An enhanced image processing method for digital radiographic inspection of pressure pipelines with insulation layers
By using enhanced image processing methods and deep neural network models in digital ray detection with insulation layer pressure pipelines, the problem of insufficient detection accuracy is solved and higher damage recognition accuracy and efficiency is achieved.
Patent Information
- Application Number
- CN202411792469.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-07
- Publication Date
- 2025-04-01
AI Technical Summary
It is difficult for the prior art to effectively detect weld damage in pressure pipes with insulation layer. Conventional detection methods are affected by insulation layer, and the digital ray detection image quality is insufficient, resulting in low detection accuracy.
The enhanced image processing method is adopted to identify and remove insulation layer interference through the distance intensity model, and combine convolutional neural networks and long-term memory networks to extract local features and spatiotemporal features of the image to improve the accuracy of damage recognition.
Effectively reduce the impact of the insulation layer and waterproof layer on the detection image, improve the identification ability and accuracy of internal defects in the pressure pipeline with the insulation layer, realize rapid detection without stopping, save detection costs, and improve detection efficiency.
Smart Images
Figure CN119741267A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of pressure pipeline detection, and more particularly to an enhanced digital ray detection image processing method for a pressure pipeline with a thermal insulation layer. Background Art
[0002] Pressure pipelines are important transmission equipment in production and life. They usually contain high-temperature or ultra-low-temperature, high-pressure, flammable, explosive, toxic and other dangerous fluids. Their safe operation is of great significance. Among them, pressure pipelines used to transmit high-temperature or ultra-low-temperature fluids are usually covered with insulation layers.
[0003] During operation, pressure pipelines are easily damaged by the long-term chemical and electrochemical effects of the internal transmission fluid and the external environment medium. In particular, the connecting welds between pressure pipelines are affected by the difference in chemical composition between the welding material and the pipeline base material, the concentrated stress at the weld, the microstructural changes of the weld and surrounding materials caused by the hot zone during welding, and the weld unevenness caused by welding process problems, making the weld a weak link in the pressure pipeline. However, conventional ultrasonic, pulsed eddy current and other detection methods are limited by the thick insulation layer and it is difficult to effectively detect pressure pipelines covered with insulation layers; although digital X-ray detection technology can penetrate the insulation layer, its detection image will still be affected and interfered by the insulation layer, resulting in insufficient accuracy in the detection of damage and defects.
[0004] Therefore, how to improve the accuracy of damage detection of pressure pipes with insulation layers is an urgent problem to be solved by technical personnel in this field. Summary of the invention
[0005] In view of this, the present invention provides an enhanced image processing method for digital radiographic inspection of pressure pipelines with insulation layers. Aiming at the problem that the image quality in digital radiographic inspection of pressure pipelines with insulation layers is affected by the insulation layer, an enhanced image processing method is developed, which can identify and remove the interference of the insulation layer on the image quality, thereby improving the accuracy of damage and defect identification.
[0006] In order to achieve the above object, the present invention provides the following technical solutions:
[0007] The present invention discloses an enhanced digital ray detection image processing method for a pressure pipeline with a thermal insulation layer, and the specific steps are as follows:
[0008] Determine the weld position and pipeline parameters of each pipe section according to the pipeline construction data;
[0009] Collect multiple initial digital radiographic inspection images along the circumferential direction of the pipeline at the weld position;
[0010] According to the pipeline parameters corresponding to each weld position, and the geometric position relationship between the radiation source, the detector array and the pipeline axis, the distance of the detection signal received by each detector in the detector array through the thermal insulation material layer and the waterproof layer is determined, and the interference intensity of each detector is determined according to the distance intensity model, and the interference map of each weld position is further determined;
[0011] Aligning and subtracting the multiple initial digital ray detection images at each weld position with the corresponding interference images to obtain a digital ray detection atlas at each weld position;
[0012] Each digital radiographic detection image sequence in the digital radiographic detection atlas is input into a damage identification model to determine the damage condition of the corresponding weld position.
[0013] Furthermore, the pipeline parameters include: pipeline outer diameter, pipeline thickness, insulation material thickness, insulation material material, insulation material density, waterproof layer thickness, and waterproof layer material.
