Method for predicting mechanical properties of sewer pipes based on hyperspectral images and machine learning

Abstract

The application discloses the technical field of intelligent detection of drainage pipeline and a mechanical property prediction method for drainage pipeline based on hyperspectral images and machine learning, and comprises the following steps: S1, preparing a test sample, and measuring and acquiring porosity and saturation rate; S2, collecting hyperspectral data, and obtaining compressive strength values through experiments; S3, using a machine learning method to fit the relationship between reflectivity and saturation rate, and analyzing the relationship between mix proportion and porosity; S4, proposing a model saturation rate-porosity-compressive strength extension model; S5, collecting spectral data, obtaining mix proportion information and measured compressive strength, substituting into the model for verification, and realizing mechanical property prediction of the drainage pipeline; the application combines spectral detection and machine learning, can nondestructively, noncontactly and accurately detect the mechanical property of the drainage pipeline, reduces the cost consumed by mechanical property detection of the drainage pipeline, greatly improves the efficiency and accuracy, and provides a new technical path for efficient evaluation of the mechanical property of the drainage pipeline.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection technology for drainage pipelines, specifically a method for predicting the mechanical properties of drainage pipelines based on hyperspectral images and machine learning. Background Technology

[0002] As a crucial component of urban infrastructure, the health of underground drainage pipes directly impacts a city's flood control and drainage capabilities, as well as residents' quality of life. With increasing service life, underground drainage pipes are affected by environmental pollution, soil erosion, and geological changes, leading to gradual aging, corrosion, or damage to the pipe structure, thus affecting drainage efficiency and safety. Particularly under long-term loads, the pipes' pressure resistance and impermeability significantly decrease, increasing the risk of urban flooding. Improper maintenance can ultimately lead to pipe structural failure and sudden accidents. To mitigate this risk, employing efficient underground drainage pipe strength testing methods and implementing timely, targeted maintenance and reinforcement measures are becoming increasingly important.

[0003] Currently, the core objective of testing the mechanical properties of drainage pipelines is to obtain information on their load-bearing capacity, deformation characteristics, and structural degradation degree under service conditions, in order to assess their safety and remaining service life. However, due to limitations imposed by underground space, complex operating conditions, and the requirement for non-destructive testing, accurately obtaining the mechanical properties of drainage pipelines still faces many challenges.

[0004] During long-term operation, drainage pipelines are subjected to the combined effects of overburden load, traffic load, groundwater pressure, and uneven soil settlement. As a result, their structural stress state and mechanical properties will continue to evolve, making them prone to problems such as pipe wall cracking, elliptical deformation, joint misalignment, and local instability.

[0005] Existing methods for testing the mechanical properties of drainage pipelines mainly rely on geometric deformation information obtained from CCTV, laser scanning, and sonar detection, combined with empirical models or standard formulas to invert the stress state and structural strength of the pipeline. The test results are highly dependent on assumptions about the working conditions and human experience. Although mechanical parameters such as stress and strain can be obtained through embedded sensors or in-situ loading tests, they suffer from problems such as large construction interference, high cost, and limited applicability, making it difficult to meet the rapid testing needs of large-scale drainage pipelines.

[0006] Based on this, the present invention designs a method for predicting the mechanical properties of drainage pipes based on hyperspectral images and machine learning to solve the above problems. Summary of the Invention

[0007] The purpose of this invention is to provide a method for predicting the mechanical properties of drainage pipes based on hyperspectral images and machine learning, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A method for predicting the mechanical properties of drainage pipes based on hyperspectral images and machine learning includes the following steps:

[0010] S1. Prepare standard concrete specimen samples, obtain the porosity of the prepared specimens by the drainage method, set the water content gradient, and measure the saturation rate;

[0011] S2. Collect hyperspectral data on the specimen prepared in S1. The hyperspectral data contains the spectral reflectance sequence of each pixel. After the data collection is completed, perform a standard compressive strength test to obtain the compressive strength value.

[0012] S3. Preprocess the hyperspectral data collected in S2 and use machine learning methods to fit the relationship between reflectance and saturation; collect concrete porosity dataset and use machine learning methods to analyze the relationship between mix proportion and porosity.

[0013] S4. Based on the classical strength model, an extended model of saturation rate-porosity-compressive strength is proposed for the saturation rate and porosity proposed in S3.

