Lightweight road damage detection method based on RT-DETR improvement

Abstract

The invention provides a lightweight road damage detection method based on RT-DETR improvement, and the method comprises the steps: obtaining road damage image data, and dividing a data set; a backbone network of the RT-DETR model is modified to be composed of a plurality of multi-scale edge information enhancement modules, a hybrid encoder of the RT-DETR model is modified to be composed of an attention-based intra-scale feature interaction module and a multi-branch cavity convolution pyramid network module, and an RDD-DETR model is obtained after modification; and the RDD-DETR model is trained and is used for road damage detection in the road image data. According to the invention, the end-to-end multi-scale feature extraction and fusion network is used for road damage detection, so that the road damage detection accuracy and detection speed can be greatly improved.

Description

Technical Field

[0001] The invention belongs to the technical field of road monitoring, and in particular relates to a lightweight road damage detection method based on RT-DETR improvement. Background Art

[0002] With the acceleration of urbanization, the scale and complexity of road networks have achieved explosive growth. However, with the rapid expansion of road scale, the problem of road damage has become increasingly serious. Damage such as cracks and potholes not only affects the appearance and service life of the road, but also poses a serious threat to the safety of vehicles and passengers.

[0003] In the field of road monitoring technology, road damage detection is crucial to ensuring road safety and performance. Traditional road damage detection methods mainly include (1) manual visual inspection, (2) detection by detection vehicles equipped with high-precision sensors, and (3) detection based on image processing technology.

[0004] The manual visual inspection method is highly dependent on the professional knowledge and practical experience of the inspectors and is significantly affected by environmental factors. This method has low detection efficiency and a high missed detection rate. It cannot meet the needs of the highway inspection field for large-scale and high-efficiency inspection, which seriously limits its widespread application in the highway inspection field.

[0005] The high-precision sensors carried by special inspection vehicles usually include ground-penetrating radars, laser scanners, etc. Laser scanning technology has strict requirements on environmental conditions. Factors such as changes in light intensity and the presence of obstructions will have a significant impact on the scanning results. It can only roughly reflect the deformation trend of the road surface, and it is difficult to provide accurate deformation data, and it cannot meet the needs of accurate assessment of road conditions. The use of ground-penetrating radar technology to detect cracks, potholes and other damage on roads has the advantages of non-destructive testing, high detection efficiency and high accuracy. However, this technology has many limitations in practical applications, such as the inability to realize the automatic classification of road cracks, serious interference from surrounding environmental clutter, and great impact when detecting in inhomogeneous media. At the same time, the use and maintenance costs of this technology are high, and it still requires manual assistance. It faces great difficulties in the promotion and application of daily road inspection work.

[0006] Traditional detection methods based on image processing mainly use algorithms such as wavelet transform and threshold segmentation to extract road damage features. Such methods mainly rely on artificially constructed features for detection. When road damage features change due to various factors (such as differences in road surface materials, diversity of damage types, etc.), the detection performance of these algorithms will drop sharply. In addition, this method is easily interfered by complex background factors such as water stains, shadows, and color differences, and has poor anti-interference ability during the detection process, which seriously restricts the further improvement of road damage recognition accuracy. Due to its lack of robustness and generalization ability, it is difficult to effectively cope with the complex challenges of road damage detection in different actual environments and cannot meet diverse detection needs.

[0007] In order to solve the above problems, deep learning technology has been widely used in the field of road damage detection, significantly improving the accuracy and efficiency of detection. For example, two-stage target detection algorithms based on R-CNN, Faster R-CNN, Mask R-CNN, and single-stage target detection algorithms based on SSD and YOLO, as well as the real-time end-to-end target detection model RT-DETR based on the Transformer architecture, etc. Deep learning has shown great advantages in the field of road damage detection and is expected to completely replace traditional road damage detection methods. However, there are many types of road damage, including cracks, potholes, ridges, etc., which show complex and changeable characteristics in shape, size and color. The current deep learning model still has problems such as difficulty in multi-scale road damage recognition and large number of high-precision model parameters that make it difficult to apply on resource-constrained devices. Summary of the invention

[0008] The present invention proposes a lightweight road damage detection method based on RT-DETR improvement, which uses an end-to-end multi-scale feature extraction and fusion network for road damage detection, which can greatly improve the detection accuracy and speed.

