Subgrade soil dry density hyperspectral nondestructive testing method based on soil compactness coefficient

Abstract

The invention discloses a subgrade soil dry density hyperspectral nondestructive testing method based on a soil compactness coefficient. The method comprises the following steps: a1, uniformly distributing physical and chemical parameters of engineering foundation soil in a space; b2, a remote sensing product used for monitoring the water content of engineering foundation soil is obtained and generated through the four-rotor unmanned aerial vehicle; c3, building model parameters in combination with indoor experiments; d4, the engineering foundation is compacted, and the soil dry density of the engineering foundation is actually measured through a sand filling method and verified; e5, a remote sensing product used for monitoring the soil water content after engineering foundation compaction is obtained and generated through the four-rotor unmanned aerial vehicle; and f6, constructing a soil compactness coefficient image based on the hyperspectral image products generated in the step b2 and the step e5, inputting the image of the corresponding characteristic parameter into the model obtained in the step d4, and estimating the dry density of the roadbed construction soil. According to the method, rapid, lossless and real-time detection of the dry density of the roadbed soil can be realized through an unmanned aerial vehicle-hyperspectral technology, and the defects of a traditional soil dry density determination method are effectively overcome.

Description

Technical Field

[0001] The present invention relates to the technical fields of engineering construction and non-destructive detection of dry density of plough layer soil, and particularly relates to a hyperspectral non-destructive detection method for dry density of subgrade soil based on soil compactness coefficient. Background Art

[0002] The compaction of highway subgrade is a key link in highway construction, and the compaction quality of the subgrade can directly affect the strength and durability of the highway pavement. Therefore, the dry density of highway subgrade soil is an important detection index for highway construction quality and also a key factor for evaluating highway construction quality. The traditional dry density of highway subgrade soil is mainly carried out by the mode of field sampling + laboratory detection. This mode has deficiencies such as subgrade damage, taking points instead of the whole, and time-consuming, and cannot meet the application requirements of obtaining the dry density information of highway subgrade soil in real time, accurately and non-destructively; in recent years, hyperspectral technology has been widely used in the monitoring of land surface information with its own advantages such as non-destructive, real-time, fast and accurate, which provides a new solution idea for the accurate detection of dry density of highway subgrade soil.

[0003] At present, the standard method for detecting the dry density of highway subgrade soil is the sand (water) replacement method, but this method has complex procedures, is time-consuming, laborious and destructive, which restricts the comprehensive detection of the dry density of highway subgrade soil. The wave method and the empirical method can also realize the non-destructive detection of the dry density of highway subgrade soil. Among them, the wave method mainly uses the propagation law of waves in different dry densities of subgrade soil to realize, and the empirical method mainly realizes by constructing the quantitative relationship between the dry density of subgrade soil and parameters such as deflection value, static penetration depth, and resilience modulus; however, due to the detection accuracy of the wave method and the empirical method not meeting the application requirements, there is an urgent need for a new type of non-destructive, fast and accurate detection method. Hyperspectral remote sensing is a new type of land surface information detection technology. Existing research shows that there is a strong internal correlation between soil spectra and soil physical and chemical parameters. By constructing the quantitative relationship between spectra and soil physical and chemical parameters, hyperspectral technology can be applied to the rapid, accurate and non-destructive detection of land surface soil physical and chemical parameters at the regional scale. The application of hyperspectral technology in the soil physical and chemical parameters of non-engineering fields provides a new idea for the monitoring of the dry density of highway subgrade soil, and the data processing and analysis methods can provide certain support for applying hyperspectral technology to the monitoring of the dry density of highway subgrade soil.

[0004] The traditional detection mode of dry density of subgrade soil is destructive to the subgrade, and the sampling points are less, so it is impossible to accurately master the dry density of non-sampling points, and thus it is impossible to timely discover and repair the hidden dangers of the foundation; in order to explore a real-time, non-destructive and accurate detection method for the dry density of highway subgrade soil, the present invention takes hyperspectral technology as the main technical means, analyzes the action mechanism between soil and spectrum before and after compaction, and couples the discrete wavelet algorithm and the partial least squares algorithm to propose a hyperspectral non-destructive detection method for dry density of subgrade soil based on soil compactness coefficient, in order to provide basic theory and method support for the detection of dry density of highway subgrade soil.

