Pavement crack detection equipment based on double spectrums

Abstract

The invention discloses pavement crack detection equipment based on double spectrums, and the equipment comprises an equipment design module which comprises a camera module, a supporting structure, an inclination angle monitoring module, an image processing unit and a mobile device; the dual-spectrum fusion detection module comprises a thermal diffusion enhancement dual-spectrum multi-scale feature fusion image preprocessing module, noise reduction and multi-scale feature fusion are carried out on images collected by the infrared thermal imaging camera and the visible light camera, and a detection result and category and position information are output through an unsupervised enhancement BiSeNet hybrid wavelet transform network to complete detection; and the pavement quality evaluation module is used for realizing pavement crack detection through image splicing and BiSeNet network segmentation, realizing pavement quality evaluation through crack distribution diagram generation and area proportion calculation, and realizing detection result display through visual labeling. According to the invention, the detection efficiency can be greatly improved, the maintenance cost can be reduced, and powerful support can be provided for timely maintenance of the road.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of target detection, and particularly relates to a movable portable complete road section pavement crack detection and diagnosis equipment based on dual spectrum. BACKGROUND

[0002] Roads are an important part of transportation infrastructure and play a crucial role in modern transportation, providing convenient and efficient paths for vehicles and public transportation, significantly reducing travel time and improving transportation efficiency. Cracks are one of the important safety hazards of roads, and if not discovered and handled in a timely manner, they can cause serious structural damage and safety accidents. The gradual expansion of cracks can affect the structural integrity of roads, and road cracks can also cause water seepage, corrosion, and potholes, affecting the durability and lifespan of the structure, increasing operating costs and economic losses in transportation.

[0003] In daily road maintenance work, the convenience of detection equipment is one of the key factors affecting detection efficiency. In recent years, multi-sensor fusion technology has received widespread attention, and the application of multi-sensor combined target detection technology can greatly reduce labor costs. How to combine the advantages and disadvantages of different sensors to accurately and effectively detect and locate has become an important direction for many scholars to study. Comprehensive detection of complete road sections can accurately determine the location of cracks and accurately determine their severity, which is of great significance for subsequent maintenance and repair. Therefore, the development and application of a portable, accurate, efficient, and full-coverage detection device can not only greatly improve detection efficiency and reduce maintenance costs, but also provide strong support for timely maintenance of roads, thereby effectively ensuring driving safety and promoting smooth and stable road transportation. SUMMARY

[0004] The purpose of the present application is achieved by the following technical solutions.

[0005] In order to overcome the defects and shortcomings of the prior art, the present application provides a movable portable complete road section pavement crack detection and diagnosis equipment based on dual spectrum.

[0006] The technical scheme of the present application is: a movable and portable complete road section pavement crack detection and diagnosis equipment based on dual-spectrum, comprising an equipment design module, a dual-spectrum fusion detection module, and a pavement quality evaluation module. The equipment design module realizes the free movement and high adaptability of an infrared thermal imaging camera and a visible light camera through a movable and portable equipment. The dual-spectrum fusion detection module comprises a thermal diffusion enhanced dual-spectrum multi-scale feature fusion image preprocessing module, which outputs detection results, categories, and location information through an unsupervised enhanced BiSeNet hybrid wavelet transform network to complete detection. The pavement quality evaluation module comprises accurate detection of pavement cracks, pavement quality evaluation through crack distribution map generation and area ratio calculation, and intuitive display of detection results through visual labeling.

[0007] Preferably, the equipment design module comprises a camera module, a support structure, an inclination monitoring module, an image processing unit, and a moving device. The camera module combines visible light and infrared thermal imaging technology to highlight pavement crack features. The support structure provides basic support through a stable triangular support. The inclination monitoring module ensures image shooting angle consistency through a mini digital inclinometer. The image processing unit realizes efficient image analysis using an NVIDIA Jetson Xavier NX and carries a visual screen to display detection conditions in real time. The moving device contains adjustable knobs and detachable small wheels to adapt to various carriers and realize free movement.