[0009] To achieve the above object, the technical solution of the present invention is achieved as follows: A lightweight road damage detection method based on RT-DETR improvement, comprising: S1, obtaining road damage image data and dividing the data set; S2. The backbone network of the RT-DETR model is modified to be composed of multiple multi-scale edge information enhancement modules to extract multi-scale features of the input image data; the hybrid encoder of the RT-DETR model is modified to be composed of an attention-based intra-scale feature interaction module and a multi-branch dilated convolutional pyramid network module, the attention-based intra-scale feature interaction module AIFI processes the multi-scale features, and the multi-branch dilated convolutional pyramid network module fuses the multi-scale features to convert features of different levels into a sequence of image features; the decoder adopts an IoU-based query selection mechanism to select a group of image features from a sequence of image features output by the modified hybrid encoder as the initial object query, and generates a prediction box and a confidence score through iterative optimization of the auxiliary prediction head; the RDD-DETR model is obtained after modification; S3. Input the training set of the divided road damage image data into the RDD-DETR model for training, adjust the model through the validation set, and evaluate the model performance through the test set; the trained model is used for road damage detection in road image data.

[0010] Furthermore, in step S2, the method for constructing the backbone network includes: alternatingly arranging a plurality of multi-scale edge information enhancement modules and a plurality of convolutional layers.

[0011] Furthermore, the multi-scale edge information enhancement module realizes multi-scale edge information enhancement through multi-scale pooling, edge enhancement and feature fusion. The multi-scale pooling is to extract multi-level features by performing adaptive average pooling of different scales on the input feature map through adaptive average pooling operation. The edge enhancement is to strengthen the edge details of each level of features through an independent EdgeInfoEnhancer module, adaptively enhance the edge information of the input feature map, and obtain multi-scale edge features. The feature fusion is to splice the original local features with the multi-scale edge features in the channel dimension to form a unified feature.

[0012] Furthermore, the multi-scale edge information enhancement module also introduces the HiLo Attention mechanism. The high-frequency attention Hi-Fi enhances the ability to capture edge details, and the low-frequency attention Lo-Fi optimizes the perception of the global structure.

[0013] Preferably, the EdgeInfoEnhancer module first uses average pooling to locally smooth the input feature map to extract low-frequency information, then subtracts the smoothed feature map from the input feature map to obtain high-frequency information, and then adjusts the high-frequency information weights of different channels through the convolution kernel. Finally, the enhanced edge is superimposed on the input feature map in the form of a residual.

[0014] Furthermore, in step S2, the backbone network outputs a shallow feature map S3, a middle feature map S4, and a deep feature map S5; the attention-based intra-scale feature interaction module AIFI processes the deep feature map S5, performs intra-scale interaction, and applies a self-attention mechanism to the deep features rich in semantic information; the multi-branch atrous convolutional pyramid network module fuses the shallow feature map S3, the middle feature map S4, and the features processed by AIFI.

[0015] Furthermore, the multi-branch dilated convolutional pyramid network module is composed of a parallel upsampling module, a parallel downsampling module, and a local multi-scale dilated convolution module PMAC. The parallel upsampling module and the parallel downsampling module provide multiple feature extraction paths, and the local multi-scale dilated convolution module PMAC realizes multi-scale feature extraction through multi-path dilated convolution.

[0016] Preferably, the parallel upsampling module sets two upsampling paths, one for upsampling through transposed convolution, and the other for upsampling through linear interpolation, and the outputs of the two paths are spliced; the spliced ​​feature map is multiplied by the attention weight, and the number of channels of the fused feature map is adjusted through a convolution layer to obtain the final upsampling feature map; the attention weight is generated by a global pooling operation and a HardSigmoid activation function.

[0017] Preferably, the parallel downsampling module sets two downsampling paths, one for downsampling through convolution, and the other for downsampling through maximum pooling, and the outputs of the two paths are spliced; the spliced ​​feature map is multiplied by the attention weight, and the number of channels of the fused feature map is adjusted through a convolution layer to obtain the final downsampled feature map; the attention weight is generated by a global pooling operation and a HardSigmoid activation function.

[0018] Preferably, the local multi-scale dilated convolution module PMAC includes a multi-scale dilated convolution module MAC, and a convolution layer is introduced to independently extract features, and the features are fused with the output of MAC; the multi-scale dilated convolution module MAC captures features of different scales by processing dilated convolutions with different expansion rates in parallel, and then splices the features of different scales.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention proposes a multi-scale edge information enhancement structure in the backbone network part to improve the feature extraction quality and edge information representation capability. Through the three core functions of multi-scale feature extraction, edge enhancement and feature fusion, the model significantly enhances the perception ability of image features and can effectively handle target detection tasks that require precise edge information.

