images as described in any one of claims 1 to 7 above.

Images