[0014] S5. Collect spectral data of drainage pipe sections, obtain mix proportion information and measured compressive strength, substitute them into the proposed strength model for verification, and realize the prediction of mechanical properties of drainage pipes based on hyperspectral images and machine learning.

[0015] Further, step S1 includes:

[0016] S1-1. Prepare 6 sets of standard concrete specimens with different mix proportions, mainly controlling the water-cement ratio. Set the water-cement ratio gradient as 0.40, 0.45, 0.50, 0.55, 0.60, 0.65, and prepare 12 specimens for each water-cement ratio grade.

[0017] S1-2. Dry the prepared specimen at 105℃ until it is completely dry, record the weight, boil the specimen until it is completely saturated, record the weight, and calculate the porosity by gravimetric method.

[0018] S1-3. Set a moisture content gradient, dry the saturated specimen at 105°C to the specified mass, remove it, seal all surfaces with aluminum foil, and place it at 20°C for one month, recording the weight.

[0019] Further, step S2 includes:

[0020] S2-1. Use a hyperspectral camera to collect hyperspectral data of the specimen obtained in S1. Before collection, perform white plate correction and dark current correction. The hyperspectral data contains the spectral reflectance sequence of each pixel.

[0021] S2-2. After the collection is completed, the specimen obtained in S1 is subjected to a standard compressive strength test to obtain the compressive strength value.

[0022] Further, step S3 includes:

[0023] S3-1. Use SG smoothing filter to remove noise from the hyperspectral data obtained in S2 and retain significant spectral features;

[0024] S3-2. Divide the surface of each specimen into 20 regular regions (ROIs), and obtain a reflectance curve for each ROI. The amount of spectral data for a single specimen is 4×20=80.

[0025] S3-3. Machine learning methods are used to analyze the relationship between saturation rate and reflectivity for specimens with different saturation gradients, and a function expressing the relationship between saturation rate and reflectivity is obtained.

[0026] S3-4. Create a concrete porosity dataset, which includes the corresponding mix proportions and porosities of the specimens prepared in S1, as well as the corresponding mix proportions and porosities of concrete specimens collected from the literature.

[0027] S3-5. Machine learning methods are used to analyze the relationship between porosity and mix ratio for specimens with different mix ratios, and a function expressing the relationship between porosity and mix ratio is obtained.

[0028] Furthermore, the standard concrete specimen in step S1 has a size of 100mm×100mm×100mm, the cement in the specimen material is ordinary Portland cement, the coarse aggregate is 5~20mm granite crushed stone, and the fine aggregate is natural river sand with a fineness modulus of 2.3.

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] This invention enables rapid surface scanning through hyperspectral imaging without damaging the structure, making it particularly suitable for on-site assessment of existing drainage pipes, etc. Moreover, hyperspectral data acquisition is completed almost instantaneously. Combined with a trained machine learning model, it can predict saturation rate, porosity, and intensity in near real-time, significantly improving detection efficiency.

[0031] This invention uses hyperspectral scanning to simultaneously invert saturation rate and indirectly correlated porosity, and further predicts intensity, achieving simultaneous acquisition of multiple parameters and integrated evaluation of multiple performance characteristics.

[0032] This invention combines spectral detection technology with machine learning to perform non-destructive, non-contact, and accurate mechanical performance testing of drainage pipes. It avoids the shortcomings of traditional methods that rely on human experience and simplistic assumptions, thereby reducing the cost of mechanical performance testing of drainage pipes and greatly improving efficiency and accuracy. It provides a new technical path for the efficient evaluation of the mechanical performance of drainage pipes and has great potential for the development of refined management and intelligent operation and maintenance of underground pipe networks. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 A flowchart for a method to predict the mechanical properties of drainage pipes;

[0035] Figure 2 A graph showing the effect of fitting the relationship between reflectance and saturation using a machine learning method;

[0036] Figure 3 The graph shows the effect of fitting the relationship between the mix ratio and porosity using a machine learning method. Detailed Implementation

[0037] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Please see the appendix Figure 1-3 The present invention provides a technical solution:

[0039] A method for predicting the mechanical properties of drainage pipes based on hyperspectral images and machine learning includes the following steps:

[0040] S1. Prepare standard concrete specimen samples, obtain the porosity of the prepared specimens by the drainage method, set the water content gradient, and measure the saturation rate.