[0020] 2. The present invention proposes a multi-branch dilated convolutional pyramid network structure in the design of the neck encoding network, which improves the diversity and effectiveness of feature expression through the combination of multi-branch sampling, feature selection and dilated convolution.

[0021] 3. The present invention realizes lightweight model, and has the practicality and efficiency of actual scene application when used for road damage detection, thereby enhancing the real-time and accuracy of road damage detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a schematic diagram of a flow chart of an embodiment of the present invention; Figure 2 is a schematic diagram of the structure of an improved RDD-DETR model according to an embodiment of the present invention; Figure 3 is a schematic diagram of the structure of a multi-scale edge information enhancement module (MEIE) according to an embodiment of the present invention; Figure 4 is a schematic diagram of the structure of an EdgeInfoEnhancer module according to an embodiment of the present invention; Figure 5 is a schematic diagram of the structure of a parallel upsampling module according to an embodiment of the present invention; Figure 6 is a schematic diagram of the structure of a parallel downsampling module according to an embodiment of the present invention; Figure 7 is a schematic diagram of the structure of a PMAC module according to an embodiment of the present invention; Figure 8 The detection effect of the embodiment of the present invention is shown in FIG. Figure 1 ; Fig. 9 The detection effect of the embodiment of the present invention is shown in FIG. Figure 2 ; Fig.10 The detection effect of the embodiment of the present invention is shown in FIG. Figure 3 ; Fig.11 The detection effect of the embodiment of the present invention is shown in FIG. Figure 4 . DETAILED DESCRIPTION

[0023] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0024] In order to make the purpose and features of the present invention more obvious and understandable, the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0025] This embodiment applies the present invention to the UAV road damage detection scenario to solve the problems of low detection accuracy and insufficient real-time performance in the prior art. The specific method is as follows: Figure 1 As shown, including: S1. Obtain road damage image data collected by drones and divide the data set; S2. Modify the backbone network of the RT-DETR model to consist of multiple multi-scale edge information enhancement modules, and modify the hybrid encoder of the RT-DETR model to consist of an attention-based intra-scale feature interaction module and a multi-branch dilated convolutional pyramid network module. After modification, the RDD-DETR model is obtained and deployed on the drone; S3. Input the training set of the divided road damage image data into the RDD-DETR model for training, adjust the model through the validation set, and evaluate the model performance through the test set; the trained model is used for UAV to collect road image data for road damage detection.

[0026] In this embodiment, the data set division described in step S1 means that after obtaining the road damage image data collected by the drone, the image data is preprocessed and labeled, and then the image data is divided into a training set, a test set, and a validation set in a ratio of 7:2:1 or 6:2:2.

[0027] In step S2, the RT-DETR model is modified to the RDD-DETR model. The structure of the modified RDD-DETR model is as follows: Figure 2 The modification process is as follows: The backbone network is modified to consist of multiple multi-scale edge information enhancement modules (MEIE), which are alternately arranged with multiple convolutional layers Conv. This alternating arrangement structure can gradually go deeper into the image data and capture feature information at different levels. The backbone network extracts multi-scale feature representations from the original image and outputs feature maps of three stages (shallow feature map s3, middle feature map s4, and deep feature map s5), which are input into the hybrid encoder of the neck encoding network. The hybrid encoder consists of an attention-based intrascale feature interaction module (AIFI) and a multiple atrous convolution pyramid network module (MAPN). AIFI processes the deep feature map s5 from the backbone network, performs intra-scale interaction only on the feature layer of S5, and applies the self-attention mechanism to the deep features rich in semantic information, which can significantly reduce the computational cost. MAPN fuses features of various scales, including shallow feature maps S3, mid-level feature maps S4, and features processed by AIFI. Features at different levels are converted into a sequence of image features to enhance the accuracy of target recognition. The decoder of the decoding prediction network adopts an IoU-based query selection mechanism to select a set of image features from the sequence output by the encoder as the initial object query, which improves the quality of the decoder's initial object query and generates prediction boxes and confidence scores through iterative optimization of the auxiliary prediction head, thereby improving the detection performance of the decoder.

[0028] The multi-scale edge information enhancement module MEIE proposed in this embodiment is used to improve the problem of insufficient feature extraction capability of the original backbone network. The MEIE module realizes multi-scale edge information enhancement through three parts: multi-scale pooling, edge enhancement and feature fusion.