[0008] Further preferably, the movable and portable equipment realizes flexible movement and high efficiency adaptability of the detection equipment by integrating an infrared thermal imaging camera and a visible light camera. The equipment is equipped with a mini digital inclinometer, which can monitor and display the inclination angle in real time to ensure the stability of the detection process. A stable triangular support provides basic support, and the support bottom is equipped with adjustable knobs to flexibly adjust the base angle according to the needs of different carrier platforms, meeting various installation requirements. In addition, the equipment also integrates a detachable small wheel moving device, which facilitates quick installation or disassembly, further improving the portability and adaptability of the equipment. At the same time, the equipment is designed with fixing holes, and users can choose to install small wheels or fix the equipment on other platforms through screws according to specific needs, enhancing the versatility and flexibility of the equipment.

[0009] Further preferably, the image processing unit employs NVIDIA Jetson Xavier NX to process and analyze the images captured by the infrared thermal imaging camera and the visible light camera in real time. The powerful capabilities of Jetson Xavier NX in the field of edge computing are fully utilized, allowing the use of trained models directly without the need for lightweight versions of the framework. This ensures fast response and efficient execution of data processing, providing a high-performance and efficient solution for road crack detection. The device is equipped with a visual screen that can display the detection situation in real time.

[0010] Further preferably, the dual-spectrum fusion detection module includes a thermal diffusion enhanced dual-spectrum multi-scale feature fusion image preprocessing module. Non-local mean filtering (NLM) is used to denoise images of two different spectra; Sigmoid transformation is used to enhance the features of infrared thermal images; and Laplacian-based thermal diffusion processing is used for pseudo-three-dimensional data processing to enhance image features. This method further includes a multi-scale fusion step based on a Gaussian pyramid, which unifies the processing of crack sizes in visible light and infrared spectrum images and implements a weighted fusion strategy to enhance the visibility of cracks.

[0011] Further preferably, the unsupervised enhanced BiSeNet hybrid wavelet transform network constructs a customized dataset containing unlabeled images and a small number of labeled images for unsupervised pre-training and model fine-tuning. The network architecture combines the Deep InfoMax unsupervised learning mechanism and the BiSeNet dual-path network structure, initializes BiSeNet using the pre-trained DeepInfoMax encoder weights, and realizes efficient transfer learning. On this basis, the network embeds the HWD feature extraction module based on Haar wavelet transform into the spatial path of BiSeNet, replacing the traditional stride convolution operation, thereby significantly enhancing the feature learning ability and the accuracy of capturing image details. In addition, the network accurately predicts the crack type through the classification head and comprehensively evaluates the model through key performance indicators. Finally, the optimized model is deployed on the NVIDIA Jetson Xavier NX platform to realize efficient road crack detection and edge computing functions.

[0012] Further preferably, the image stitching, segmentation and detection determination method comprises collecting road section images and retaining 10% overlapping parts for seamless stitching, dividing the stitched images into appropriate blocks to retain crack details. The image of each block is input into an unsupervised enhanced BiSeNet hybrid network, thereby obtaining a crack segmentation mask and a classification label for each block. The segmentation mask represents a binary mask of the crack area, and the classification label can clearly distinguish between small cracks, large cracks and subsidence cracks, preparing for subsequent road surface quality determination. Finally, by stitching the segmentation masks of all blocks, a complete road section crack distribution map is constructed. This distribution map can accurately determine the location of the cracks, and the classification label can also lock the crack level at the same time, thereby providing a clear target for subsequent road maintenance.

[0013] Further preferably, the comprehensive evaluation system based on the segmentation mask and the classification label accurately analyzes the road crack condition. By calculating the crack area ratio and assigning weights to different crack types, the weighted ratio is used to quantify the severity of the cracks, which are divided into five levels: no cracks, mild, moderate, severe and extremely severe. On this basis, the system superimposes crack description information on the complete road section image after stitching to generate a visual image with crack annotations.

[0014] The advantages of the present application are: 1. The movable and portable complete road section pavement crack detection and diagnosis device based on dual-spectrum provided by the present application solves the image deviation problem caused by environmental changes or unstable installation of traditional detection equipment by integrating a visible light camera and an infrared thermal imaging camera and combining a mini digital tilt angle meter for real-time monitoring of the angle. This design not only ensures clear capture of crack features, but also compensates for the shortcomings of single imaging mode through dual-spectrum imaging technology, further improving detection accuracy. The device uses the NVIDIA Jetson Xavier NX platform for real-time image processing and analysis, fully utilizes its powerful edge computing capability, directly runs the trained model without the need for a lightweight framework, thereby ensuring fast response and efficient execution of data processing, and displays the detection situation in real time through the visual screen. This technical choice not only reduces bandwidth consumption and operating costs, but also simplifies the detection process and improves detection efficiency.