[0041] The specific steps in this process include:

[0042] S1-1. Prepare 6 sets of standard concrete specimens with different mix proportions, mainly controlling the water-cement ratio. Set the water-cement ratio gradients as 0.40, 0.45, 0.50, 0.55, 0.60, and 0.65, and prepare 12 specimens for each water-cement ratio grade.

[0043] The concrete specimens were standard cubes with dimensions of 100mm×100mm×100mm. Six different mix proportions were set with water-cement ratios of 0.40, 0.45, 0.50, 0.55, 0.60, and 0.65, and 12 specimens were prepared for each group, for a total of 72 specimens.

[0044] Ordinary Portland cement was used for preparing the specimens. 5-20mm granite crushed stone was used as the coarse aggregate, and natural river sand with a fineness modulus of 2.3 was used as the fine aggregate. The specimens were cured in accordance with the "Standard for Test Methods of Mechanical Properties of Ordinary Concrete" (GB / T 50081-2019).

[0045] S1-2. Dry the prepared specimen at 105℃ until it is completely dry, record the weight, boil the specimen until it is completely saturated, record the weight, and calculate the porosity by gravimetric method.

[0046] Specifically, the specimen prepared in S1-1 was dried at 105°C until its weight no longer changed, reaching a completely dry state, and the weight was recorded. The boiling method was performed according to the "Test Method for Strength of Cement Mortar" (GB / T 17671) until the mortar reached complete saturation, and the weight was recorded. Completely immerse the saturated specimen in water, but do not let it touch the bottom of the container, and record the weight. Porosity was calculated using the drainage method. The calculation formula is as follows:

[0047]

[0048] In the formula, Dry weight For saturated weight, Weight in water Porosity.

[0049] S1-3. Set a moisture content gradient, dry the saturated specimen at 105°C to the specified mass, remove it, seal all surfaces with aluminum foil, and place it at 20°C for one month, recording the weight.

[0050] Specifically, for each group of water-cement ratio specimens, five different saturation levels were set: 0, 0.25, 0.50, 0.75, and 1. Three specimens were used for each saturation gradient, and the mass of each specimen reaching the corresponding saturation level was calculated using a formula. And dry at 105℃ to the specified weight. Remove from the container and immediately seal all surfaces with aluminum foil. Store at 20°C for one month and record the weight.

[0051] The calculation formula is as follows:

[0052]

[0053] In the formula, Dry weight; Saturated weight; For the specified weight; This represents saturation.

[0054] S2. Collect hyperspectral data on the specimen prepared in S1. The hyperspectral data contains the spectral reflectance sequence of each pixel. After the data collection is completed, perform a standard compressive strength test to obtain the compressive strength value.

[0055] The specific steps in this process include:

[0056] S2-1. Use a hyperspectral camera to collect hyperspectral data of the specimen obtained in S1. Before collection, perform whiteboard correction and dark current correction. The hyperspectral data contains the spectral reflectance sequence of each pixel.

[0057] Specifically, the specimen obtained in S1 was scanned using a hyperspectral near-infrared wavelength camera under full-band light source conditions. The camera was connected to a laptop computer via data acquisition software for parameter setting and image acquisition. The hyperspectral near-infrared wavelength camera was a FigSpec FS-25 with a spectral range of 900~1700nm and a spectral resolution of 8.0nm.

[0058] Before formal data acquisition, the system is calibrated using a white calibration plate with a reflectivity of 80%. The camera distance and light source intensity are adjusted to ensure that the detected reflected light intensity reaches 60% of the saturation light intensity, thus avoiding data loss due to insufficient lighting.

[0059] To eliminate noise caused by dark current and uneven illumination intensity in hyperspectral cameras, it is necessary to calibrate hyperspectral images by collecting and processing the average light intensity and dark current of the white reflectance calibration plate; calculate the corrected reflectance according to the formula, and extract the calibrated reflectance spectrum of each pixel in the image to obtain corrected reflectance hyperspectral image data.

[0060] The calculation formula is as follows:

[0061]

[0062] In the formula, The calibration value represents the corrected reflectance; , and These represent the light intensity values ​​of the measurement area, the calibration whiteboard, and the dark current, respectively. To calibrate the reflectivity of the whiteboard (fixed value 80%).