[0014] Furthermore, the distance strength model includes a first sub-model and a second sub-model, which are respectively used to determine the absorption strength of the detection signal passing through the thermal insulation material and the signal absorption strength of the waterproof layer; the first sub-model and the second sub-model are both BP neural network models;
[0015] The first sub-model and the second sub-model are trained based on the first sample set and the second sample set respectively; the sample characteristics of the samples in the first sample set include: detection signal strength, frequency, path distance, density and material of the thermal insulation material, and the sample label is the signal strength received by the detector; the sample characteristics of the samples in the second sample set include: detection signal strength, frequency, path distance and material of the waterproof material, and the sample label is the signal strength received by the detector.
[0016] Furthermore, the BP neural network model is trained based on an improved chromosome algorithm, specifically including:
[0017] Step 1: randomly generate an initial population, where each individual in the initial population represents the weight and threshold of each neuron in the neural network, and the weight and threshold are encoded as chromosomes;
[0018] Step 2, calculate the fitness value of each individual in the current population;
[0019] Step 3: Determine whether the stopping condition is met. If so, stop the iteration, decode the chromosome of the individual with the smallest fitness value, and output the neural network weight and threshold; otherwise, execute step 4.
[0020] Step 4: Perform chromosome screening in the current population and remove individuals whose selection probability is lower than the threshold;
[0021] Step 5, calculate the crossover probability and mutation probability of each individual in the current population, perform chromosome crossover and mutation, generate new individuals, and return to step 2.
[0022] Furthermore, the stopping condition is: the fitness value of the best individual in the current population is less than a threshold or the current number of iterations is the maximum number of iterations; the fitness value is calculated as:
[0023]
[0024] Among them, f i represents the fitness value of the i-th individual, n is the total number of training samples, y j , They represent the label value and predicted value of the jth training sample respectively.
[0025] Furthermore, the formula for the selection probability is:
[0026]
[0027] Among them, P i 、f i Respectively represent the selection probability and fitness value of the i-th individual, f min Indicates the minimum fitness value in the current population.
[0028] Furthermore, the formula for the crossover probability is:
[0029]
[0030] Among them, P ic 、f i Respectively represent the crossover probability and fitness value of the i-th individual, f avg 、f max They represent the average and maximum fitness of the current population, e represents the natural logarithm, P cmin = 0.5 indicates the minimum crossover probability, P cmax =((0.9-0.4) / num max )*num+0.5 represents the maximum crossover probability value, num, num max Respectively represent the current number of iterations and the maximum number of iterations;
[0031] The formula for the mutation probability is:
[0032]
[0033] Among them, P im represents the mutation probability of the i-th individual, P mmax Represents the maximum mutation probability P mmax=((0.09-0.05) / num max )*num+0.09,P mmin Represents the minimum mutation probability, 0<P mmin ≤0.05.
[0034] Furthermore, obtaining a digital radiographic detection atlas of each weld position specifically includes:
[0035] Perform edge detection on all the initial digital radiographic detection images at each weld position, and determine the axis centerline of the pressure pipeline based on the detected upper and lower edges of the pressure pipeline with the insulation layer;
[0036] Performing edge detection on the interference graph at each weld position, and determining the axis centerline of the pressure pipeline according to the upper and lower edges in the interference graph;
[0037] Aligning the axis line of the initial digital ray detection image with the axis line of the corresponding interference image, subtracting the grayscale values and performing Gaussian filtering to obtain a de-interferenced digital ray detection image;
[0038] All the interference-free digital ray detection images of each weld position are sorted according to the rotation angle of the circumferential rotation, so as to obtain a sequence of digital ray detection images of each weld position, and all the digital ray detection image sequences constitute the digital ray detection atlas.
[0039] Furthermore, the damage identification model is a deep neural network model including: a convolutional neural network, a long short-term memory network and a fully connected layer; the convolutional neural network is used to extract local features of a single digital ray detection image, and output a local feature sequence based on the digital ray detection image sequence; the long short-term memory network is used to extract the spatiotemporal features of the local feature sequence; the fully connected layer performs logistic regression on the local features and the spatiotemporal features to determine the output result.
[0040] Furthermore, the convolutional neural network includes a first convolutional layer, a first pooling layer and a second convolutional layer in sequence, the first pooling layer is a maximum pooling layer, the first pooling layer is used for downsampling, the sizes of the convolution kernels of the two convolutional layers are 5 and 3 respectively, the step size is 1, and the number of convolutional layer kernels is 64 and 128 respectively; the number of hidden layer neurons in the long short-term memory network is 128.