[0005] Content of the invention patent

[0006] In order to solve the problems existing in the detection of the dry density of traditional subgrade soil, this application takes drones and hyperspectral as the main technical means to study and analyze the soil components before and after compaction and their effects on the spectrum. By analyzing the subgrade construction process and combining the soil water content, soil dry density, and soil compactness coefficient, a detection technology for the dry density of highway subgrade soil based on the "drone-hyperspectral" technology is determined, and a non-destructive hyperspectral detection method for the dry density of subgrade soil based on the soil compactness coefficient is proposed, providing basic theory and technical support for the detection of subgrade construction quality. The constructed method can meet the accuracy requirements of subgrade construction quality monitoring, and the subgrade is repaired according to the monitoring method provided by the present invention, thereby ensuring the subgrade construction quality and improving the safety performance of the subgrade.

[0007] The technical solution adopted to achieve the above object is as follows:

[0008] A non-destructive hyperspectral detection method for the dry density of subgrade soil based on the soil compactness coefficient includes the following steps:

[0009] a1. The engineering foundation is turned over and loosened, and the soil of the engineering foundation is fully mixed with the filler to achieve a uniform spatial distribution of the physical and chemical parameters of the engineering foundation soil;

[0010] b2. Use a quadcopter drone to obtain the hyperspectral image of the processed engineering foundation in step a1, and perform smoothing and denoising processing to generate a remote sensing product that can be used to monitor the soil water content of the engineering foundation, denoted as H_Water;

[0011] c3. Collect the subgrade soil in the construction area and construct model parameters in combination with indoor experiments. The construction of model parameters includes the following steps:

[0012] c3.1 Compaction test: Place the compaction instrument steadily on a rigid foundation, firmly connect the compaction cylinder and the base, install the protection cylinder, and evenly apply a layer of lubricating oil on the inner wall of the compaction cylinder; Weigh a certain amount of the sample, pour it into the compaction cylinder, and compact it in layers. It is divided into 3 layers, with 25 blows for each layer; The height of each layer of the sample should be equal, and the soil surface at the junction of the two layers should be roughened. When the compaction is completed, the height of the sample exceeding the top of the compaction cylinder should be less than 6 mm; Remove the protection cylinder, use a straight scraper to level the sample on the top of the compaction cylinder, remove the bottom plate, and if the bottom of the sample exceeds the outside of the cylinder, it should also be leveled. Wipe the outer wall of the compaction cylinder clean, and accurately weigh the total mass of the cylinder and the sample to 1 g, and calculate the soil water density of the sample;

[0013] c3.2 Spectral determination. In the soil spectral determination experiment, spectral data of the soil surface before and after compaction are measured. An ASD handheld ground object spectrometer is selected to collect spectral data of soils with different moisture contents. Before measurement, the spectrometer is preheated, and optimized with a calibrated whiteboard. After meeting the established standards, spectral measurement of the soil surface is carried out. Each spectral determination must collect more than 10 spectra and average them as the spectral data for this determination.

[0014] c3.3 Calculation of key parameters of soil compactness. Based on the spectral data collected in steps c3.1 and c3.2 and the experimental data of soil sample weight, soil moisture content and soil dry density are calculated; the soil compactness coefficient is calculated using the spectral data.

[0015] c3.4 Information extraction and resampling processing of spectral data. The discrete wavelet algorithm is used to process the soil compactness coefficient generated in step c3.3 to achieve the aggregation of effective information and the reduction of spectral resolution.

[0016] c3.5 Correlation analysis. The correlation coefficient is obtained by performing correlation analysis on the soil compactness coefficient generated in step c3.4 and the soil dry density in step c3.3, using the following formula:

[0017]

[0018] where r j is the correlation coefficient, X i is the compactness coefficient of the i-th sample in the j-th band, Y i is the soil dry density of the sample, and n is the total number of samples; X A is the average value of the compactness coefficients of all samples in the j-th band, and Y A is the average value of the maximum dry densities of all samples in the j-th band.

[0019] c3.6 Extraction of characteristic parameters. Based on the influence characteristics of external factors on the spectrum, combined with the correlation coefficient between the soil compactness coefficient and the soil dry density in step c3.5, characteristic parameters are extracted according to the principle of the largest square of the correlation coefficient, and the square of the correlation coefficient should be greater than 0.5. The selection interval of the characteristic parameters is limited within [350, 1350].

