Spectrum-based pollutant concentration fitting method, unmanned intelligent device and application

Abstract

The application relates to the technical field of water quality monitoring, in particular to a pollutant concentration fitting method based on spectrum, an unmanned intelligent device and application, a plurality of target grid regions are screened out through the pollutant influence similarity of every adjacent two grid regions, meanwhile, sampling redundancy of spectrum data and sampling omission of spectrum data are avoided, and the accuracy of pollutant concentration fitting of sampling spectrum is improved; a first similarity index between a blank grid region and a target grid region is obtained through the pollutant influence similarity of the blank grid region and the target grid region, the accuracy of reconstructed spectrum data of the blank region is improved, and the accuracy of pollutant concentration fitting of sampling spectrum is improved; the key target grid region is resampled through the reconstructed spectrum data of the blank grid region, a fusion fitting model is obtained based on direct inversion and indirect inversion, the interference on spectrum data is reduced, and the accuracy of pollutant concentration fitting of sampling spectrum is improved.

Description

Technical Field

[0001] This application relates to the field of water quality monitoring technology, and in particular to a spectral-based pollutant concentration fitting method, unmanned intelligent equipment, and its application. Background Technology

[0002] Spectrum-based pollutant concentration fitting is a technique that quantitatively inverts pollutant concentration by analyzing the characteristic spectra generated after the interaction of substances with light. It has important applications in environmental monitoring, industrial process control and other fields.

[0003] Currently, environmental monitoring technologies do not fully consider the similarity of pollutant concentration influencing factors in different grid areas, leading to the following problems: First, sampling redundancy occurs because spectral data are collected from multiple grid areas with similar pollutant concentrations; second, sampling omission occurs because spectral data are not collected from grid areas with unique pollutant concentration influencing factors; third, during the reconstruction of spectral data in unsampled blank areas, interpolation is performed only based on the spatial distance between two grid areas, which is not conducive to improving the accuracy of fitting pollutant concentrations using sampled spectra. Summary of the Invention

[0004] In view of the above problems, this application provides a method for fitting pollutant concentration based on spectrum, an unmanned intelligent device and application, to solve the above-mentioned technical problems that are not conducive to improving the accuracy of fitting pollutant concentration based on sampled spectrum.

[0005] In a first aspect, embodiments of this application provide a method for fitting pollutant concentrations based on spectral data, comprising:

[0006] The target monitoring area is divided into multiple grid areas. Based on the similarity of pollutant impact between any two adjacent grid areas, multiple target grid areas are selected from the multiple grid areas. At least some of the multiple target grid areas are key target grid areas.

[0007] Collect spectral data of the target grid region to obtain pollutant concentration data of the key target grid region;

[0008] The first similarity index between the blank grid region and the target grid region is obtained based on the similarity of pollutant impact and spatial similarity between the blank grid region and the target grid region. The reconstructed spectral data of the blank grid region is obtained based on the spectral data of the target grid region and the first similarity index between the blank grid region and the target grid region. The blank grid region is any other grid region among the plurality of grid regions except the target grid region.

[0009] Based on the reconstructed spectral data of the blank grid region, the key target grid region is resampled to obtain the resampled spectral data of the key target grid region;

[0010] A first fitting model is established based on the spectral data and pollutant concentration data of the key target grid region; a second fitting model is established based on the resampled spectral data and pollutant concentration data of the key target grid region; and a fusion fitting model is obtained based on the first fitting model and the second fitting model.

[0011] Based on the fusion fitting model, pollutant concentration data for other grid areas in the grid region, excluding the key target grid area, are obtained.

[0012] Secondly, embodiments of this application provide an unmanned intelligent device, including a processor and a memory coupled to the processor, the memory storing program instructions executable by the processor; when the processor executes the program instructions stored in the memory, it implements the above-mentioned spectral-based pollutant concentration fitting method.