[0015] 2.The portable complete road section pavement crack detection and diagnosis device based on dual-spectrum provided by the application, in terms of mobility and adaptability of the device, the application realizes free movement and flexible fixation of the device by designing a set of comprehensive support and movement solutions. The scheme takes a stable triangular support as the core support structure, is equipped with adjustable knobs, and can flexibly adjust the angle according to different mounting platforms. In addition, the device integrates detachable small wheel moving devices, which are convenient for quick installation or removal, further improving the portability of the device. At the same time, the device is designed with fixing holes, and users can choose to install small wheels or fix the device on other platforms according to specific needs, ensuring its stability and adaptability in diversified environments.

[0016] 3.The portable complete road section pavement crack detection and diagnosis device based on dual-spectrum provided by the application, through the thermal diffusion enhanced dual-spectrum multi-scale feature fusion image preprocessing module in the dual-spectrum fusion detection module, by comprehensively utilizing visible light and infrared spectrum data and a multi-scale analysis method, the features of road cracks are strengthened and highlighted, and image fusion is performed, effective image preprocessing is completed for visible light and infrared spectrum data, and a data set is prepared for subsequent crack detection.

[0017] 4.The portable complete road section pavement crack detection and diagnosis device based on dual-spectrum provided by the application, through the unsupervised enhanced BiSeNet hybrid wavelet transform network in the dual-spectrum fusion detection module, the recognition ability of the network for road crack features is enhanced, the detection result, category and position information are output, efficient crack detection is realized, and edge computing deployment is facilitated.

[0018] 5.The portable complete road section pavement crack detection and diagnosis device based on dual-spectrum provided by the application, through the pavement quality evaluation module, the pavement condition is comprehensively evaluated, the crack position is accurately and efficiently judged, and the detection result is intuitively displayed through visual labeling. This greatly facilitates the decision-making process of road maintenance personnel and engineers, improves work efficiency, reduces labor costs, and provides a new idea for the monitoring and maintenance of road infrastructure. BRIEF DESCRIPTION OF DRAWINGS

[0019] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments with reference made to the accompanying drawings. The drawings are for purposes of illustration only and are not considered as limiting the application thereto. Moreover, like reference numerals are used to designate like parts throughout the specification and drawings. In the drawings: Figure 1 The structure flow chart of the portable complete road section pavement crack detection and diagnosis device based on dual-spectrum described in the application; Figure 2The device structure schematic diagram of the dual-spectrum-based movable portable complete road section pavement crack detection and diagnosis equipment according to the application; Figure 3 The device design module structure principle diagram according to the application; Figure 4 The heat diffusion enhanced dual-spectrum multi-scale feature fusion image preprocessing module logic diagram according to the application; Figure 5 The unsupervised enhanced BiSeNet hybrid wavelet transform network principle diagram according to the application; Figure 6 The pavement quality evaluation method logic diagram according to the application. DETAILED DESCRIPTION

[0020] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0021] The application discloses a dual-spectrum-based movable portable complete road section pavement crack detection and diagnosis equipment. A new movable portable equipment is designed to meet the convenience requirement of detection, adapt to various road conditions, adapt to different carriers, and improve the detection efficiency. Long-wave infrared spectrum and visible spectrum images are fused to improve the detection accuracy. A pavement quality evaluation method is standardized to meet the requirements of pavement quality determination and crack accurate positioning.

[0022] As Figure 1As shown, the present invention provides a portable, mobile, and complete road surface crack detection and diagnosis device based on dual-spectrum imaging. The device includes a device design module, a dual-spectrum fusion detection module, and a road surface quality assessment module. The device design module includes a support structure, a camera module, a tilt monitoring module, an image processing unit, and a mobile device. The portable device enables free movement and high adaptability of the infrared thermal imaging camera and the visible light camera. It is equipped with an edge computing device (e.g., NVIDIA Jetson Xavier NX) for efficient image analysis and a visual screen to display the detection status in real time. The dual-spectrum fusion detection module includes a thermal diffusion-enhanced dual-spectrum multi-scale feature fusion image preprocessing module. This module denoises the images acquired by the infrared thermal imaging camera and the visible light camera and performs multi-scale feature fusion. It outputs the detection results, category, and location information through an unsupervised enhanced BiSeNet hybrid wavelet transform network. The road surface quality assessment module achieves accurate detection of road surface cracks through image stitching and BiSeNet network segmentation, assesses road surface quality through crack distribution map generation and area ratio calculation, and provides an intuitive display of the detection results through visual annotation.