[0020] c3.7 Based on the characteristic parameters extracted in c3.6, a detection model for the dry density of subgrade soil is constructed using the partial least squares algorithm, and the detection accuracy is evaluated using evaluation indexes; among them, the determination coefficient and the root mean square error are selected as the evaluation indexes, and their calculation formulas are as follows:

[0021]

[0022] Wherein, SUM is the measured dry density of the soil, SMP is the predicted value based on the soil dry density estimation model, and SM A is the mean value of the measured dry density of the soil;

[0023] d4. Compact the engineering foundation, and use the sand replacement method to measure the dry density of the soil of the engineering foundation. Verify the estimation results based on the "UAV - hyperspectral" technology through the measured values. When R 2 is above 0.9 and RMSE is below 0.03, the subgrade soil dry density detection model meets the verification requirements, and the best estimation model is obtained;

[0024] e5. Use a quadcopter UAV to obtain the hyperspectral image of the engineering foundation, and perform smoothing and denoising processing to generate a remote sensing product that can be used to monitor the water content of the soil after the engineering foundation is compacted, denoted as H_Soil;

[0025] f6. Based on the hyperspectral image products generated in step b2 and step e5, construct a soil compactness coefficient image, and use the wavelet algorithm to process the soil compactness image pixel by pixel to generate a series of soil compactness decomposition images. Then, input the images of the corresponding characteristic parameters obtained in step c3 into the model obtained in step d4 to estimate the dry density of the soil during subgrade construction.

[0026] Furthermore, in the said step a1, for the loosening treatment, first turn the soil to the left, then to the right, and then to the middle, and the number of turning times is not less than 3 times to achieve the uniformity of the physical and chemical parameters of the soil of the engineering foundation in the spatial distribution.

[0027] Furthermore, in the said step b2 and step e5, the hyperspectral images are regarded as a set of spectral curves with the same spectral resolution and the same interval, and the spectral intervals have the characteristic of spatial continuity. The processing of the hyperspectral images is divided into 2 steps, including:

[0028] h2.1 Use a low-pass filter with a Hamming window length of 9 to smooth and denoise each spectral curve,

[0029] SM = [0.0800, 0.2147, 0.5400, 0.8653, 1.0000, 0.8653, 0.5400, 0.2147, 0.0800];

[0030] h2.2 Use a 3X3 Gaussian low-pass filter to process each band of the hyperspectral image respectively to achieve smoothing and denoising processing in the spatial dimension.

[0031] Furthermore, in the said step c3.3, the following formula is used to calculate the soil water content:

[0032] Place the compacted soil in an oven and dry it to a constant weight at 107°C, and then weigh its dry weight respectively to calculate the soil water content. The calculation formula is as follows:

[0033]

[0034] Wherein, WC is the soil water content, and W fresh is the fresh weight, and W dry is the dry weight;

[0035] In the step c3.3, the following formula is adopted to calculate the soil dry density:

[0036]

[0037] Wherein, ρ d is the soil dry density, ρ0 is the density containing soil water, and WC is the soil water content (%) of a sample at a certain point.

[0038] Furthermore, the calculation formula for the soil compactness coefficient in the step c3.3 is:

[0039]

[0040] Wherein C s is the soil compactness coefficient, C w is the water influence coefficient, R i is the spectrum of the soil with the uncompacted water content of i, R k is the spectral reflectance of the soil with the uncompacted water content of k, R i_j is the spectrum of the soil with the dry density of j and the water content of i.

[0041] Furthermore, the principles for selecting characteristic parameters in the step c3.6 include:

[0042] When the resolution is 1 nm, the band position interval of the characteristic parameters is more than 50 nm; when the resolution is 2 nm, the band position interval of the characteristic parameters is more than 60 nm; when the resolution is 4 nm, the band position interval of the characteristic parameters is more than 80 nm; when the resolution is 8 nm, the band position interval of the characteristic parameters is more than 90 nm; when the resolution is 16 nm, the band position interval of the characteristic parameters is more than 100 nm; when the resolution is 32 nm, the band position interval of the characteristic parameters is more than 120 nm; when the resolution is 64 nm, the band position interval of the characteristic parameters is more than 150 nm.

[0043] Furthermore, the wavelet algorithm adopted in the step f6 is the discrete wavelet algorithm. Regarding the hyperspectral image as being composed of spectral curves that are finite and of equal length, without considering the spatial position correlation between the spectral curves, the adopted processing flow includes:

[0044] f6.1 Number the spectra forming the hyperspectral image according to their spatial positions in the image as 1, 2, 3,..., n;

[0045] f6.2 Use the discrete wavelet algorithm to process the spectral curves numbered 1 - n respectively, and the decomposition scale can be adjusted by itself;

[0046] f6.3 Separate the information of each scale decomposed by the discrete wavelet algorithm according to the decomposition scale to generate datasets for each scale; f6.4 Re - organize the images of each scale within the datasets for each scale according to the numbering.