[0013] Thirdly, embodiments of this application provide a spectroscopic pollutant concentration fitting device, comprising:

[0014] A grid division module is used to divide the target monitoring area into multiple grid regions, and to select multiple target grid regions from the multiple grid regions based on the similarity of pollutant impact between each pair of adjacent grid regions, wherein at least some of the multiple target grid regions are key target grid regions;

[0015] The data acquisition module is used to collect spectral data of the target grid area and obtain pollutant concentration data of the key target grid area;

[0016] The reconstruction module is used to obtain a first similarity index between the blank grid area and the target grid area based on the similarity of pollutant influence and spatial similarity between the blank grid area and the target grid area, and to obtain reconstructed spectral data of the blank grid area based on the spectral data of the target grid area and the first similarity index between the blank grid area and the target grid area, wherein the blank grid area is other grid areas besides the target grid area among the plurality of grid areas;

[0017] The resampling module is used to resample the key target grid region based on the reconstructed spectral data of the blank grid region, and obtain the resampled spectral data of the key target grid region;

[0018] The fitting module is used to establish a first fitting model based on the spectral data and pollutant concentration data of the key target grid area, establish a second fitting model based on the resampled spectral data and pollutant concentration data of the key target grid area, and obtain a fusion fitting model based on the first fitting model and the second fitting model.

[0019] The inversion module is used to obtain pollutant concentration data of other grid areas in the grid region, excluding the key target grid area, based on the fusion fitting model.

[0020] Fourthly, embodiments of this application provide a computer-readable storage medium storing program instructions that, when executed by a processor, implement the aforementioned spectral-based pollutant concentration fitting method.

[0021] The spectral-based pollutant concentration fitting method, unmanned intelligent device, and application provided in this application's embodiments select multiple target grid regions by screening the similarity of pollutant influence between every two adjacent grid regions. This avoids sampling redundancy and omissions in spectral data, thus improving the accuracy of pollutant concentration fitting based on sampled spectra. A first similarity index is obtained between blank grid regions and target grid regions based on the similarity of pollutant influence. Reconstructing the spectral data of blank regions based on this first similarity index improves the accuracy of the reconstructed spectral data, further enhancing the accuracy of pollutant concentration fitting based on sampled spectra. Resampling of key target grid regions using the reconstructed spectral data from blank grid regions, and obtaining a fusion fitting model based on direct inversion of spectral data and pollutant concentration data from key target grid regions, as well as indirect inversion based on resampled spectral data and pollutant concentration data from key target grid regions, reduces the interference of spectral data sampling device posture and ambient light changes on spectral data, further improving the accuracy of pollutant concentration fitting based on sampled spectra.

[0022] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0023] Figure 1 A schematic flowchart of the spectroscopic pollutant concentration fitting method provided in the embodiments of this application is shown.

[0024] Figure 2 An example diagram of grid division and path is shown in the spectrum-based pollutant concentration fitting method provided in the embodiments of this application.

[0025] Figure 3 The diagram illustrates the principle of the fusion fitting model in the spectroscopic pollutant concentration fitting method provided in this application embodiment.

[0026] Figure 4 An example of a GIS thematic map of pollutant concentration data in the spectral-based pollutant concentration fitting method provided in this application embodiment is shown.

[0027] Figure 5 A schematic diagram of the structure of the spectroscopic pollutant concentration fitting device provided in an embodiment of this application is shown.

[0028] Figure 6 A schematic diagram of the structure of the unmanned intelligent device provided in an embodiment of this application is shown.

[0029] Figure 7 A schematic diagram of the structure of a computer-readable storage medium provided in an embodiment of this application is shown. Detailed Implementation

[0030] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0031] To enable those skilled in the art to better understand the solutions of this application, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0032] In the embodiments of this application, it should be noted that, in this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0033] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0034] In the description of the embodiments of this application, the words "example" or "for example" are used to indicate exemplification, illustration, or description. Any embodiment or design described as "example" or "for example" in the embodiments of this application is not to be construed as being more preferred or having more advantages than another embodiment or design. The use of the words "example" or "for example" is intended to present relative concepts in a clear manner.