[0029] In this embodiment, the specific structure of the MEIE module is as follows: Figure 3As shown in the figure, first, the information of different scales corresponding to different branches is adaptively averaged by the Adaptive Average Pooling operation, and the input feature map is adaptively averaged by different scales to extract multi-level features. This pooling method can automatically adjust the size of the pooling kernel according to the preset target size, so as to extract local information at different resolutions, thereby capturing multi-level features in the image, while taking into account global and local information. Then, the feature map is reduced in dimension through two convolutional layers Conv, which reduces the amount of calculation while separating feature channels of different scales and extracting local features. Subsequently, the feature map of each branch is restored to the original input size through linear interpolation Upsample to ensure that the multi-scale features can be spliced, and then the edge details are enhanced through an independent EdgeInfoEnhancer module, which adaptively enhances the edge information of the input feature map, helping the model to capture more detailed features. Finally, the original local features after a convolutional layer are spliced ​​with the multi-scale edge features enhanced by EdgeInfoEnhancer in the channel dimension to form a unified feature representation. In addition, the MEIE module also introduces HiLo Attention, so that the multi-scale edge information enhancement module can process multi-scale features and edge information more efficiently. High-frequency attention (Hi-Fi) enhances the ability to capture edge details, and low-frequency attention (Lo-Fi) optimizes the perception of global structures. The combination of the two makes the fusion of multi-scale features more efficient, which not only improves the model's sensitivity to edge information, but also significantly improves computational efficiency.

[0030] The EdgeInfoEnhancer module (edge ​​information enhancement module) is set in the MEIE module because edge information plays a key role in the road damage detection task. Clear edge information helps to locate the target boundary more accurately. The EdgeInfoEnhancer module is used to enhance edge features. Edge information usually contains important details such as the outline, texture and shape of objects in the image. By enhancing this information, the network can more sensitively capture the edge features in the image. The EdgeInfoEnhancer module structure is as follows Figure 4 As shown in the figure, first use the average pooling AvgPool to locally smooth the input feature map and extract the low-frequency information. Then, subtract the smoothed feature map from the original feature map to obtain the high-frequency information, i.e., edges, textures and other details. Then, the convolutional layer Conv is used to adjust the weights of the high-frequency information of different channels. Finally, the enhanced edges are superimposed on the original input in the form of residuals, which not only retains the original features but also strengthens the edge details.

[0031] The multi-branch dilated convolutional pyramid network MAPN in the hybrid encoder of the neck coding network is also one of the improvements to the RT-DETR model. The multi-branch dilated convolutional pyramid network module MAPN is composed of a parallel upsampling module, a parallel downsampling module, and a local multi-scale dilated convolution module PMAC.

[0032] The parallel upsampling module is as follows Figure 5 As shown in the figure, two different upsampling paths are used to extract features: one is upsampling achieved by transposed convolution ConvTranspose, and the other is upsampling completed by linear interpolation Upsample reconvolution. These two paths work in parallel to capture important information in the input feature map from different angles. At the same time, the module generates attention weights through global pooling operation GlobalPool and HardSigmoid activation function, which reflect the importance of each channel. In the feature fusion stage, the outputs of the two upsampling paths are first concat-concatenated in the channel dimension to form a richer feature representation. The concatenated feature map is multiplied by the attention weight calculated by the previous parallel upsampling module. This multiplication operation realizes the weighted fusion of features based on attention weights, thereby strengthening those features that are more important to the task and suppressing irrelevant or redundant features. Finally, the number of channels of the fused feature map is adjusted through a convolution layer to obtain the final upsampled feature map.

[0033] Similar to the upsampling module, the parallel downsampling module is Figure 6 As shown in the figure, two parallel downsampling paths are also used to extract features: one is downsampling achieved by convolution with a stride of 2, and the other is downsampling completed by reconvolution with maximum pooling MaxPool. These two paths are also executed in parallel to capture key information in the input feature map from different angles. At the same time, the parallel downsampling module also generates attention weights through the global pooling operation GlobalPool and the HardSigmoid activation function. In the feature fusion stage, the outputs of the two downsampling paths are first concatenated (Concat) in the channel dimension to form a richer feature representation. Then, these concatenated feature maps are multiplied by the attention weights calculated by the previous parallel downsampling module. This multiplication operation also realizes feature weighted fusion based on attention weights, strengthening important features and suppressing irrelevant features. Finally, the number of channels of the fused feature map is adjusted through a convolution layer to obtain the final downsampled feature map.