[0023] Specifically, the implementation process of the portable, dual-spectrum-based pavement crack detection and diagnosis device for complete road sections described in this invention includes the following:

[0024] 1. Equipment design module for a portable, dual-spectrum-based, full-section pavement crack detection and diagnosis device: like Figure 2 As shown, the present invention provides a portable, mobile, and complete road surface crack detection and diagnosis device based on dual-spectrum imaging. The device design modules include a camera module, a support structure, a tilt monitoring module, an image processing unit, and a mobile device.

[0025] The camera module is equipped with both an infrared thermal imaging camera and a visible light camera. The presence of water or air at the crack causes differences in thermal conductivity, hindering or exacerbating heat conduction and thus affecting heat transfer. Furthermore, the crack area differs from the surrounding road surface in terms of structure, resulting in variations in illumination and ventilation. Therefore, the infrared thermal imaging camera can clearly capture the characteristics of the crack, complementing the images captured by the visible light camera.

[0026] The support structure uses a sturdy triangular bracket as its core, providing a solid foundation for the equipment. Its triangular design utilizes the principle of geometric stability to ensure the overall structural robustness. This not only provides a solid foundation for the equipment but also effectively disperses and absorbs external impacts, ensuring stable operation in various complex environments. Furthermore, the adjustable design of the triangular bracket allows it to adapt to equipment of different sizes and weights, further enhancing its versatility and practicality.

[0027] The tilt monitoring module uses a lightweight and portable digital tilt meter. This device has a measurement range of 4×90°, an accuracy of ±0.2°, a resolution of 0.01°, and an EBTN display. The device measures 60×60×30mm and is designed with a common magnetic base for easy fixed measurement.

[0028] The image processing unit utilizes edge computing devices (including but not limited to NVIDIA Jetson Xavier NX) to process and analyze the acquired images in real time. This choice fully leverages the powerful capabilities of edge computing, allowing the direct use of pre-trained models without requiring a lightweight version of the framework, ensuring rapid response and efficient execution of data processing. The device is equipped with a visual screen to display the detection status in real time, thus providing a high-performance and high-efficiency solution for road crack detection.

[0029] The mobile device consists of a knob, detachable wheels, and mounting holes. An adjustable knob at the bottom of the bracket allows for flexible adjustment of the tripod base angle to meet the required angle when fixed to different platforms. Furthermore, the device integrates detachable wheels for quick installation and removal to adapt to various movement needs. The device also features mounting holes, allowing users to choose between installing wheels or securing the device to other platforms with screws, depending on their specific requirements.

[0030] like Figure 3 The diagram shown is a structural schematic of the portable, dual-spectrum-based pavement crack detection and diagnosis device for complete road sections provided by the present invention.

[0031] 2. Thermal diffusion-enhanced dual-spectral multi-scale feature fusion image preprocessing module: like Figure 4 The diagram shown is a logic diagram of the thermally diffused dual-spectral multi-scale feature fusion image preprocessing module. The thermally diffused dual-spectral multi-scale feature fusion image preprocessing can be divided into four steps: (1) Non-Local Means (NLM) filtering for complex cracks based on similarity principle.

[0032] The complex fracture non-local mean filter based on similarity principle is used to denoise the visible spectrum image and long-wave infrared spectrum image. By searching for similar pixel blocks in the whole image range and using these redundant information, the NLM filter effectively suppresses the noise while preserving the image detail features, providing high-quality image data for the accurate extraction of fracture features. The principle formula is as follows: Assuming that the noisy image is , the denoised image is , The pixel gray value at is calculated as follows: ; Among them, the weight represents the similarity between the pixel points and , and its value is determined by the distance between the rectangular neighborhood , centered at : ; ; ; ; Among them, is the normalization coefficient, is the smoothing parameter, d is the side length of the small neighborhood window, is the small search window area, and controls the attenuation degree of the Gaussian function. The larger the Gaussian function, the more gentle the change, and the higher the denoising level, but it will also cause the image to be more blurred; The smaller, the more edge detail components are preserved, but there will be too many noise points left; The specific value of

[0033] (2) Sigmoid transformation of fracture features based on logical function. The long-wave infrared spectrum image after NLM processing is subjected to Sigmoid transformation, effectively enhancing the fracture features and making them more prominent in the image. The principle formula is as follows: ; Among them, controls the steepness of the function, is the center point of the function. The function maps the input value to the range of (0, 1) and has smooth transition.