[0035] Furthermore, in the embodiments of this application, "multiple" refers to two or more. Therefore, in the embodiments of this application, "multiple" can also be understood as "at least two". "At least one" can be understood as one or more, such as one, two, or more. For example, including at least one means including one, two, or more, and is not limited to which ones are included. For example, including at least one of A, B, and C, then it could include A, B, C, A and B, A and C, B and C, or A and B and C.

[0036] It should be noted that in the embodiments of this application, "and / or" describes the relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. In addition, the character " / ", unless otherwise specified, generally indicates that the associated objects before and after it are in an "or" relationship.

[0037] It should be noted that in the embodiments of this application, "connection" can be understood as electrical connection. The connection between two electrical components can be a direct or indirect connection between the two electrical components. For example, the connection between A and B can be a direct connection between A and B, or an indirect connection between A and B through one or more other electrical components.

[0038] One embodiment of this application provides a method for fitting pollutant concentrations based on spectral data. Please refer to [link to relevant documentation]. Figure 1 As shown, the method may include the following steps S11 to S16:

[0039] Step S11: Divide the target monitoring area into multiple grid areas. Based on the similarity of pollutant impact between any two adjacent grid areas, select multiple target grid areas from the multiple grid areas. Among these, at least some of the multiple target grid areas are key target grid areas.

[0040] Dividing the target monitoring area into a grid, abstracting it into a regular grid, provides a spatial reference for spectral data acquisition. The number of grid regions can be a natural number. The grid area can be denoted as , The mesh generation results can be as follows: Figure 2 As shown, two adjacent grid regions can be Figure 2Two adjacent grid regions on the left and right can also be Figure 2 Two adjacent grid regions, one above the other.

[0041] As one implementation method, the size of the grid area can be determined based on the minimum resolution of existing pollutant concentration impact data, and the size of the grid area is no larger than the minimum resolution of the pollutant concentration impact data. The pollutant concentration impact data can be the value of at least one influencing factor that affects the pollutant concentration in the grid area. Specifically, readily available data can be preferred as the pollutant concentration impact data. For water bodies, the pollutant concentration impact data can include temperature values, depth values, and distance from the discharge outlet. For example, the target monitoring area is the entire shallow lake or at least a portion of a continuous area of ​​a shallow lake, and the pollutant concentration impact data for the grid area can be at least one of temperature values, depth values, and distance from the discharge outlet; that is, the influencing factors of the pollutant concentration in the grid area can be temperature, depth, and distance from the discharge outlet. For example, assuming the minimum resolution of the pollutant concentration impact data is... The size of the grid area can be , The value is 1 / 3 to 1 / 2. For example, data on the impact of pollutant concentration can be derived from historical literature. Taking data on the impact of pollutant concentration on lakes as an example, the temperature values ​​of each grid area can be obtained from large-scale, long-term surface water temperature grid data provided in literature on lake remote sensing research; the depth values ​​of each grid area can be obtained from grid data of average water depth or historical topographic data of the grid area provided in literature on lake remote sensing research; and the distance between the grid area and the discharge outlet can be obtained based on the location of the discharge outlet. For example, data on the impact of pollutant concentration can be derived from the applicant's historical research data.

[0042] As one implementation method, the similarity of pollutant impact between two adjacent grid regions can be obtained as follows: First, obtain the grid regions separately. Pollutant concentration impact data and grid area The pollutant concentration impact data includes the values ​​of multiple pollutant concentration impact factors. Then, a pollutant impact feature vector for a grid region is constructed based on the values ​​of these multiple pollutant concentration impact factors. Pollutant impact feature vector , The number of impact factors, For the first The values ​​of each influencing factor; grid area Pollutant impact feature vector , The number of impact factors, For the first The numerical values ​​of each influencing factor are then determined. Next, the magnitudes of the two pollutant influence feature vectors are obtained, along with the Euclidean distance between them. Finally, based on the Euclidean distance and the magnitudes of the two pollutant influence feature vectors, the similarity of pollutant influence between the two grid regions is calculated using the following formula:

[0043]

[0044] in, For grid area and grid area The similarity is affected by the pollutants. The number of impact factors, For grid area The pollutant impact feature vector of the th The values ​​of each influence factor. For grid area The pollutant impact feature vector of the th The values ​​of each influencing factor; finally, when two grid regions and Pollutant impact similarity Greater than or equal to the preset similarity threshold At that time, determine the grid area and grid area For similar grid regions, one can be deleted while keeping the other, with a preset similarity threshold. It can be set according to actual needs or experience.