[0034] The multi-branch atrous convolutional pyramid network module MAPN provides multiple feature extraction paths for the network through parallel upsampling modules and parallel downsampling modules. This multi-path design enriches the diversity of feature expression, enabling the network to capture target information from different scales and levels, and extract diverse features through different sampling strategies such as transposed convolution, interpolation, and pooling. In addition, it also introduces a gating mechanism to selectively enhance the sampled features. By learning the importance weights of the features, the gating mechanism can strengthen the features related to road damage and suppress redundant features, thereby improving the effectiveness of feature expression. The parallel upsampling module and the parallel downsampling module correspond to Figure 2 U and D in .

[0035] The local multi-scale atrous convolution module PMAC (Partial Multiple Atrous Convolution) module is one of the core structures of the multi-branch atrous convolution pyramid network module. Figure 7 As shown in the figure, multi-scale feature extraction is achieved through multi-way atrous convolution. In PMAC, in addition to multi-way atrous convolution, an additional convolution layer Conv is introduced. This convolution layer can extract features independently, and then concat the output of the multi-scale atrous convolution module MAC to further enhance the expressiveness of the features, and can reduce the amount of computation while improving performance, which has significant advantages for resource-constrained devices or application scenarios that require fast inference. Atrous convolution expands the receptive field by introducing a dilation rate in the convolution kernel, allowing the network to capture a wider range of contextual information without increasing the amount of computation. The multi-scale atrous convolution module MAC (Multiple Atrous Convolution) in PMAC uses atrous convolutions with different dilation rates, so that local details and global contextual information can be extracted simultaneously.

[0036] The MAC module structure is as follows Figure 7As shown in the figure, features of different scales are captured by processing AtrousConv with different dilation rates in parallel. Atrous convolution branches with different dilation rates can capture features of different scales. A larger dilation rate can capture a wider range of contextual information, while a smaller dilation rate can retain more detailed information. This multi-scale feature extraction capability enables the MAC module to better understand the local and global information in the input feature map. The outputs of the atrous convolution branches with different dilation rates and the output of the convolution layer are concatenated in the channel dimension. This concatenation operation fuses features of different scales together to form a richer feature representation. This feature fusion helps to retain feature information of different scales and improve the model's ability to recognize complex patterns. Finally, the concatenated feature map is processed by a convolution layer Conv to generate the final output feature map. This convolution layer can be used to adjust the number of channels so that the number of channels of the output feature map matches that of the subsequent layers. In the road damage detection task, small target damage usually requires more refined local features, while large target damage requires more extensive contextual information. The MAC module can meet both requirements at the same time through parallel multi-scale feature extraction, significantly improving the network's ability to represent complex features.

[0037] The improved RDD-DETR model is deployed on the drone and trained using the divided data set. The trained model is used for the drone to collect road image data for road damage detection, which can significantly improve the efficiency of road damage detection.

[0038] Figure 8 , Fig. 9 , Fig.10 , Fig.11 The performance of the improved RDD-DETR model is shown in the detection of some images in the public dataset UAPD (UAV image dataset for detecting road cracks). Figure 8-11 The Longitudinal crack, Transverse crack, Oblique crack, Alligator crack, and Repair represent the damage types. Longitudinal crack is a longitudinal crack, Transverse crack is a transverse crack, Oblique crack is an oblique crack, and Alligator crack is a reticular crack. Repair means repair, and the number after the damage type indicates the detection accuracy. Figure 8-11 It can be seen from the display that the present invention has optimized the accuracy and speed of road damage image recognition, and can identify road damage faster and more accurately, thereby improving the accuracy of the detection results and meeting the needs of drones for real-time detection of road damage.

[0039] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A lightweight road damage detection method based on RT-DETR improvement, characterized in that: include: S1. Obtain road damage image data and divide the data set; S2, modifying the backbone network of the RT-DETR model to consist of multiple multi-scale edge information enhancement modules to extract multi-scale features of the input image data; The hybrid encoder of the RT-DETR model is modified to consist of an attention-based intra-scale feature interaction module and a multi-branch dilated convolutional pyramid network module. The attention-based intra-scale feature interaction module AIFI processes the multi-scale features, and the multi-branch dilated convolutional pyramid network module fuses the multi-scale features to convert features at different levels into a sequence of image features. The decoder adopts an IoU-based query selection mechanism to select a set of image features from a sequence of image features output by the modified hybrid encoder as the initial object query, and generates prediction boxes and confidence scores through iterative optimization of the auxiliary prediction head; After modification, the RDD-DETR model is obtained; S3. Input the training set of the divided road damage image data into the RDD-DETR model for training, adjust the model through the validation set, and evaluate the model performance through the test set; the trained model is used for road damage detection in road image data.