[0047] The beneficial effects of this invention patent are as follows:

[0048] (1) This application is a non - destructive hyperspectral detection method for the dry density of subgrade soil based on the soil compactness coefficient, with unmanned aerial vehicles (UAVs) and hyperspectral as the main technical means. Through the analysis of the interaction mechanism between soil and spectrum before and after compaction, and by coupling the discrete wavelet algorithm and the partial least squares algorithm, combined with soil moisture content, soil dry density, and soil compactness coefficient, a detection model for the dry density of subgrade soil is constructed, and the best estimation model is obtained through verification; This invention can realize the rapid, non - destructive, and real - time detection of the dry density of subgrade soil with UAV - hyperspectral technology, thus realizing the comprehensive monitoring of the dry density of subgrade soil, obtaining the overall information of the dry density of subgrade soil, effectively making up for the deficiencies of traditional methods for measuring soil dry density, providing basic theory and technical support for the quality detection of subgrade construction, and the constructed method can meet the accuracy requirements of subgrade construction quality monitoring, and the subgrade can be repaired according to the monitoring method provided by this invention, thus ensuring the quality of subgrade construction and improving the safety performance of the subgrade.

[0049] (2) This invention can help to discover the parts with insufficient subgrade compaction, thus improving the compaction quality of the subgrade, enhancing the stability and safety of the subgrade, and reducing the occurrence of highway diseases.

[0050] (3) This invention can realize the detection of the dry density of subgrade soil of the highways under construction in the region, effectively improve the detection efficiency of subgrade, and save the construction cost of subgrade.

[0051] (4) This invention can also be used for the monitoring of the dry density of foundation soil of other projects, and also provides a basic method and theoretical support for the monitoring of the dry density of foundation soil of other projects. Description of the Drawings

[0052] Figure 1 is the flowchart of the method of this invention;

[0053] Figure 2 is the spectral analysis diagram of the soil before compaction in the patent;

[0054] Figure 3 is the spectral analysis diagram of the soil after compaction in the patent;

[0055] Figure 4 is the soil compactness coefficient diagram in the patent;

[0056] Figure 5 It is the correlation analysis diagram of soil compactness coefficient and soil dry density in the patent;

[0057] Figure 6 It is the result of the soil dry density estimation model constructed based on the discrete wavelet algorithm in the patent;

[0058] Figure 7 It is based on Figure 6 The scatter plot of the estimation model constructed by D1 in it. Specific implementation mode

[0059] The following further describes the present invention patent with reference to the accompanying drawings.

[0060] In this application, the soil spectrum is the result of the overall action of soil physical and chemical parameters within the sensor's field of view. When the non-aqueous components of the soil remain unchanged, soil water content and soil compactness are the main factors affecting the soil spectrum. Therefore, the soil spectrum in any state can be regarded as the product of the dry soil spectrum and the sum of the soil humidity coefficient and the soil compactness coefficient, where the soil humidity coefficient is the ratio of the spectrum of soil water content to the dry soil spectrum, and the soil compactness coefficient is the ratio of the spectrum of the compacted soil (after drying) to the dry soil before compaction; based on the analysis of the action mechanism between the spectrum and soil components, the present invention proposes a non-destructive detection method for the dry density of subgrade (compacted) soil based on the soil compactness coefficient using an unmanned aerial vehicle hyperspectral.

[0061] According to the above principle, as Figure 1 shown, this application proposes a non-destructive detection method for the dry density of subgrade soil based on the soil compactness coefficient using hyperspectral, including the following steps:

[0062] a1. Turn over and loosen the engineering foundation, and fully mix the engineering foundation soil with the filler to achieve a uniform spatial distribution of the physical and chemical parameters of the engineering foundation soil;

[0063] b2. Use a quadcopter unmanned aerial vehicle to obtain the hyperspectral image of the engineering foundation processed in step a1, and perform smoothing and denoising processing to generate a remote sensing product that can be used to monitor the soil water content of the engineering foundation, denoted as H_Water. The parameters during the shooting process of the multi-rotor unmanned aerial vehicle should meet the basic requirements of aerial photography; as Figure 2 shown is the spectral analysis diagram of the remote sensing product of the soil water content of the engineering foundation before soil compaction.