[0045] During the screening of target grid areas, all critical target grid areas must be retained. Critical target grid areas are the key areas where pollutant concentration data must be actually collected and measured. In other words, the target grid area includes multiple critical target grid areas and multiple non-critical target grid areas. Pollutant concentration data in non-critical target grid areas may not necessarily need to be actually collected and measured.

[0046] Step S12: Collect spectral data of the target grid area to obtain pollutant concentration data of the key target grid area.

[0047] Mobile platforms can be used for spectral data acquisition, and monitoring vessels can be used for sample collection (on-site or laboratory determination of pollutant concentration data). For example, mobile platforms include drones equipped with spectral sensors or monitoring vessels equipped with spectral sensors; specifically, drones are selected as multi-rotor models with a payload capacity greater than or equal to 2 kg and an endurance greater than or equal to 30 min, or monitoring vessels are selected as small monitoring vessels with a speed of 3–6 km / h and an endurance less than or equal to 4 h; the spectral sensor is selected as a point-type fiber optic spectrometer with a wavelength range of 400–950 nm, a spectral resolution less than or equal to 3 nm, and a sampling frequency of 1–5 Hz. During installation and fixing, the spectral sensor is fixed to the underside of the drone's fuselage using a shock-absorbing bracket, at least 5 m above the ground; or the spectral sensor on the monitoring vessel is installed in the middle of the hull, at least 0.5 m above the ground; ensure that the spectral sensor probe is perpendicular to the ground surface to reduce interference from platform vibration on sampling.

[0048] For example, radiometric calibration is required when collecting spectral data. A standard white board (with known reflectivity, such as a 99% diffuse reflectance white board) is used for radiometric calibration at the monitoring site. Based on the spectral sensor's response value to the standard white board, a conversion model of "sensor response value (V) - actual radiance (L)" is established: ,in This is the sensitivity coefficient. This is the dark current correction value.

[0049] For example, attitude calibration is required when acquiring spectral data. Attitude parameters (pitch angle, roll angle, measurement accuracy ≤ 0.1°) are acquired in real time using the IMU (Inertial Measurement Unit) module mounted on the mobile platform to establish an attitude deviation correction model. ,in This is the sensitivity coefficient. This is the dark current correction value.

[0050] For example, during the sampling process, a mobile platform and a spectral sensor are activated, traveling or flying along a predetermined path. The spectral sensor collects point-based spectral data at a set frequency, and simultaneously records the following types of data at millisecond-level synchronization based on GPS timestamps:

[0051] Spectral data: Spectral values ​​are continuously collected multiple times (e.g., 3 to 5 times) at each sampling point (target grid area), and the arithmetic mean of the multiple samples is taken as the spectral data of the target grid area. , , The number of target grid regions;

[0052] Pollutant concentration data: At each sampling point (key target grid area), samples are collected using a sampler, and experimental data are obtained through testing and analysis of the samples. Pollutant concentration data is then derived from this experimental data. , , The number of grid regions for key targets;

[0053] Auxiliary data: GPS coordinates of each sampling point (target grid area), environmental parameters, attitude parameters of the mobile platform, etc.

[0054] The target grid area contains spectral data. Specifically, the critical target grid area within the target grid area contains both spectral data and pollutant concentration data, while the non-critical target grid area only contains spectral data and no pollutant concentration data. Other grid areas within the multiple grid areas besides the target grid area can be called blank grid areas, which contain neither spectral data nor pollutant concentration data. In other words, the grid areas are divided into three categories: critical target grid areas, non-critical target grid areas, and blank grid areas.