[0063] S2-2. After the collection is completed, the specimen obtained in S1 is subjected to a standard compressive strength test to obtain the compressive strength value.

[0064] Specifically, uniaxial compression tests were conducted on the specimens obtained from S1, and the uniaxial compressive strength values ​​of each concrete specimen were obtained through standard uniaxial compression tests. The uniaxial compression tests were carried out using a WDW-600C hydraulic servo testing machine, with the loading rate set to 0.2 mm / min until the specimen was completely destroyed. The uniaxial compressive strength of each specimen was then measured.

[0065] S3. Preprocess the hyperspectral data collected in S2 and use machine learning methods to fit the relationship between reflectance and saturation; collect concrete porosity dataset and use machine learning methods to analyze the relationship between mix proportion and porosity.

[0066] The specific steps in this process include:

[0067] S3-1. Use SG smoothing filter to remove noise from the hyperspectral data obtained in S2 while retaining significant spectral features.

[0068] Specifically, the hyperspectral data acquired by S2 is smoothed using Savitzky-Golay filtering (SG smoothing). Polynomial smoothing preserves useful information in the spectral data, eliminates random noise, and retains significant spectral features. The SG smoothing formula used is as follows:

[0069]

[0070] In the formula, This represents the smoothed reflectance value at a point within the smoothed window. For position Spectral reflectance at that location The size of the sliding convolution box. For the weighting factors in the smoothing process, denoted as the order of the polynomial fitting.

[0071] S3-2. Divide the surface of each specimen into 20 regular regions (ROIs), and obtain a reflectance curve for each ROI. The amount of spectral data for a single specimen is 4×20=80.

[0072] Specifically, 20 regular regions (ROIs) were precisely divided from the hyperspectral image of the 100mm×100mm specimen surface. The average spectral reflectance curve of each region was calculated to reduce the error caused by local variation points. Four surfaces were selected for each specimen, and the spectral data of a single specimen was 4×20=80, for a total of 5760 data points.

[0073] S3-3. Machine learning methods are used to analyze the relationship between saturation rate and reflectivity for specimens with different saturation gradients, and a display function expressing the relationship between saturation rate and reflectivity is obtained.

[0074] Specifically, the fitted data includes the spectral data collected in S3-2 and the specimen saturation rate set in S1. The spectral data contains 256 channels, each representing a specific band within the 900-1700 nm wavelength range, with the value representing the reflectance of the concrete specimen in that channel. The set specimen saturation rate corresponds to the spectral data of that specimen; that is, one spectral data point for a region of interest (ROI) of each specimen corresponds to one saturation rate for that specimen, resulting in a total of 5760 corresponding saturation rate and reflectance data points.

[0075] The input feature is the reflectance values ​​of 256 channels. The data is randomly divided into stratified groups with 80% of the data in the training set and 20% in the test set to ensure the balance of the data distribution. The Least Squares Support Vector Machine (LSSVM) machine learning model is selected for training, and the performance of the model is evaluated using R² and RMSE.

[0076] The least squares support vector machine model is expressed as follows:

[0077]

[0078] st

[0079] In the formula, This is the weight vector; For bias terms; This is the error term; It is a nonlinear mapping function that maps the input to a high-dimensional feature space; This is the regularization parameter (penalty coefficient).

[0080] The relationship between saturation and reflectance obtained from training is as follows:

[0081]

[0082] In the formula, This represents the number of training samples; Let be the weights (Lagrange multipliers) of the i-th training sample; For bias terms; The spectral data of the i-th training sample; The predicted concrete saturation rate is expressed in % (%). For the new sample spectral data.

[0083] S3-4. Create a concrete porosity dataset, which includes the corresponding mix proportions and porosities of the specimens prepared in S1, as well as the corresponding mix proportions and porosities of concrete specimens collected from the literature.

[0084] Specifically, a database was established by collecting experimental data on concrete porosity and porosity data of specimens prepared in S1 from publicly available literature according to specific selection criteria.