[0064] c3. Collect the subgrade soil in the construction area and construct model parameters in combination with indoor experiments. The construction of model parameters includes the following steps:

[0065] c3.1 Hammer test: Place the compaction apparatus steadily on a rigid foundation, firmly connect the compaction cylinder to the base, install the protective cylinder, and evenly apply a layer of lubricating oil on the inner wall of the compaction cylinder. Weigh a certain amount of the sample, pour it into the compaction cylinder, and compact it in layers. There are 3 layers, with 25 blows for each layer. The height of each layer of the sample should be equal, and the soil surface at the junction of the two layers should be roughened. When the compaction is completed, the height of the sample exceeding the top of the compaction cylinder should be less than 6 mm. Remove the protective cylinder, use a straight scraper to level the sample on the top of the compaction cylinder, remove the bottom plate, and if the bottom of the sample extends beyond the cylinder, it should also be leveled. Wipe the outer wall of the compaction cylinder clean, accurately weigh the total mass of the cylinder and the sample to 1 g, and calculate the soil water content density of the sample.

[0066] c3.2 Spectral determination: The soil spectral determination experiment measures the spectral data on the surface of the soil before and after compaction. Select an ASD handheld ground object spectrometer to collect the spectral data of soils with different water contents. Before measurement, preheat the spectrometer first and optimize it with a calibrated whiteboard. After reaching the established standard, carry out the spectral measurement on the soil surface. Each spectral determination must collect more than 10 spectra and take the average as the spectral data for this determination.

[0067] c3.3 Calculation of key parameters of soil compactness: Based on the spectral data and experimental data of soil sample weights collected in steps c3.1 and c3.2, calculate the soil water content and soil dry density; calculate the soil compactness coefficient using the spectral data.

[0068] c3.4 Information extraction and resampling processing of spectral data: Use the discrete wavelet algorithm to process the soil compactness coefficient generated in step c3.3 to achieve the aggregation of effective information and the reduction of spectral resolution, as Figure 4 shown;

[0069] c3.5 Correlation analysis: Conduct a correlation analysis on the soil compactness coefficient generated in step c3.4 and the soil dry density in step c3.3 to obtain the correlation coefficient. Use the following formula:

[0070]

[0071] In the formula, r j is the correlation coefficient, X i is the soil compactness coefficient of the i-th sample in the j-th band, Y i is the soil dry density of the sample, and n is the total number of samples; X A is the average value of the soil compactness coefficients of all samples in the j-th band, and Y A is the average value of the soil dry densities of all samples in the j-th band.

[0072] Extraction of c3.6 characteristic parameters: Based on the influence characteristics of external factors on the spectrum, combined with the correlation coefficient between the soil compaction coefficient and the soil dry density in step c3.5, according to the principle of the largest square of the correlation coefficient, and the square of the correlation coefficient should be greater than 0.5 to extract the characteristic parameters, and the selection range of the characteristic parameters is limited within [350, 1350], such as Figure 5 as shown;

[0073] c3.7 Based on the characteristic parameters extracted in c3.6, use the partial least squares algorithm to construct a detection model for the soil dry density of the subgrade, and use evaluation indicators to evaluate the detection accuracy; among them, the determination coefficient and the root mean square error are selected as the evaluation indicators, and their calculation formulas are as follows:

[0074]

[0075] In the formula, SUM is the measured soil dry density, SMP is the predicted value based on the soil dry density estimation model, and SM A is the mean value of the measured soil dry density;

[0076] d4. Compact the engineering foundation, and use the sand replacement method to measure the soil dry density of the engineering foundation. Verify the estimation results based on the "UAV - hyperspectral" technology through the measured values. When R 2 is above 0.9 and RMSE is below 0.03, the detection model of the soil dry density of the subgrade meets the verification requirements, and the best estimation model is obtained; as Figure 6 shown, after verification and analysis, only D1 and D2 meet the verification requirements, as Figure 7 shown is the scatter plot of the estimation model constructed based on D1.

[0077] e5. Use a quadcopter UAV to obtain the hyperspectral image of the engineering foundation, and perform smoothing and denoising processing to generate a remote sensing product that can be used to monitor the soil water content after the engineering foundation is compacted, denoted as H_Soil; as Figure 3 shown is the spectral analysis diagram of the remote sensing product of the soil water content of the engineering foundation after soil compaction.

[0078] f6. Based on the hyperspectral image products generated in step b2 and step e5, construct a soil compaction coefficient image, and use the wavelet algorithm to process the soil compaction image pixel by pixel to generate a series of soil compaction decomposition images, and then input the images of the corresponding characteristic parameters obtained in step c3 into the model obtained in step d4 to estimate the soil dry density during subgrade construction.