[0034] ​(3) Thermal diffusion processing based on the Laplacian operator. By simulating the diffusion process of heat in the image, the visibility of key information in the image, especially crack features in the visible light and infrared images, is further improved. The visible light and infrared images processed in the first two steps are stacked into pseudo-3D data, so that the two images are processed simultaneously as different channels. Then, the thermal diffusion algorithm is applied to smooth high-frequency noise and unnecessary details in the image by iteratively calculating each pixel, thereby enhancing the prominence of important features such as cracks.

[0035] During thermal diffusion processing, heat (or changes) in the image diffuses to surrounding pixels via the Laplacian operator, smoothing abrupt crack edges and stabilizing smooth areas. Through multiple iterations, the thermal diffusion algorithm effectively removes noise from visible and infrared images and enhances the saliency of cracks and other features. By using stacked pseudo-3D data, thermal diffusion not only processes data within a single image channel but also shares information across channels, allowing crack features to be complementary between visible and infrared images, further improving crack identifiability. The image after thermal diffusion retains more key information, and crack features are clearer, resulting in two independent datasets that further improve image quality and feature extraction accuracy.

[0036] (4) Gaussian pyramid fusion of dual-spectral images based on multi-scale image fusion. Gaussian pyramids are used to unify the size of cracks in visible and infrared spectral images, ensuring a one-to-one correspondence between features in the two images. Based on this, a weighted fusion strategy is further implemented, assigning higher weights to images with more prominent crack features. This fusion method not only preserves key information in the dual-spectral images but also enhances crack visibility, thus providing richer and more accurate image data for accurate crack detection and analysis. It completes effective image preprocessing of visible and infrared spectral data, preparing a dataset for subsequent crack detection.

[0037] 3. Unsupervised Enhancement BiSeNet Hybrid Wavelet Transform Network: like Figure 5 The diagram shows the principle of the unsupervised enhanced BiSeNet hybrid wavelet transform network. It combines DeepInfoMax (DIM) unsupervised learning with the BiSeNet dual-path network, further integrating a fully connected layer and a wavelet transform-based feature extraction module. This not only enhances the model's feature learning ability but also improves the ability to capture image details through the wavelet transform module, thus achieving higher accuracy and efficiency in image segmentation tasks. The unsupervised enhanced BiSeNet hybrid wavelet transform network can be divided into 8 steps: (1) Dataset preparation. The dataset consists of two parts: one part is unlabeled images used for unsupervised pre-training phase to enhance the network's ability to recognize road crack features; the other part is a small amount of labeled images used for fine-tuning the segmentation model. The labeled images contain crack segmentation labels indicating the presence or absence of cracks in the image, as well as crack classification labels to distinguish crack types.

[0038] (2) Unsupervised task learning based on Deep InfoMax. Unlabeled images are used for Deep InfoMax unsupervised task learning training. After learning image features, pre-trained encoder parameters can be obtained.

[0039] (3) Transfer learning. The DIM pre-trained encoder is used as the initialization weight of BiSeNet to adapt to specific downstream tasks, providing pre-trained feature representations that help improve the performance and efficiency of new tasks, and transferring parameters to the spatial path and context path of the BiSeNet network.

[0040] (4) HWD module based on Haar wavelet transform. The HWD (Haar Wavelet-based Downsampling) module based on Haar wavelet transform is integrated into the spatial path of the BiSeNet network, replacing the original stride convolution. Not only does it enhance the expression ability of features, but also effectively captures and preserves the low-frequency and high-frequency components of the image through Haar wavelet transform, achieving better performance in detail feature extraction and overall structure preservation, significantly improving the accuracy and efficiency of segmentation.

[0041] (5) Model fine-tuning. To further improve the accuracy of crack segmentation, a small amount of labeled data set is used to fine-tune the model. During the fine-tuning process, loss functions suitable for segmentation tasks such as Dice loss or cross-entropy loss are used to optimize model performance. Road crack detection is performed on the fine-tuned network model, and the model can automatically segment the crack area in the image and generate crack position and size information for subsequent data analysis or system integration.