2. The improved lightweight road damage detection method based on RT-DETR according to claim 1 is characterized in that: In step S2, the method for constructing the backbone network includes: alternatingly arranging a plurality of multi-scale edge information enhancement modules and a plurality of convolutional layers.

3. The improved lightweight road damage detection method based on RT-DETR according to claim 2 is characterized in that: The multi-scale edge information enhancement module realizes multi-scale edge information enhancement through three parts: multi-scale pooling, edge enhancement and feature fusion; the multi-scale pooling is to extract multi-level features by performing adaptive average pooling of different scales on the input feature map through adaptive average pooling operation; The edge enhancement is to strengthen the edge details of each level of features through an independent EdgeInfoEnhancer module, adaptively enhance the edge information of the input feature map, and obtain multi-scale edge features; the feature fusion is to splice the original local features with the multi-scale edge features in the channel dimension to form a unified feature.

4. The improved lightweight road damage detection method based on RT-DETR according to claim 3 is characterized in that: The multi-scale edge information enhancement module also introduces the HiLo Attention mechanism. The high-frequency attention Hi-Fi enhances the ability to capture edge details, and the low-frequency attention Lo-Fi optimizes the perception of the global structure.

5. The improved lightweight road damage detection method based on RT-DETR according to claim 3 or 4, characterized in that: The EdgeInfoEnhancer module first uses average pooling to locally smooth the input feature map to extract low-frequency information, then subtracts the smoothed feature map from the input feature map to obtain high-frequency information, and then adjusts the high-frequency information weights of different channels through convolution kernels. Finally, the enhanced edge is superimposed on the input feature map in the form of a residual.

6. The improved lightweight road damage detection method based on RT-DETR according to claim 1 is characterized in that: In step S2, the backbone network outputs a shallow feature map S3, a middle feature map S4, and a deep feature map S5; the attention-based intra-scale feature interaction module AIFI processes the deep feature map S5, performs intra-scale interaction, and applies a self-attention mechanism to the deep features rich in semantic information; the multi-branch atrous convolutional pyramid network module fuses the shallow feature map S3, the middle feature map S4, and the features processed by AIFI.

7. The improved lightweight road damage detection method based on RT-DETR according to claim 6 is characterized in that: The multi-branch dilated convolutional pyramid network module is composed of a parallel upsampling module, a parallel downsampling module, and a local multi-scale dilated convolution module PMAC. The parallel upsampling module and the parallel downsampling module provide multiple feature extraction paths, and the local multi-scale dilated convolution module PMAC realizes multi-scale feature extraction through multi-path dilated convolution.

8. The improved lightweight road damage detection method based on RT-DETR according to claim 7 is characterized in that: The parallel upsampling module sets two upsampling paths, one of which implements upsampling through transposed convolution and the other completes upsampling through linear interpolation, and the outputs of the two paths are spliced; the spliced ​​feature map is multiplied by the attention weight, and the number of channels of the fused feature map is adjusted through a convolution layer to obtain the final upsampling feature map; the attention weight is generated by a global pooling operation and a HardSigmoid activation function.

9. The improved lightweight road damage detection method based on RT-DETR according to claim 7 is characterized in that: The parallel downsampling module sets two downsampling paths, one for downsampling through convolution and the other for downsampling through maximum pooling, and the outputs of the two paths are spliced; the spliced ​​feature map is multiplied by the attention weight, and the number of channels of the fused feature map is adjusted through a convolution layer to obtain the final downsampled feature map; the attention weight is generated by a global pooling operation and a HardSigmoid activation function.

10. The improved lightweight road damage detection method based on RT-DETR according to claim 7, characterized in that: The local multi-scale dilated convolution module PMAC includes a multi-scale dilated convolution module MAC, and also introduces a convolution layer to independently extract features and fuse them with the output of MAC; the multi-scale dilated convolution module MAC captures features of different scales by processing dilated convolutions with different expansion rates in parallel, and then splices the features of different scales.

Images