CN-121999324-A - Intelligent pavement paving temperature uniformity detection method and system based on infrared thermal image

CN121999324ACN 121999324 ACN121999324 ACN 121999324ACN-121999324-A

Abstract

The invention relates to the technical field of road engineering intelligent detection, in particular to an infrared thermal image-based pavement paving temperature uniformity intelligent detection method and system, wherein in the paving construction process, an infrared thermal imager behind a paver is used for collecting original infrared thermal image data, abnormal region segmentation and overall temperature uniformity grade marking are completed, and a training sample set is constructed; the method comprises the steps of carrying out normalization and denoising processing on collected data, realizing self-adaptive region segmentation based on temperature gradients, extracting multi-dimensional statistical features and generating a region feature map, constructing a multi-source feature neural network, carrying out feature fusion and space weighting on the region feature map, denoising data and masks, integrating priori knowledge to realize abnormal region pixel level segmentation and temperature uniformity level assessment, and designing a multi-task cooperative loss function to complete model training and parameter optimization. The invention can realize real-time, accurate and intelligent detection of the pavement paving temperature distribution, and effectively improve the automation level of temperature uniformity evaluation and defect identification.

Inventors

SONG MIN
SU DI
XU GANGNIAN

Assignees

鄄城县公路事业发展中心
菏泽市智慧公路与应急抢险保障中心
山东交通学院

Dates

Publication Date: 20260508
Application Date: 20260211

Claims (10)