[0041] Through the above technical solutions, it can be known that compared with the prior art, the present invention discloses an enhanced digital radiographic detection image processing method for pressure pipelines with insulation layers. By introducing a distance intensity model and processing the signal absorption of insulation materials and waterproof layers respectively, the interference degree can be quantified more accurately, and interference analysis and de-interference processing can be performed on the weld position, which can effectively reduce the influence of the insulation layer and the waterproof layer on the detection image; by using a combination of a convolutional neural network and a long short-term memory network, the local features and spatiotemporal features of the image can be effectively extracted, and damage identification can be performed based on the image sequence with higher accuracy. The present invention can effectively improve the recognition ability and accuracy of internal defects of pressure pipelines with insulation layers, avoid removing the insulation layer, and realize rapid detection of pressure pipelines without stopping, thereby saving detection costs and improving detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0043] Figure 1 It is a schematic diagram of the overall process of an embodiment of the present invention.
[0044] Figure 2 FIG. 4 is a schematic diagram of digital ray detection according to an embodiment of the present invention. DETAILED DESCRIPTION
[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0046] The embodiment of the present invention discloses an enhanced digital ray detection image processing method for a pressure pipeline with a thermal insulation layer. Figure 1 As shown, the specific steps are as follows:
[0047] Determine the weld position and pipeline parameters of each pipe section according to the pipeline construction data;
[0048] Collect multiple initial digital radiographic inspection images along the circumferential direction of the pipeline at the weld position;
[0049] According to the pipeline parameters corresponding to each weld position, and the geometric position relationship between the radiation source, the detector array and the pipeline axis, the distance of the detection signal received by each detector in the detector array through the insulation material layer and the waterproof layer is determined, and the interference intensity of each detector is determined according to the distance intensity model, and the interference map of each weld position is further determined;
[0050] Align and subtract multiple initial digital radiographic detection images at each weld position from the corresponding interference images to obtain a digital radiographic detection atlas at each weld position;
[0051] Each digital radiographic detection image sequence in the digital radiographic detection atlas is input into the damage identification model to determine the damage condition of the corresponding weld position.
[0052] Specifically, the digital ray detection image is obtained using digital ray technology, such as Figure 2 As shown in the figure, the ray source emits x-rays as detection signals. The x-rays are absorbed by different layers of the pressure pipe with insulation layer and reach the detector array on the left. The leftmost side of the detector array is a lead protective layer, which is used to reduce the impact of x-rays on the surrounding environment. Each detector in the detector array detects different ray intensities, and then generates an initial digital ray detection map according to the absorption intensity of the ray. The digital ray detection map is a grayscale map. The greater the ray absorption intensity, the higher the grayscale value. During the detection, the distance between the ray source and the detector array is kept unchanged, and multiple initial digital ray detection maps are collected in sequence along the axis of the pipeline at a fixed rotation angle.
[0053] In a specific embodiment, after the weld positions of each pipe section are determined according to the pipeline construction data, the process further includes scanning the pressure pipeline with an infrared imager to list all possible weld positions.
[0054] In a specific embodiment, the pipeline parameters include: pipeline outer diameter, pipeline thickness, insulation material thickness, insulation material material, insulation material density, waterproof layer thickness, and waterproof layer material.
[0055] Among them, the thermal insulation material is usually silicate cotton, polyurethane thermal insulation cotton and other materials wrapped outside the pressure pipe, and the waterproof layer is usually polyurethane waterproof material, PVC waterproof material, iron sheet, etc. Here, the outer diameter and thickness of the pipe refer to the outer diameter of the metal pressure pipe. The outer diameter of the thermal insulation layer is obtained according to its outer diameter and the thickness of the thermal insulation layer, and the outer diameter of the waterproof layer is further determined according to the thickness of the waterproof layer. The symmetry line and the intersection of the symmetry line and the detector array are determined based on the ray source and the axis of the pipe; according to the distance between each detector and the intersection and the distance between the ray source and the detector array, the angle between the x-ray received by each detector and the symmetry line is determined, and then the length of the chord of the x-ray passing through each circle is calculated according to the distance between the axis of the pipe and the ray source and the radius of each layer, and the distance of the detection signal received by each detector through the thermal insulation layer and the waterproof layer is determined.
[0056] In a specific embodiment, the distance strength model includes a first sub-model and a second sub-model, which are respectively used to determine the absorption intensity of the detection signal passing through the thermal insulation material and the signal absorption intensity of the waterproof layer; the first sub-model and the second sub-model are both BP neural network models.