[0085] The data selection criteria are as follows: First, ordinary Portland cement was used to prepare the concrete; second, all concrete specimens contained both coarse and fine aggregates; third, the curing regimes were divided into air curing and water curing; fourth, the effect of carbonation on pore structure refinement was not considered in the specimens. Furthermore, a balanced mix proportion was maintained for each type of concrete. The final database contains 300 data records, covering 80 unique concrete mix designs.

[0086] The data includes eight input features: water-cement ratio, cementitious material dosage (kg / m³), fly ash content (%), slag powder content (%), water-reducing agent content (%), coarse and fine aggregate ratio, curing conditions (categorical variable), and curing age.

[0087] In this study, the dosage of fly ash and slag powder was recorded as the percentage of replacement of cementitious materials (cement and auxiliary cementitious materials). The amount of water-reducing agent added was uniformly converted to the mass percentage of cementitious materials in the concrete. If the water-reducing agent data in the original literature was in volume form, the density of the water-reducing agent was assumed to be 1.2 kg / L for conversion. All eight predictor variables in the established database had no missing values.

[0088] S3-5. Machine learning methods are used to analyze the relationship between porosity and mix ratio for specimens with different mix ratios, and a function expressing the relationship between porosity and mix ratio is obtained.

[0089] Specifically, the data is randomly partitioned into stratified groups of 80% for training and 20% for testing to ensure balanced data distribution. The model is trained on a regularized gradient boosting model built on XGBoost. A Bayesian optimization algorithm is used to automatically optimize the hyperparameters of the random forest and gradient boosting tree, targeting out-of-bag error and 10-fold cross-validation error respectively. On an independent test set, the prediction accuracy of the model is evaluated using multiple metrics such as RMSE, MAPE, and R².

[0090] The XGBoost regularized gradient boosting model is expressed as follows:

[0091]

[0092] Among them, the regularization term Defined as:

[0093]

[0094] In the formula, MSE is the mean squared error loss; The penalty coefficient is used to control the number of leaf nodes; The number of leaf nodes in the tree; is the L2 regularization coefficient, which penalizes the sum of squares of the weights of the leaf nodes; is the L1 regularization coefficient, which penalizes the sum of the absolute values ​​of the leaf node weights; The leaf node weight vector; Let q be the q-norm of vector w.

[0095] The L1 regularization coefficient is set to 0.01, and the L2 regularization coefficient is set to 2.

[0096] The relationship between porosity and mix proportion obtained from the training is shown in the following function:

[0097]

[0098] In the formula, The predicted concrete porosity is expressed in % (%). For the input feature vector, , These are the input features from the database; The learned mapping function is constructed using an XGBoost regularized gradient boosting model; These are the basic parameters of the model, i.e., the specific structure of the XGBoost regularized gradient boosting model, such as split features, split thresholds, leaf node values, etc. Hyperparameters of the model control the overall architecture and training process of the model, such as the number of trees, depth, and learning rate.

[0099] S4. Based on the classical strength model, an extended model of saturation rate-porosity-compressive strength is proposed for the saturation rate and porosity proposed in S3.

[0100] Specifically, based on several existing strength-porosity models, such as Balshin, Ryshkevitch, Schiller, and Hasselman, it was found that the Ryshkevitch index model performs best in fitting the relationship between strength and porosity. Furthermore, related studies have shown that changes in saturation significantly affect strength. Based on the classic Ryshkevitch porosity-strength index model, a strength prediction model that combines pore structure and moisture state is established by introducing a saturation variable and its nonlinear modulation term. This compensates for the original model's failure to reflect changes in material mechanical properties under water-bearing conditions. The prediction accuracy of the model is evaluated using RMSE, R², and R².

[0101] The model is expressed as follows:

[0102]

[0103] In the formula, The compressive strength of the test specimen; The porosity of the test specimen; The saturation level of the test specimen; This is the baseline strength term; This is the porosity attenuation coefficient; The water-pore coupling modulation coefficient; This is the nonlinear influence coefficient of saturation. The porosity–critical saturation coupling coefficient; This is the baseline critical saturation parameter.

[0104] in, , , , , , The result is obtained by fitting the input porosity, saturation, and compressive strength.

[0105] S5. Collect spectral data of drainage pipe sections, obtain mix proportion information and measured compressive strength, substitute them into the proposed strength model for verification, and realize the prediction of mechanical properties of drainage pipes based on hyperspectral images and machine learning.