[0055] Spectral values ​​of multiple pollutant concentration parameters can be extracted from the spectral data of key target grid areas, with each pollutant concentration parameter corresponding to a characteristic sensitive band. Taking a shallow lake as an example, the pollutant concentration parameters can be chlorophyll a, chemical oxygen demand (COD), and turbidity. The spectral values ​​of chlorophyll a in the sensitive band of 680–690 nm, COD in the sensitive band of 420–440 nm, and turbidity in the sensitive band of 550–570 nm can be extracted from the spectral data, respectively.

[0056] It should be noted that the spectral data of the key target grid area and the spectral data of the non-key target grid area collected in this step can also be referred to as the raw spectral data in the following steps.

[0057] Step S13: Obtain the first similarity index between the blank grid area and the target grid area based on the similarity of pollutant influence and spatial similarity between the blank grid area and the target grid area. Obtain the reconstructed spectral data of the blank grid area based on the spectral data of the target grid area and the first similarity index between the blank grid area and the target grid area. The blank grid area refers to the other grid areas in the multiple grid areas besides the target grid area.

[0058] Blank grid areas are the grid areas other than the target grid area among multiple grid areas. Blank grid areas do not have spectral data, and the spectral data of blank grid areas needs to be reconstructed.

[0059] For each blank grid region, spectral data is reconstructed using spectral data from a portion of the target grid regions or spectral data from all target grid regions. The interpolation weights of the spectral data from different target grid regions in the reconstructed spectral data of the blank grid region are different. These interpolation weights are calculated based on the first similarity index between the corresponding blank grid region and each target grid region.

[0060] The first similarity index between two grid regions is calculated based on the spatial similarity and the pollutant impact similarity between the two grid regions. The calculation method for the pollutant impact similarity can be found in the section on calculating the pollutant impact similarity between two adjacent grid regions described above.

[0061] First, obtain the blank grid areas respectively. Pollutant concentration impact data and target grid area The pollutant concentration impact data includes the values ​​of multiple pollutant concentration impact factors. Then, a pollutant impact feature vector is constructed for the grid area based on the values ​​of these multiple pollutant concentration impact factors. The blank grid area... Pollutant impact feature vector , The number of impact factors, The first blank area The values ​​of each influencing factor; target grid area Pollutant impact feature vector , The number of impact factors, For target grid area The The numerical values ​​of each influencing factor are then determined. Next, the magnitudes of the two pollutant influence feature vectors are obtained, along with the Euclidean distance between them. Finally, based on the Euclidean distance and the magnitudes of the two pollutant influence feature vectors, the similarity of pollutant influence between the two grid regions is calculated using the following formula:

[0062]

[0063] in, blank grid area and target grid area The similarity is affected by the pollutants. The number of impact factors, blank grid area The pollutant impact feature vector of the th The values ​​of each influence factor. For target grid area The pollutant impact feature vector of the th The numerical values ​​of each influencing factor.

[0064] Step S14: Resample the key target grid region based on the reconstructed spectral data of the grid region to obtain the resampled spectral data of the key target grid region.

[0065] The posture of the spectral data sampling device and changes in ambient lighting can interfere with the acquisition of spectral data. For each key target grid region, the spectral data of the key target grid region can be resampled based on the reconstructed spectral data. The resampled spectral data takes into account a richer surrounding information field, so it can be used to optimize and correct the original spectral data of the key target grid region, making the results more reliable and smooth, and eliminating possible boundary effects or errors.

[0066] Similarly, for each non-critical target grid region, the spectrum of the non-critical target grid region can be resampled based on the reconstructed spectral data. The richer surrounding information field reconstructed can be used to optimize and correct the original spectral data of the non-critical target grid region, making the results more reliable and smooth, and eliminating possible boundary effects or errors.

[0067] Step S15: Establish a first fitting model based on the spectral data and pollutant concentration data of the key target grid area, establish a second fitting model based on the resampled spectral data and pollutant concentration data of the key target grid area, and obtain a fusion fitting model based on the first fitting model and the second fitting model.