[0057] The first sub-model and the second sub-model are trained based on the first sample set and the second sample set respectively; the sample characteristics of the samples in the first sample set include: detection signal strength, frequency, path distance, density and material of the thermal insulation material, and the sample label is the signal strength received by the detector; the sample characteristics of the samples in the second sample set include: detection signal strength, frequency, path distance and material of the waterproof material, and the sample label is the signal strength received by the detector.
[0058] Specifically, in order to make the digital X-ray detection image have better contrast, it is necessary to adjust the detection signal intensity emitted by the X-ray source according to the thickness of the pressure pipeline and the thickness of the insulation layer, so as to improve the detection accuracy. Similarly, the penetration of X-rays of different frequencies into different materials is different. In actual detection, a reasonable detection frequency will be selected according to the pipeline construction data. Among the sample features, signal strength, frequency, path distance and density can directly generate corresponding feature vectors based on numerical values, and the material quality is converted into numerical format through unique hot encoding.
[0059] In a specific embodiment, the BP neural network model is trained based on an improved chromosome algorithm, specifically including:
[0060] Step 1: randomly generate an initial population, where each individual in the initial population represents the weight and threshold of each neuron in the neural network, and the weight and threshold are encoded as chromosomes;
[0061] Step 2, calculate the fitness value of each individual in the current population;
[0062] Step 3: Determine whether the stopping condition is met. If so, stop the iteration, decode the chromosome of the individual with the smallest fitness value, and output the neural network weight and threshold; otherwise, execute step 4.
[0063] Step 4: Perform chromosome screening in the current population and remove individuals whose selection probability is lower than the threshold;
[0064] Step 5, calculate the crossover probability and mutation probability of each individual in the current population, perform chromosome crossover and mutation, generate new individuals, and return to step 2.
[0065] Among them, the fitness value is obtained based on the prediction accuracy of the BP neural network under the weights and thresholds represented by different chromosomes. When the fitness value is lower than the set threshold, that is, the prediction accuracy meets the requirements, the iteration is stopped, and the corresponding weights and thresholds are obtained based on the current chromosome decoding. Binary encoding and decoding can be used for encoding and decoding. In the iterative process of genetic variation:
[0066] For chromosomes with smaller fitness, elimination based on selection probability and threshold can make the individuals in the population have higher prediction accuracy as a whole.
[0067] For chromosomes with lower fitness values, since they have higher prediction accuracy, and chromosome crossover will cause greater changes in chromosomes, the probability of obtaining chromosomes with higher prediction accuracy is not high, but may cause the offspring of the original chromosomes with higher prediction accuracy to become worse. Therefore, appropriately reducing their probability can ensure that chromosomes with low fitness values are retained in the population as a whole; for chromosomes with higher fitness values, by increasing their crossover probability, gene exchange is performed to generate more offspring chromosomes, thereby screening chromosomes with low fitness values from many offspring chromosomes. During the crossover process, the crossover probability product of any two parent chromosomes is first calculated, and the parent with a value greater than the set threshold is selected to generate offspring chromosomes through random coding exchange.
[0068] By introducing a mutation mechanism, local optimality can be prevented. Similar to chromosome crossing, chromosomes with lower fitness values have lower mutation probabilities, while chromosomes with higher fitness values have higher mutation probabilities. Chromosomes with mutation probabilities greater than a set threshold are selected for mutation to obtain mutant offspring chromosomes. Mutation can be simulated by flipping the random positions of chromosome encodings.
[0069] Finally, the unchanged chromosomes retained after chromosome elimination and the generated new offspring chromosomes enter the next round of iteration as a new population.
[0070] In a specific embodiment, the stopping condition is: the fitness value of the best individual in the current population is less than the threshold or the current iteration number is the maximum iteration number; the calculation formula of the fitness value is:
[0071]
[0072] Among them, f i represents the fitness value of the i-th individual, n is the total number of training samples, y j , They represent the label value and predicted value of the jth training sample respectively.
[0073] In a specific embodiment, the formula for selecting probability is:
[0074]
[0075] Among them, P i 、f i Respectively represent the selection probability and fitness value of the i-th individual, f min Indicates the minimum fitness value in the current population.
[0076] In a specific embodiment, the formula for the crossover probability is:
[0077]
[0078] Among them, P ic 、f i Respectively represent the crossover probability and fitness value of the i-th individual, f avg 、f max They represent the average and maximum fitness of the current population, e represents the natural logarithm, P cmin = 0.5 indicates the minimum crossover probability, P cmax =((0.9-0.4) / num max )*num+0.5 represents the maximum crossover probability value, num, num max Respectively represent the current number of iterations and the maximum number of iterations;
[0079] The formula for mutation probability is:
[0080]
[0081] Among them, P im represents the mutation probability of the i-th individual, P mmax Represents the maximum mutation probability P mmax =((0.09-0.05) / num max )*num+0.09,P mmin Represents the minimum mutation probability, 0<P mmin ≤0.05.