[0106] Specifically, the spectral camera described in S2 is used to scan the drainage pipe section on site to obtain spectral data. The saturation rate of the scanned drainage pipe section is predicted using the reflectance-saturation rate relationship model proposed in S3. The porosity of the scanned drainage pipe section is predicted using the mix ratio-porosity relationship model proposed in S3. The predicted saturation rate and porosity are then used to predict the compressive strength using the saturation rate-porosity-compressive strength model proposed in S4, thereby achieving non-destructive and rapid detection of the mechanical properties of the drainage pipe.

[0107] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0108] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for predicting the mechanical properties of drainage pipes based on hyperspectral images and machine learning, characterized in that, Includes the following steps: S1. Prepare standard concrete specimen samples, obtain the porosity of the prepared specimens by the drainage method, set the water content gradient, and measure the saturation rate; S2. Collect hyperspectral data on the specimen prepared in S1. The hyperspectral data includes the spectral reflectance sequence of each pixel. After the data collection is completed, a standard compressive strength test is performed to obtain the compressive strength value. S3. Preprocess the hyperspectral data collected in S2 and use machine learning methods to fit the relationship between reflectance and saturation; collect concrete porosity dataset and use machine learning methods to analyze the relationship between mix proportion and porosity. S4. Based on the classical strength model, an extended model of saturation rate-porosity-compressive strength is proposed for the saturation rate and porosity proposed in S3. S5. Collect spectral data of drainage pipe sections, obtain mix proportion information and measured compressive strength, substitute them into the proposed strength model for verification, and realize the prediction of mechanical properties of drainage pipes based on hyperspectral images and machine learning.

2. The method for predicting the mechanical properties of drainage pipes based on hyperspectral images and machine learning according to claim 1, characterized in that, Step S1 includes: S1-1. Prepare 6 sets of standard concrete specimens with different mix proportions, and set the water-cement ratio gradients as 0.40, 0.45, 0.50, 0.55, 0.60, and 0.65 respectively. Prepare 12 specimens for each water-cement ratio grade. S1-2. Dry the prepared specimen at 105℃ until it is completely dry, record the weight, boil the specimen until it is completely saturated, record the weight, and calculate the porosity by gravimetric method. S1-3. Set a moisture content gradient, dry the saturated specimen at 105°C to the specified mass, remove it, seal all surfaces with aluminum foil, and place it at 20°C for one month, recording the weight.

3. The method for predicting the mechanical properties of drainage pipes based on hyperspectral images and machine learning according to claim 1, characterized in that, Step S2 includes: S2-1. Use a hyperspectral camera to collect hyperspectral data of the specimen obtained in S1. Before collection, perform white plate correction and dark current correction. The hyperspectral data contains the spectral reflectance sequence of each pixel. S2-2. After the collection is completed, the specimen obtained in S1 is subjected to a standard compressive strength test to obtain the compressive strength value.

4. The method for predicting the mechanical properties of drainage pipes based on hyperspectral images and machine learning according to claim 1, characterized in that, Step S3 includes: S3-1. Use SG smoothing filter to remove noise from the hyperspectral data obtained in S2 and retain significant spectral features; S3-2. Divide the surface of each specimen into 20 regular regions (ROIs), and obtain a reflectance curve for each ROI. The amount of spectral data for a single specimen is 4×20=80. S3-3. Machine learning methods are used to analyze the relationship between saturation rate and reflectivity for specimens with different saturation gradients, and a function expressing the relationship between saturation rate and reflectivity is obtained. S3-4. Create a concrete porosity dataset, which includes the corresponding mix proportions and porosities of the specimens prepared in S1, as well as the corresponding mix proportions and porosities of concrete specimens collected from the literature. S3-5. Use machine learning methods to analyze the relationship between porosity and mix ratio for specimens with different mix ratios, and obtain an explicit relationship function that expresses the relationship between porosity and mix ratio.

5. The method for predicting the mechanical properties of drainage pipes based on hyperspectral images and machine learning according to claim 1, characterized in that, The standard concrete specimen in step S1 has a size of 100mm×100mm×100mm. The cement in the specimen material is ordinary Portland cement, the coarse aggregate is 5~20mm granite crushed stone, and the fine aggregate is natural river sand with a fineness modulus of 2.3.

Images