[0042] (6) Model classification. To further distinguish crack types, a classification head - fully connected layer is added to the output layer to predict the type of each crack region, such as small cracks, large cracks, and subsidence cracks.

[0043] (7) Model Evaluation. The model evaluation comprehensively utilizes key indicators such as F1 score, mean squared error (MSE), and recall to quantify the model's performance and reliability. The F1 score, through the harmonic mean of precision and recall, balances the model's accuracy and completeness in identifying positive samples; the mean squared error measures the deviation between predicted and true values, reflecting the model's prediction accuracy; and the recall rate evaluates the model's ability to detect positive samples. The comprehensive analysis of these indicators provides a scientific basis for model optimization.

[0044] (8) Model Deployment and Visualization. In order to achieve high-efficiency crack detection and facilitate edge computing deployment, the trained model is optimized and integrated into the edge computing device platform, equipped with a visual screen to display the detection status in real time, thus providing a high-performance and high-efficiency solution for road crack detection and providing strong technical support for on-site detection and immediate processing.

[0045] 4. Road surface quality assessment: like Figure 6 The diagram shown is a logic diagram of the road surface quality assessment method. This invention proposes a road surface quality assessment method based on image analysis, aiming to assess road quality by calculating the proportion of cracks in the road surface. The road surface quality assessment method can be divided into four steps: (1) Image stitching and segmentation. The image stitching and segmentation are achieved by acquiring images of the road segment, retaining a 10% overlap between images to achieve seamless stitching and form a continuous image of the road segment. Subsequently, the stitched image is divided into blocks of appropriate size to ensure that crack details are fully preserved.

[0046] (2) Acquisition of crack segmentation masks and classification labels. The image of each block is input into a self-supervised BiSeNet hybrid network to obtain the crack segmentation mask and classification label for each block. The segmentation mask represents the binary mask of the crack region, and the classification label can clearly distinguish between small cracks, large cracks, and subsidence cracks, preparing for subsequent pavement quality assessment. Finally, by stitching together the segmentation masks of all blocks, a complete crack distribution map of the road segment is constructed. This distribution map can accurately determine the location of cracks, and the classification label can also lock the crack level while locating the crack, thus providing a clear target for subsequent road maintenance.

[0047] (3) Based on segmentation masks and classification labels, this invention provides a method for calculating the crack area ratio of each target region. This invention assigns different weights to each type of crack; for small cracks: weight... Large Crack: Weight ; Subsidence cracks: weight For each type of crack (let's assume it's type i, i=1, 2, 3), the area of ​​each crack type is calculated using a segmentation mask. According to the defined weight coefficient and the area of the crack type, the weighted ratio can be calculated , and the formula is as follows: ; Wherein is the weight coefficient of crack type i; is the area of the i-th crack type; is the total area of the image. When , it is no crack or slight crack; when , it is mild crack; when , it is moderate crack; when , it is severe crack; when , it is extremely serious crack.

[0048] (4) Visual display. In order to realize the visualization of the detection result, after the calculation and severity determination of the crack, the device superimposes the text description of the crack on the spliced image, thereby generating a complete road section image with crack annotation, intuitively marking the crack position and the pavement crack condition, facilitating the decision-making process of road maintenance personnel and engineers, improving the work efficiency and reducing the labor cost.

[0049] The movable portable complete road section pavement crack detection and diagnosis device based on dual spectrum provided by the application not only can greatly improve the detection efficiency and reduce the maintenance cost, but also can provide strong support for the timely maintenance of the road, so as to effectively guarantee the driving safety and promote the smoothness and stability of the road transportation.

[0050] The above is only the preferred specific embodiment of the application, but the protection scope of the application is not limited to this, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the application, which should be covered in the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.

Claims

1. A dual-spectrum based road surface crack detection apparatus, characterized by, The device comprises a device design module, a dual-spectrum fusion detection module, and a road surface quality evaluation module. The device design module comprises a camera module, a support structure, an inclination monitoring module, an image processing unit, and a mobile device. The dual-spectrum fusion detection module comprises a heat diffusion enhanced dual-spectrum multi-scale feature fusion image preprocessing module, which denoises and performs multi-scale feature fusion on images collected by an infrared thermal imaging camera and a visible light camera, outputs detection results, categories, and location information through an unsupervised enhanced BiSeNet hybrid wavelet transform network to complete detection. The road surface quality evaluation module detects road surface cracks through image stitching and BiSeNet network segmentation, evaluates road surface quality through crack distribution map generation and area ratio calculation, and displays detection results through visual labeling.