[0082] In a specific embodiment, obtaining a digital radiographic detection atlas of each weld position specifically includes:
[0083] Perform edge detection on all initial digital radiographic inspection images of each weld position, and determine the axis of the pressure pipeline based on the detected upper and lower edges of the pressure pipeline with insulation layer;
[0084] Perform edge detection on the interference graph of each weld position, and determine the axis centerline of the pressure pipeline according to the upper and lower edges in the interference graph;
[0085] Align the axis of the initial digital ray detection image with the axis of the corresponding interference image, subtract the grayscale values (i.e., subtract the grayscale values to remove the interference of the insulation layer), and then perform Gaussian filtering to obtain a de-interference digital ray detection image;
[0086] All interference-free digital radiographic detection images of each weld position are sorted according to the rotation angle of the circumferential rotation shooting, so as to obtain a sequence of digital radiographic detection images of each weld position, and all digital radiographic detection image sequences constitute a digital radiographic detection atlas.
[0087] In a specific embodiment, the damage recognition model is a deep neural network model including: a convolutional neural network, a long short-term memory network and a fully connected layer; the convolutional neural network is used to extract the local features of a single digital ray detection image, and output a local feature sequence based on the digital ray detection image sequence; the long short-term memory network is used to extract the spatiotemporal features of the local feature sequence; the fully connected layer performs logistic regression on the local features and spatiotemporal features to determine the output result. Affected by the shooting angle, a single image cannot well identify pipeline damage, and since x-rays usually need to pass through the pipeline wall twice, the accuracy of damage detection can be improved by combining the local features of a single detection image and the spatiotemporal features of the sequence through the detection image sequence.
[0088] In a specific embodiment, the convolutional neural network includes a first convolutional layer, a first pooling layer, and a second convolutional layer in sequence. The first pooling layer is a maximum pooling layer, and the first pooling layer is used for downsampling. The sizes of the convolution kernels of the two convolutional layers are 5 and 3 respectively, the step size is 1, and the number of convolutional layer kernels is 64 and 128 respectively; the number of hidden layer neurons in the long short-term memory network is 128.
[0089] In a specific embodiment, the training samples of the damage identification model are obtained by setting high-pressure pipeline samples with various damages (such as cracks, rust, holes, etc.) at the weld positions, and performing simulated detection to obtain a sequence of digital radiographic detection images of the weld positions.
[0090] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0091] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. An enhanced digital radiographic inspection image processing method for a pressure pipeline with an insulation layer, characterized in that: The specific steps are as follows: Determine the weld position and pipeline parameters of each pipe section according to the pipeline construction data; Collect multiple initial digital radiographic inspection images along the circumferential direction of the pipeline at the weld position; According to the pipeline parameters corresponding to each weld position, and the geometric position relationship between the radiation source, the detector array and the pipeline axis, the distance of the detection signal received by each detector in the detector array through the thermal insulation material layer and the waterproof layer is determined, and the interference intensity of each detector is determined according to the distance intensity model, and the interference map of each weld position is further determined; Aligning and subtracting the multiple initial digital ray detection images at each weld position with the corresponding interference images to obtain a digital ray detection atlas at each weld position; Each digital radiographic detection image sequence in the digital radiographic detection atlas is input into a damage identification model to determine the damage condition of the corresponding weld position.
2. The enhanced digital radiographic inspection image processing method for pressure pipes with thermal insulation layer according to claim 1 is characterized in that: The pipeline parameters include: pipeline outer diameter, pipeline thickness, insulation material thickness, insulation material material, insulation material density, waterproof layer thickness, and waterproof layer material.
3. The enhanced digital radiographic inspection image processing method for pressure pipes with thermal insulation layer according to claim 1 is characterized in that: The distance strength model includes a first sub-model and a second sub-model, which are respectively used to determine the absorption strength of the detection signal passing through the thermal insulation material and the signal absorption strength of the waterproof layer; the first sub-model and the second sub-model are both BP neural network models; The first sub-model and the second sub-model are trained based on the first sample set and the second sample set respectively; The sample characteristics of each sample in the first sample set include: detection signal strength, frequency, path distance, density and material of the thermal insulation material, and the sample label is the signal strength received by the detector; the sample characteristics of each sample in the second sample set include: detection signal strength, frequency, path distance and material of the waterproof material, and the sample label is the signal strength received by the detector.