1. An intelligent detection method for pavement paving temperature uniformity based on infrared thermal imaging is characterized by comprising the following steps: S1, acquiring original infrared thermal image data through an infrared thermal imager arranged at the rear of a paver in the pavement paving construction process, and marking the acquired data, wherein the marking comprises abnormal region segmentation marking and overall temperature uniformity grade marking, so as to form a training sample set; S2, performing wavelet transformation on the original infrared thermal image data, and combining with soft threshold processing to obtain a wavelet denoising mask; S3, calculating a temperature gradient according to the normalized denoising data and the wavelet denoising mask, dynamically dividing regions according to the temperature gradient, performing self-adaptive region segmentation, and extracting multi-dimensional statistical features from each region to form a region feature map for local temperature distribution characteristics; S4, constructing a neural network model based on multi-source features, inputting a regional feature map, normalized denoising data and wavelet denoising masks, carrying out multi-source feature fusion and space weighting to obtain a fusion feature map, optimizing the fusion feature map by combining features of each region to obtain an optimized feature map, and then carrying out pixel level segmentation and overall temperature uniformity level assessment of a pavement paving abnormal region based on the optimized feature map; s5, constructing a multi-task cooperative loss function, and calculating a total loss function of the model; S6, training the model by using a training sample set, and optimizing model parameters by minimizing a multi-task cooperative loss function to obtain a trained model; and S7, using the trained model for intelligent detection of the paving temperature uniformity of the road surface, and performing intelligent detection on the paving temperature uniformity of the road surface in a new paving construction scene by the trained model.
2. The intelligent detection method for the pavement paving temperature uniformity based on the infrared thermal image according to claim 1, wherein the step S2 is specifically as follows: S2.1, performing two-dimensional discrete wavelet transformation on original infrared thermal image data to obtain a wavelet coefficient matrix containing low-frequency approximate coefficients and high-frequency detail coefficients, and then combining soft threshold processing operation and the two-dimensional discrete wavelet transformation to obtain a wavelet denoising mask; s2.2, the self-adaptive normalization function fuses sliding window statistics and wavelet threshold denoising strategies, calculates a local statistics matrix for the original infrared thermal image data, centers and scales the local statistics matrix, and then multiplies the local normalization data and the wavelet denoising mask element by element to obtain normalized denoising data.
3. The intelligent detection method for the pavement paving temperature uniformity based on the infrared thermal image, which is disclosed in claim 1, is characterized in that: s3 is specifically as follows: s3.1, combining the normalized denoising data and the wavelet denoising mask, calculating a temperature gradient by using a gradient operator, and obtaining a temperature gradient map by calculating the gradient amplitude of the normalized denoising data; s3.2, carrying out region division on the temperature gradient map by adopting an iterative threshold method and combining with the analysis of a communication component, dynamically determining a division threshold value to obtain a plurality of temperature uniform regions, and dividing to obtain a temperature region set; s3.3, extracting a plurality of statistical feature vectors from the normalized denoising data for each divided region, wherein the statistical feature vectors are used for comprehensively representing the temperature distribution characteristic of each region; and S3.4, mapping the statistical feature vector back to the image space to form a regional feature map, and coding the statistical feature of the region to which the statistical feature vector belongs at each spatial position.
4. The intelligent detection method for the pavement paving temperature uniformity based on the infrared thermal image according to claim 3, wherein the area division of the temperature gradient map is specifically as follows: each temperature region is a set of spatially connected pixels with lower temperature gradients for adaptively capturing temperature uniform regions of different sizes; Specifically, an iterative threshold method is adopted to traverse all possible gray threshold values of the temperature gradient map, and the gray threshold values are calculated to obtain The sum of intra-class variances of the segmented high-gradient class and low-gradient class is selected to minimize the sum of intra-class variances As an optimal segmentation threshold value, Representing an optimal segmentation threshold Generating a binary mask by comparing the temperature gradient map with an optimal segmentation threshold, wherein if the pixel value of the binary mask is 1, the pixel is represented as an edge pixel, and if the pixel value is 0, the pixel is represented as an intra-area pixel, and the pixel value in the binary mask is represented as Labeling the connected pixels to obtain each temperature region; And performing connectivity analysis on all pixels with the pixel value of 0 in the binary mask, and marking all mutually communicated pixel sets with the value of 0 as independent uniform temperature areas so as to obtain a temperature area set.
5. The intelligent detection method for the pavement paving temperature uniformity based on the infrared thermal image according to claim 1, wherein the step S4 is specifically as follows: S4.1, fusing the normalized denoising data, the regional feature map and the wavelet denoising mask, introducing spatial weight based on temperature gradient and regional area, and enhancing sensitivity of the feature to an abnormal region to obtain a fused feature map; S4.2, carrying out regional average pooling on the fusion feature map to obtain representative features of each region, then enhancing the contrast between the regions through differential operation, and obtaining an optimized feature map by combining the fusion feature map, wherein the optimized feature map is used for highlighting the consistency inside the regions and the difference between the regions; s4.3, adopting a dual-task output layer, and simultaneously generating a pixel-level abnormal segmentation map and an image-level temperature uniformity level probability vector, wherein the generation process of the abnormal segmentation map is modulated by regional characteristics, and the prediction of the temperature uniformity level probability vector is obtained based on the weighted global characteristics of the segmentation map.