[0068] The spectral data and pollutant concentration data of the key target grid area were fitted to obtain the first fitting model. , For pollutant concentration data, For the raw spectral data, The functional relationship between pollutant concentration data and raw spectral data in the key target grid area is given by the first fitting model, which has a goodness of fit of [value missing]. For example, a linear regression model or a nonlinear regression model can be used to fit the spectral data and pollutant concentration data of the key target grid area to obtain the first fitting model. Among them, for nonlinear fitting, classic machine learning models such as random forests are recommended.

[0069] The second fitting model is obtained by fitting the resampled spectral data and pollutant concentration data of the key target grid area. , For pollutant concentration data, To resample spectral data, The functional relationship between pollutant concentration data and resampled spectral data in the key target grid area is given by the goodness of fit of the second fitting model. For example, a linear regression model or a nonlinear regression model can be used to fit the resampled spectral data and pollutant concentration data of the key target grid area to obtain a second fitting model. Similarly, for nonlinear fitting, classic machine learning models such as random forests are recommended.

[0070] like Figure 3 As shown, the fusion fitting model , For pollutant concentration data, This refers to reconstructed spectral data for blank areas, or spectral data or resampled spectral data for non-critical target grid areas. For fusion weighting coefficients, fusion weighting coefficients It is calculated based on the goodness of fit of the first fitting model and the goodness of fit of the second fitting model. .

[0071] Step S16: Obtain pollutant concentration data for other grid areas in the grid region, excluding the key target grid area, based on the fusion fitting model.

[0072] The reconstructed spectral data of the blank area is used as Substitute into the fusion fitting model In the calculation of pollutant concentration data, spectral data or resampled spectral data from non-critical target grid areas are used as the basis for calculation. Substitute into the fusion fitting model The calculated values ​​of pollutant concentration data are used to calculate the values.

[0073] The fusion fitting model was validated using pollutant concentration data from 10%–15% of the key target grid regions. Specifically, resampled spectral data from the key target grid regions were used as... Substitute into the fusion fitting model The calculated pollutant concentration data is then compared with the corresponding pollutant concentration data, and the error between the two is less than 5%.

[0074] In this embodiment, multiple target grid regions are selected by screening the similarity of pollutant impact between every two adjacent grid regions. This avoids sampling redundancy and omission of spectral data, which helps improve the accuracy of fitting pollutant concentrations using the sampled spectra. A first similarity index is obtained between the pollutant impact similarity between the blank grid region and the target grid region. The spectral data of the blank region is then reconstructed based on this first similarity index, improving the accuracy of the reconstructed spectral data and thus enhancing the accuracy of fitting pollutant concentrations using the sampled spectra. The key target grid regions are resampled using the reconstructed spectral data from the blank grid regions. A fusion fitting model is obtained through direct inversion of spectral data and pollutant concentration data from the key target grid regions, as well as through indirect inversion of resampled spectral data and pollutant concentration data from the key target grid regions. This reduces the interference of the spectral data sampling device's posture and ambient light changes on the spectral data, further improving the accuracy of fitting pollutant concentrations using the sampled spectra.

[0075] As one implementation method, after step S16, the following steps are also included: generating a thematic map for each pollutant indicator based on the pollutant concentration data of the grid area. Specifically, low concentration range, medium concentration range, and high concentration range can be set for each pollutant indicator (e.g., chlorophyll a concentration, chemical oxygen demand, or turbidity), with different colors corresponding to the low, medium, and high concentration ranges, and different colors are matched to the concentration range corresponding to the pollutant concentration data of the grid area. For example, as... Figure 4 As shown, a tiered color-coding method can be used to generate thematic maps of pollutant concentrations. Low-concentration (compliant) areas are marked in blue, medium-concentration areas in yellow, and high-concentration (exceeding) areas in red. The map update frequency is consistent with the sampling frequency of the spectral sensor, achieving real-time dynamic updates. The red area represents the maximum exceeding value; the lower limit threshold of the high-concentration range can be obtained through calculation. = , = ,in Referring to the Class III standard in the "Surface Water Environmental Quality Standard" (GB3838-2002), For the pollutant concentration data of the first grid area, This represents the number of grid regions.