[0079] Since the uniformity of the distribution of soil and fill in the foundation is one of the important factors affecting the detection of soil dry density, it is necessary to treat the soil and fill in the engineering foundation. In step a1 of this application, the engineering foundation is turned over and loosened. When loosening the soil, it is first turned over to the left, then to the right, and then to the middle, and the number of turnovers is not less than 3 times, so as to fully mix the soil and fill in the engineering foundation and achieve the uniformity of the physical and chemical parameters of the soil in the engineering foundation in the spatial distribution.

[0080] In this application, in steps b2 and e5, the hyperspectral image is regarded as a set of spectral curves with the same spectral resolution and the same interval, and the spectral intervals have the characteristic of spatial continuity. The hyperspectral image processing is divided into 2 steps, including:

[0081] h2.1 Use a low-pass filter with a Hamming window length of 9 to smooth and denoise each spectral curve.

[0082] SM = [0.0800, 0.2147, 0.5400, 0.8653, 1.0000, 0.8653, 0.5400, 0.2147, 0.0800];

[0083] h2.2 Use a 3X3 Gaussian low-pass filter to process each band of the hyperspectral image respectively to achieve spatial dimension smoothing and denoising.

[0084] 0.0007 0.0256 0.0007 0.0256 0.8948 0.0256 0.0007 0.0256 0.0007

[0085] In this application, the following formula is used to calculate the soil water content in step c3.3:

[0086] Place the compacted soil in an oven and dry it at 107°C to a constant weight, weigh its dry weight respectively, and calculate the soil water content. The calculation formula is as follows:

[0087]

[0088] In the formula, WC is the soil water content, W fresh is the fresh weight, and W dry is the dry weight;

[0089] The following formula is used to calculate the soil dry density in step c3.3:

[0090]

[0091] In the formula, ρ d is the soil dry density, ρ0 is the soil density with water content, and WC is the soil water content (%) of a certain point sample.

[0092] When the non-aqueous components of the soil remain unchanged, soil water content and soil compaction are the main factors affecting soil spectra. However, the influence of soil water content on soil spectra is too large, causing significant interference to the spectral response of soil compaction. To analyze the detection technology of dry density of compacted soil, the calculation formula for the soil compaction coefficient in step c3.3 of the present invention is as follows:

[0093]

[0094] In the formula, C s is the soil compaction coefficient, C w is the water influence coefficient, R i is the soil spectrum with water content i before compaction, R k is the soil spectral reflectance with water content k before compaction, R i_j is the soil spectrum with water content i and dry density j after drying.

[0095] In this application, the soil compaction coefficient C s The reasoning process is as follows:

[0096]

[0097] R i_j =R k *C w +R k *C s

[0098]

[0099] In the formula, C s is the influence of soil particle compactness on soil spectra, C w is the water influence coefficient, R i is the soil spectrum with water content i before compaction, R k is the soil spectral reflectance with water content k before compaction, R j is the soil spectrum after drying with dry density j, R i_j is the soil spectrum with water content i and dry density j.

[0100] In this application, the principle of selecting characteristic parameters in step c3.6 also includes:

[0101] When the resolution is 1 nm, the band position interval of the characteristic parameters is more than 50 nm; when the resolution is 2 nm, the band position interval of the characteristic parameters is more than 60 nm; when the resolution is 4 nm, the band position interval of the characteristic parameters is more than 80 nm; when the resolution is 8 nm, the band position interval of the characteristic parameters is more than 90 nm; when the resolution is 16 nm, the band position interval of the characteristic parameters is more than 100 nm; when the resolution is 32 nm, the band position interval of the characteristic parameters is more than 120 nm; when the resolution is 64 nm, the band position interval of the characteristic parameters is more than 150 nm.

[0102] In this application, the wavelet algorithm adopted in step f6 is the discrete wavelet algorithm. The hyperspectral image is regarded as composed of spectral curves that are finite and of equal length, without considering the spatial position correlation between the spectral curves. The processing flow adopted includes:

[0103] f6.1 Number the spectra that make up the hyperspectral image according to their spatial positions in the image, 1, 2, 3,..., n;

[0104] f6.2 Use the discrete wavelet algorithm to process the spectral curves numbered 1 - n respectively, and the decomposition scale can be adjusted by itself;

[0105] f6.3 Separate the information of each scale decomposed by the discrete wavelet algorithm according to the decomposition scale to generate datasets of each scale;

[0106] f6.4 Recombine the images of each scale according to the numbering within the datasets of each scale.