4. The enhanced digital radiographic inspection image processing method for pressure pipes with thermal insulation layer according to claim 3 is characterized in that: The BP neural network model is trained based on an improved chromosome algorithm, specifically including: Step 1: randomly generate an initial population, where each individual in the initial population represents the weight and threshold of each neuron in the neural network, and the weight and threshold are encoded as chromosomes; Step 2, calculate the fitness value of each individual in the current population; Step 3: Determine whether the stopping condition is met. If so, stop the iteration, decode the chromosome of the individual with the smallest fitness value, and output the neural network weight and threshold; otherwise, execute step 4. Step 4: Perform chromosome screening in the current population and remove individuals whose selection probability is lower than the threshold; Step 5, calculate the crossover probability and mutation probability of each individual in the current population, perform chromosome crossover and mutation, generate new individuals, and return to step 2.
5. The enhanced digital radiographic inspection image processing method for pressure pipes with thermal insulation layer according to claim 4 is characterized in that: The stopping condition is: the fitness value of the best individual in the current population is less than the threshold or the current number of iterations is the maximum number of iterations; the calculation formula of the fitness value is: Among them, f i represents the fitness value of the i-th individual, n is the total number of training samples, y j , They represent the label value and predicted value of the jth training sample respectively.
6. The enhanced digital ray detection image processing method for pressure pipelines with thermal insulation layer according to claim 4 is characterized in that: The formula for the selection probability is: Among them, P i 、f i Respectively represent the selection probability and fitness value of the i-th individual, f min Indicates the minimum fitness value in the current population.
7. The enhanced digital radiographic inspection image processing method for pressure pipes with thermal insulation layer according to claim 4 is characterized in that: The formula for the crossover probability is: Among them, P ic 、f i Respectively represent the crossover probability and fitness value of the i-th individual, f avg 、f max They represent the average and maximum fitness of the current population, e represents the natural logarithm, P cmin = 0.5 indicates the minimum crossover probability, P cmax =((0.9-0.4) / num max )*num+0.5 represents the maximum crossover probability value, num, num max Respectively represent the current number of iterations and the maximum number of iterations; The formula for the mutation probability is: Among them, P im represents the mutation probability of the i-th individual, P mmax Represents the maximum mutation probability P mmax =((0.09-0.05) / num max )*num+0.09,P mmin Represents the minimum mutation probability, 0<P mmin ≤0.
05.
8. The enhanced digital radiographic inspection image processing method for pressure pipes with thermal insulation layer according to claim 1 is characterized in that: The step of obtaining a digital radiographic detection atlas of each weld position specifically includes: Performing edge detection on all the initial digital radiographic detection images at each weld position, and determining the axis centerline of the pressure pipeline based on the detected upper and lower edges of the pressure pipeline with the insulation layer; Performing edge detection on the interference graph at each weld position to determine the axis centerline of the pressure pipeline; Aligning the axis line of the initial digital ray detection image with the axis line of the corresponding interference image, subtracting the grayscale values and performing Gaussian filtering to obtain a de-interferenced digital ray detection image; All the interference-free digital ray detection images of each weld position are sorted according to the rotation angle of the circumferential rotation, so as to obtain a sequence of digital ray detection images of each weld position, and all the digital ray detection image sequences constitute the digital ray detection atlas.
9. The enhanced digital radiographic inspection image processing method for pressure pipes with thermal insulation layer according to claim 8 is characterized in that: The damage identification model is a deep neural network model including: a convolutional neural network, a long short-term memory network and a fully connected layer; the convolutional neural network is used to extract the local features of a single digital ray detection image, and output a local feature sequence based on the digital ray detection image sequence; the long short-term memory network is used to extract the spatiotemporal features of the local feature sequence; the fully connected layer performs logistic regression on the local features and the spatiotemporal features to determine the output result.
10. The enhanced digital radiographic inspection image processing method for pressure pipelines with thermal insulation layer according to claim 9 is characterized in that: The convolutional neural network includes a first convolutional layer, a first pooling layer and a second convolutional layer in sequence, the first pooling layer is a maximum pooling layer, the first pooling layer is used for downsampling, the sizes of the convolution kernels of the two convolutional layers are 5 and 3 respectively, the step size is 1, and the number of convolutional layer kernels is 64 and 128 respectively; the number of hidden layer neurons in the long short-term memory network is 128.