[0076] As one implementation method, in step S13, the reconstructed spectral data of the blank grid region is obtained based on the spectral data of the target grid region and the first similarity index between the blank grid region and the target grid region. This specifically includes the following steps:

[0077] Step S21: For each blank grid region, normalize the first similarity index between the blank grid region and the target grid region to obtain the first similarity weight between the blank grid region and each target grid region.

[0078] Specifically, the first similarity weight between the blank grid region and each target grid region can be calculated as follows:

[0079]

[0080] in, For blank grid areas and the first The first similarity weight of each target grid region For blank grid areas and the first The first similarity index of each target grid region For blank grid areas and the first The first similarity index of each target grid region The number of target grid regions.

[0081] Step S22: For each blank grid region, establish an improved Kriging equation system based on the first similarity weight:

[0082]

[0083] in, For the first Interpolation weights for each target grid region The number of target grid regions, For blank grid areas and the first The first similarity weight of each of the target grid regions The blank grid area and the first The distance between the target grid regions Distance The corresponding semi-mutation function value, For the first The target grid region and the first The first similarity weight of each of the target grid regions For the first The target grid region and the first The distance between the target grid regions Distance The corresponding semi-mutation function value, It is a Lagrange multiplier.

[0084] Specifically, in practice, even if two grid regions are not spatially close, if they are under similar environmental conditions (e.g., downwind of the same industrial area, with similar water temperatures or depths), their contamination processes will be similar, and their pollutant concentrations (or associated spectral characteristics) should be closer. Therefore, a first similarity weight is introduced into the improved Kriging equations. This first similarity weight is calculated based on spatial similarity and the similarity of pollutant influence. The calculated interpolation weight is more accurate and can improve the accuracy of reconstructed spectral data in blank areas.

[0085] Specifically, ,in, The distance is The number of target grid region pairs, with a distance of The target grid region consists of two points at a distance of... The target grid region is composed of, and The distances are respectively Spectral data of two target grid regions.

[0086] Specifically, ,in, The distance is The number of target grid region pairs, with a distance of The target grid region consists of two points at a distance of... The target grid region is composed of, and The distances are respectively Spectral data of two target grid regions.

[0087] Step S23: Solve for the interpolation weights of each target grid region according to the improved Kriging equations, and calculate the reconstructed spectral data of the blank grid region according to the Kriging interpolation algorithm. The Kriging interpolation algorithm is as follows:

[0088]

[0089] in, Reconstructing spectral data for blank grid regions, For the first Interpolation weights for each target grid region For the first Spectral data for each target grid region The number of target grid regions.

[0090] In this embodiment, an improved Kriging equation is constructed by obtaining the first similarity weight between the pollutant influence similarity between the blank grid region and the target grid region. This makes the Kriging interpolation algorithm more accurate, improves the accuracy of the reconstructed spectral data in the blank region, and helps to improve the accuracy of fitting pollutant concentrations to the sampled spectrum.

[0091] Those skilled in the art should understand that, in addition to using the first similarity weight to construct the improved Kriging interpolation algorithm, other types of interpolation algorithms can also be constructed using the first similarity weight. For example, improved inverse distance weight interpolation algorithms, improved polynomial interpolation algorithms, improved natural neighborhood methods, improved spline function methods, and improved trend surface methods can be constructed using the first similarity weight.

[0092] As one implementation method, the steps for calculating the similarity of pollutant impact between any two grid regions include:

[0093] Step S31: Obtain pollutant concentration impact data for each of the grid regions, wherein the pollutant concentration impact data includes the values ​​of multiple pollutant concentration impact factors.

[0094] Step S32: Construct the pollutant influence feature vector of the grid region based on the values ​​of the corresponding multiple pollutant concentration influence factors.

[0095] Step S33: Obtain the magnitude of the two pollutant influence feature vectors respectively, and obtain the Euclidean distance between the two pollutant influence feature vectors.