[0107] This embodiment does not impose any formal restrictions on the shape, material, structure, etc. of the invention patent. Any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the invention patent all belong to the protection scope of the technical solution of the invention patent.

[0108] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the invention patent, rather than to limit it; although the invention patent has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features, but these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the invention patent.

Claims

1. A hyperspectral non-destructive detection method for the dry density of subgrade soil based on the soil compaction coefficient, characterized in that, It includes the following steps: a1. Turn over and loosen the engineering foundation, and fully mix the engineering foundation soil with the filler to achieve uniform spatial distribution of the physical and chemical parameters of the engineering foundation soil; b2. Use a quadrotor UAV to obtain the hyperspectral image of the engineering foundation processed in step a1, and perform smoothing and denoising processing to generate a remote sensing product for monitoring the water content of the engineering foundation soil, denoted as H_Water; c3. Collect the subgrade soil in the construction area and construct model parameters in combination with indoor experiments. The construction of model parameters includes the following steps: c3.1 Compaction test: Place the compactor steadily on a rigid foundation, firmly connect the compaction cylinder to the base, install the casing, and evenly apply a layer of lubricating oil on the inner wall of the compaction cylinder; Weigh a quantitative sample, pour it into the compaction cylinder, and compact it in layers, divided into 3 layers, with 25 blows for each layer; The height of each layer of the sample should be equal, and the soil surface at the junction of the two layers should be roughened. When the compaction is completed, the height of the sample exceeding the top of the compaction cylinder should be less than 6 mm; Remove the casing, use a straight scraper to level the sample on the top of the compaction cylinder, remove the bottom plate, and if the bottom of the sample extends outside the cylinder, it should also be leveled. Wipe the outer wall of the compaction cylinder clean, and accurately measure the total mass of the cylinder and the sample to 1 g, and calculate the water content density of the sample; c3.2 Spectral measurement: The soil spectral measurement experiment measures the spectral data on the surface of the soil before and after compaction. Select an ASD handheld ground object spectrometer to collect the spectral data of soils with different water contents. Preheat the spectrometer before measurement and optimize it with a calibrated whiteboard. After reaching the established standard, carry out the spectral measurement of the soil surface. Each spectral measurement must collect more than 10 spectra and take the average as the spectral data for this measurement; c3.3 Calculation of key parameters of soil compactness: Based on the spectral data and experimental data of soil sample weights collected in steps c3.1 and c3.2, calculate the soil water content and soil dry density; Calculate the soil compactness coefficient using spectral data; c3.4 Information extraction and resampling processing of spectral data: Use the discrete wavelet algorithm to process the soil compactness coefficient generated in step c3.3 to achieve the aggregation of effective information and the reduction of spectral resolution; c3.5 Correlation analysis: Conduct correlation analysis on the soil compactness coefficient generated in step c3.4 and the soil dry density in step c3.3 to obtain the correlation coefficient, using the following formula: where r j is the correlation coefficient, X i is the compactness coefficient of the i-th sample in the j band, Y i is the dry density of the soil of the sample, and n is the total number of samples; X A is the average value of the compactness coefficients of all samples in the j band, and Y A is the average value of the maximum dry densities of all samples in the j band; c3.6 Extraction of characteristic parameters: Based on the influence characteristics of external factors on the spectrum, combined with the correlation coefficient between the soil compactness coefficient and the soil dry density in step c3.5, extract characteristic parameters according to the principle of the largest square of the correlation coefficient, and the square of the correlation coefficient should be greater than 0.

5. The selection range of the characteristic parameters is limited within [350, 1350]; c3.7 Based on the characteristic parameters extracted in c3.6, use the partial least squares algorithm to construct a subgrade soil dry density detection model, and evaluate the detection accuracy using evaluation indicators; Among them, the evaluation indicators select the coefficient of determination and the root mean square error as evaluation indicators, and their calculation formulas are as follows: Wherein, SUM is the measured soil dry density, SMP is the predicted value based on the soil dry density estimation model, and SM A is the mean value of the measured soil dry density; d4. Compact the engineering foundation, and use the sand replacement method to measure the dry density of the soil in the engineering foundation. Verify the estimation results based on the "UAV - hyperspectral" technology through the measured values. When R 2 is above 0.9 and the RMSE is lower than 0.03, the subgrade soil dry density detection model meets the verification requirements, and the best estimation model is obtained; e5. Use a quadrotor UAV to obtain the hyperspectral image of the engineering foundation, and perform smoothing and denoising processing to generate a remote sensing product that can be used to monitor the soil moisture content of the compacted engineering foundation, denoted as H_Soil; f6. Based on the hyperspectral image products generated in step b2 and step e5, construct a soil compactness coefficient image, and use the wavelet algorithm to process the soil compactness image pixel by pixel to generate a series of soil compactness decomposition images. Then, input the images of the corresponding characteristic parameters obtained in step c3 into the model obtained in step d4 to estimate the dry density of the subgrade construction soil.