2. The dual-spectrum-based road surface crack detection device according to claim 1, wherein the camera module comprises a visible light camera and an infrared thermal imaging camera. The inclination monitoring module is a digital inclinometer. The image processing unit adopts an edge computing device and has a visual screen to display detection conditions in real time. The mobile device includes an adjustable knob and a detachable small wheel.

3. The dual-spectrum-based road surface crack detection device according to claim 1 or 2, wherein the heat diffusion enhanced dual-spectrum multi-scale feature fusion image preprocessing module comprises: Complex crack non-local mean filtering based on similarity principle is performed on infrared spectrum images and visible spectrum images. Only the filtered infrared spectrum images are subjected to crack feature Sigmoid transformation based on a logic function to enhance the features of the infrared thermal images. The processed infrared spectrum images and visible spectrum images are subjected to heat diffusion processing based on a Laplacian operator for pseudo-three-dimensional data processing to enhance image features. Based on multi-scale image fusion, dual-spectrum image Gaussian pyramid fusion is performed to unify the crack size in visible light and infrared spectrum images, implement weighted fusion, and strengthen the visibility of cracks.

4. The dual-spectrum-based road surface crack detection device according to claim 3, wherein the heat diffusion processing performs iterative calculation on each pixel point to smooth high-frequency noise and details in the image, thereby enhancing the prominence of crack features.

5. The dual-spectrum-based road surface crack detection device according to claim 1, wherein the unsupervised enhanced BiSeNet hybrid wavelet transform network integrates a fully connected layer and a feature extraction module based on wavelet transform by mixing DIM unsupervised learning and BiSeNet dual-path network.

6. The dual-spectrum-based road surface crack detection device according to claim 1 or 5, wherein the detection is completed by the unsupervised enhanced BiSeNet hybrid wavelet transform network outputting detection results, categories, and location information, comprising: ​ ​ ​ ​ ​ ​ constructing a dataset containing unannotated images for unsupervised pre-training and a small amount of annotated images for fine-tuning the segmentation model; the annotated images contain crack segmentation labels indicating the presence or absence of cracks in the image and crack classification labels distinguishing between crack types; based on DIM unsupervised task learning, using unannotated images for DIM unsupervised task learning training to obtain pre-trained encoder parameters; using the DIM pre-trained encoder as the initialization weight of BiSeNet for transfer learning, transferring parameters to the spatial path and context path of the BiSeNet network; integrating the HWD feature extraction module based on Haar wavelet transform into the spatial path of the BiSeNet network to replace the preset stride convolution, enhancing the feature learning ability and image detail capture; fine-tuning the model through the annotated images; predicting the type of each crack region through the fully connected layer; evaluating the model through key indicators; integrating the trained model into the image processing unit.

7. The dual-spectrum-based road crack detection device according to claim 6, wherein the fine-tuning of the model through the annotated images includes using a loss function suitable for the segmentation task to optimize the performance of the model.

8. The dual-spectrum-based road crack detection device according to claim 1, wherein the image stitching includes collecting road segment images and retaining a 10% overlap for seamless stitching, and dividing the stitched images into blocks of a predetermined size to preserve crack details.

9. The dual-spectrum-based road crack detection device according to claim 8, wherein the BiSeNet network segmentation includes: inputting each divided block into the unsupervised enhanced BiSeNet hybrid network to generate crack segmentation masks and classification labels, the segmentation masks representing binary masks of crack regions, and the classification labels being used to distinguish between small cracks, large cracks, and sinkhole cracks; constructing a complete road segment crack distribution map by stitching the segmentation masks, and positioning the crack location and level in combination with the classification labels.

10. The dual-spectrum-based road crack detection device according to claim 9, wherein the road surface quality evaluation through the crack distribution map and the area ratio calculation includes: based on the segmentation masks and the classification labels, calculating the crack area ratio and assigning different weights to different crack types, and evaluating the severity of the cracks through the weighted ratio.

Images