6. The intelligent detection method for the pavement paving temperature uniformity based on the infrared thermal image according to claim 1, wherein the step S5 is specifically as follows: Defining a multi-task cooperative loss function comprising segmentation loss, grade classification loss and region consistency regularization loss, wherein the multi-task cooperative loss function is used for training a total objective function of a model, unifying a pixel grade segmentation task, an image grade classification task and a regularization constraint based on region prior into a frame in a weighted summation mode, guiding the model to learn and generate an accurate abnormal segmentation map and a correct overall temperature uniformity grade at the same time, and improving the segmentation rationality by utilizing a region consistency hypothesis; the abnormal segmentation loss adopts a calculation mode of a weighted binary cross entropy loss function and is used for processing class unbalance of an abnormal region and a background; The class classification loss is realized by adopting a calculation mode of a cross entropy loss function and is used for calculating the difference between a temperature uniformity class probability vector predicted by a model and a real class label; The region consistency regularization penalty is used to encourage consistency of the intra-region predicted segmentation values.
7. The intelligent detection method for the pavement paving temperature uniformity based on the infrared thermal image according to claim 1, wherein the step S6 is specifically as follows: Dividing a training sample set into a training set and a verification set, sequentially executing the operations of the steps S2 and S3 on input data to serve as input features of a neural network model, carrying out iterative optimization on the neural network model by adopting a gradient descent algorithm, randomly extracting batch samples from the training set in each iteration, inputting the batch samples into the neural network model based on multi-source features, carrying out forward propagation calculation to obtain a prediction output of an abnormal segmentation graph and a temperature uniformity level probability vector, then calculating a multi-task cooperative loss function according to a real segmentation label and a level label, then calculating the gradient of the loss function on each layer of parameters of the model by using a reverse propagation algorithm, and updating the model parameters by using an optimizer such as Adam to gradually reduce the loss value, improve the performance of the model on segmentation and classification tasks, and ending training when the judgment condition of iteration stopping is met in each iteration in the training process, and selecting the model parameters which are optimal on the verification set as training results to obtain a trained model.
8. The intelligent detection method for the pavement paving temperature uniformity based on the infrared thermal image according to claim 1, wherein the method is characterized in that the method S1 specifically comprises the following steps: S1.1, data acquisition: Setting a high-resolution thermal infrared imager at a fixed position or on a movable platform behind the paver, continuously or regularly shooting, and combining environmental factors and construction process factors in the acquisition process, wherein the environmental factors comprise wind speed sunlight and temperature, and the construction process factors comprise paving materials, speed and thickness; s1.2, data marking: (1) The abnormal region segmentation marking is pixel-level marking, a temperature abnormal region is manually sketched on the infrared thermal image data or marked by an auxiliary tool according to the temperature distribution characteristics by a person, a binary segmentation mask is generated as a real segmentation label, wherein the normal region is marked as 0, and the abnormal region is marked as 1; (2) The whole temperature uniformity grade label is an image grade label, a whole temperature uniformity grade label is manually distributed for each piece of infrared thermal image data, different labels respectively correspond to different temperature distribution quality grades, and the labels comprise uniformity, more uniformity and non-uniformity; S1.3, rechecking the marked data, and storing the marked data in a pairing way with the original infrared thermal image data to form a complete training sample set.
9. The intelligent detection method for the pavement paving temperature uniformity based on the infrared thermal image according to claim 1, wherein the step S7 is specifically as follows: After the trained model is detected, two parts of results are output, namely a pixel-level abnormal segmentation graph, wherein each pixel value represents the probability that the position belongs to a temperature abnormal region, binarization is carried out through a set threshold value, abnormal spots with too low, too high or uneven temperature on a paving road surface are visually identified, an image-level temperature uniformity grade probability vector is converted into probability belonging to each grade through a Softmax function, the grade with the highest probability is taken as an overall uniformity evaluation result, and feedback and decision support are provided for construction quality control based on the output result.
10. An infrared thermal image-based pavement paving temperature uniformity intelligent detection system for executing the infrared thermal image-based pavement paving temperature uniformity intelligent detection method as set forth in any one of claims 1 to 9, characterized by comprising the following modules: the data acquisition labeling module is used for acquiring infrared thermal image data at the rear of the paver and finishing labeling of abnormal areas and temperature uniformity grades to construct a training sample set; the data preprocessing module is used for carrying out wavelet transformation, soft threshold processing and self-adaptive normalization on the data to obtain a wavelet denoising mask and normalized denoising data; the regional characteristic extraction module is used for realizing self-adaptive regional segmentation based on the temperature gradient, extracting multidimensional statistical characteristics and generating a regional characteristic diagram; Constructing a multi-source characteristic neural network, carrying out fusion weighting on the regional characteristic map, the denoising data and the mask, and combining priori knowledge to realize abnormal regional segmentation and uniformity grade assessment; The model training optimization module is used for completing model training and parameter optimization by adopting a multi-task cooperative loss function; And the intelligent detection execution module is used for deploying the trained model to a construction scene and performing intelligent detection on the paving temperature uniformity of the road surface.