2. The non-destructive hyperspectral detection method for the dry density of subgrade soil based on the soil compaction coefficient according to claim 1, wherein, In the said step a1, for the loosening treatment, first turn the soil to the left, then to the right, and then to the middle, and the number of turning times is not less than 3 times to achieve the uniformity of the physical and chemical parameters of the engineering foundation soil in the spatial distribution.

3. The method for non-destructive hyperspectral detection of the dry density of subgrade soil based on the soil compaction coefficient according to claim 2, wherein In the said step b2 and step e5, the hyperspectral image is regarded as a set of spectral curves with the same spectral resolution and the same interval, and the spectral intervals have the characteristic of spatial continuity. The hyperspectral image processing is divided into 2 steps, including: h2.1 Use a low-pass filter with a Hamming window length of 9 to smooth and denoise each spectral curve. SM = [0.0800, 0.2147, 0.5400, 0.8653, 1.0000, 0.8653, 0.5400, 0.2147, 0.0800]; h2.2 Use a 3X3 Gaussian low-pass filter to process each band of the hyperspectral image respectively to achieve spatial dimension smoothing and denoising processing.

4. The non-destructive hyperspectral detection method for the dry density of subgrade soil based on the soil compaction coefficient according to claim 3, characterized in that, In the said step c3.3, the following formula is used to calculate the soil moisture content: Place the compacted soil in an oven and dry it to a constant weight at 107 °C, and then weigh its dry weight respectively to calculate the soil moisture content. The calculation formula is as follows: where WC is the soil water content, W fresh is the fresh weight, and W dry is the dry weight; In the said step c3.3, the following formula is used to calculate the dry density of the soil: where ρ d is the dry density of the soil, ρ0 is the density of the soil with water content, and WC is the water content of the soil sample at a certain point (%).

5. The non-destructive hyperspectral detection method for the dry density of subgrade soil based on the soil compactness coefficient according to claim 4, characterized in that, In the said step c3.3, the calculation formula of the soil compactness coefficient is: where C s is the soil compaction coefficient, C w is the moisture influence coefficient, R i is the soil spectrum with an uncompacted water content of i, R k is the soil spectral reflectance with an uncompacted water content of k, R i_j is the soil spectrum with a dry density of j and a water content of i.

6. The non-destructive hyperspectral detection method for the dry density of subgrade soil based on the soil compaction coefficient according to claim 5, wherein In the said step c3.6, the principles for selecting characteristic parameters include: When the resolution is 1 nm, the band position interval of the characteristic parameters is more than 50 nm; when the resolution is 2 nm, the band position interval of the characteristic parameters is more than 60 nm; when the resolution is 4 nm, the band position interval of the characteristic parameters is more than 80 nm; when the resolution is 8 nm, the band position interval of the characteristic parameters is more than 90 nm; when the resolution is 16 nm, the band position interval of the characteristic parameters is more than 100 nm; when the resolution is 32 nm, the band position interval of the characteristic parameters is more than 120 nm; when the resolution is 64 nm, the band position interval of the characteristic parameters is more than 150 nm.

7. The non-destructive hyperspectral detection method for the dry density of subgrade soil based on the soil compaction coefficient according to claim 6, characterized in that, In the said step f6, the wavelet algorithm used is the discrete wavelet algorithm. The hyperspectral image is regarded as composed of finite and equal-length spectral curves, and the spatial position correlation between the spectral curves is not considered. The processing flow used includes: f6.1 Number the spectra that make up the hyperspectral image according to their spatial positions in the image, 1, 2, 3,..., n; f6.2 Use the discrete wavelet algorithm to process the spectral curves numbered 1 - n respectively, and the decomposition scale can be adjusted by itself; f6.3 Separate the information of each scale decomposed by the discrete wavelet algorithm according to the decomposition scale to generate datasets of each scale. Within each scale dataset of f6.4, reorganize the images of each scale according to the numbers.

Images