Description

Intelligent pavement paving temperature uniformity detection method and system based on infrared thermal image Technical Field The invention relates to the technical field of intelligent detection of road engineering, in particular to an intelligent detection method and system for pavement paving temperature uniformity based on infrared thermal images. Background Along with the continuous improvement of the construction scale and quality requirements of highway engineering, the temperature control in the paving construction process of the pavement has become one of key factors influencing the structural performance and the service life of the pavement. The asphalt mixture is highly sensitive to temperature in the paving and rolling processes, and uneven temperature distribution can directly cause quality problems such as insufficient compactness, segregation, early cracking, reduced durability and the like. Therefore, the method and the device can detect and evaluate the paving temperature uniformity of the road in real time and accurately in the construction process, and have important significance for improving the construction quality control level, reducing reworking and prolonging the service life of the road. The infrared thermal imaging technology is gradually applied to the field of pavement temperature monitoring due to the advantages of non-contact, full-range imaging, visual reflection of pavement surface temperature distribution and the like. The prior art objectively has the defects that the prior paving temperature detection method based on infrared thermal images mostly adopts a global normalization or simple filtering denoising strategy, is difficult to consider local temperature mutation and random noise suppression, is easy to cause key temperature gradient information loss and influence the recognition accuracy of abnormal areas, is generally mainly based on fixed window or single-scale pixel characteristics, lacks self-adaptive modeling of temperature distribution area structures, cannot effectively characterize statistical characteristics of temperature uniform areas with different scales, and is insufficient in adaptability to complex temperature distribution scenes, a definite space significance modeling mechanism is not generally introduced in the feature fusion process of the prior method, is difficult to carry out targeted feature enhancement on high temperature gradient areas or large-area abnormal areas, has limited sensitivity to temperature unevenness problems, is difficult to treat abnormal area detection and overall temperature uniformity evaluation as independent tasks, lacks information coordination and constraint among tasks, and is difficult to simultaneously ensure the accuracy of local abnormal recognition and the consistency and reliability of overall uniformity evaluation results. Therefore, the invention provides an intelligent detection method and system for the pavement paving temperature uniformity of an infrared thermal image to solve the problems. Disclosure of Invention Aiming at the defects of the prior art, the invention provides the intelligent detection method and the system for the pavement paving temperature uniformity of the infrared thermal image. On the one hand, the technical scheme for solving the technical problem is that the intelligent detection method for the pavement paving temperature uniformity of the infrared thermal image comprises the following steps: S1, acquiring original infrared thermal image data through an infrared thermal imager arranged at the rear of a paver in the pavement paving construction process, and marking the acquired data, wherein the marking comprises abnormal region segmentation marking and overall temperature uniformity grade marking, so as to form a training sample set; S2, performing wavelet transformation on the original infrared thermal image data, and combining with soft threshold processing to obtain a wavelet denoising mask; S3, calculating a temperature gradient according to the normalized denoising data and the wavelet denoising mask, dynamically dividing regions according to the temperature gradient, performing self-adaptive region segmentation, and extracting multi-dimensional statistical features from each region to form a region feature map for local temperature distribution characteristics; S4, constructing a neural network model based on multi-source features, inputting a regional feature map, normalized denoising data and wavelet denoising masks, carrying out multi-source feature fusion and space weighting to obtain a fusion feature map, optimizing the fusion feature map by combining features of each region to obtain an optimized feature map, and then carrying out pixel level segmentation and overall temperature uniformity level assessment of a pavement paving abnormal region based on the optimized feature map; s5, constructing a multi-task cooperative loss function, and calculating a total loss functio