[0096] Step S34: Obtain the pollutant influence similarity between the two grid regions based on the Euclidean distance and the magnitudes of the two pollutant influence feature vectors.

[0097] Specifically, the formula for calculating the similarity of pollutant impact between any two grid regions is as follows:

[0098]

[0099] in, For grid area and grid area The similarity is affected by the pollutants. The number of impact factors, For grid area The pollutant impact feature vector of the th The values ​​of each influence factor. For grid area The pollutant impact feature vector of the th The numerical values ​​of each influencing factor.

[0100] As one implementation method, the steps for calculating the spatial similarity between any two grid regions include:

[0101] Step S41: Calculate the spatial similarity between two grid regions based on the distance decay function, which is:

[0102]

[0103] in, For the first The grid region and the first The distance between grid regions The attenuation coefficient is... It is a natural exponential function. For the first The grid region and the first Spatial similarity of grid regions.

[0104] For example, the distance between two grid regions can be the Euclidean distance between the center points of the two grid regions.

[0105] As one implementation method, the calculation steps for the first similarity index between any two grid regions include:

[0106]

[0107] in, For the first The grid region and the first The first similarity index of each grid region For grid area and grid area The similarity is affected by the pollutants. For the first The grid region and the first Spatial similarity of grid regions These are the weighting coefficients. .

[0108] In one implementation method, step S14 specifically includes the following steps:

[0109] Step S51: Obtain the first similarity index between the key target grid region and the blank grid region based on the similarity of pollutant impact and spatial similarity between the key target grid region and the blank grid region.

[0110] Step S52: Based on the reconstructed spectral data of the blank grid region and the first similarity index between the key target grid region and the blank grid region, obtain the resampled spectral data of the key target grid region.

[0111] In this embodiment, for each key target grid region, the reconstructed spectral data of all blank grid regions is used to resample the spectral data. The reconstructed spectral data of different blank grid regions have different interpolation weights in the resampled spectral data of the key target grid region. The interpolation weights are calculated based on the first similarity index between the corresponding key target grid region and each blank grid region.

[0112] The first similarity index between two grid regions is calculated based on the spatial similarity between the two grid regions and the pollutant impact similarity between the two grid regions. The calculation method for the pollutant impact similarity can be found in steps S31 to S34 above, and the calculation method for the spatial similarity can be found in step S41 above.

[0113] In some implementations, the improved Kriging interpolation algorithm, inverse distance weighted interpolation, radial basis function interpolation, etc., described above can be used to resample the spectral data of the key target grid region. Step S52 specifically includes the following steps:

[0114] Step S61: For each key target grid region, normalize the first similarity index between the key target grid region and the blank grid region to obtain the first similarity weight between the key target grid region and each blank grid region.

[0115] Specifically, the first similarity weight between the key target grid region and each blank grid region can be calculated as follows:

[0116]

[0117] in, For the key target grid area and the first The first similarity weight of each blank grid region For the key target grid area and the first The first similarity index of the blank grid regions, For the key target grid area and the first The first similarity index of the blank grid regions, This represents the number of blank grid areas.

[0118] Step S62: For each key target grid region, establish an improved Kriging equation system based on the first similarity weight:

[0119]

[0120] in, For the first Interpolation weights for each blank grid region The number of blank grid areas. For the key target grid region and the first The first similarity weight of the blank grid regions For the key target grid region and the first The distance between the blank grid regions, Distance The corresponding semi-mutation function value, For the first The blank grid region mentioned above and the first The first similarity weight of the blank grid regions For the first The blank grid region mentioned above and the first The distance between the blank grid regions, Distance The corresponding semi-mutation function value, It is a Lagrange multiplier.

[0121] Specifically, ,in, The distance is The number of blank grid region pairs, with a distance of The blank grid region is composed of two distances. Composed of blank grid areas, and The distances are respectively Spectral data of two blank grid regions.

[0122] Specifically, ,in, The distance is The number of blank grid region pairs, with a distance of The blank grid region is composed of two distances. Composed of blank grid areas, and The distances are respectively Spectral data of two blank grid regions.

[0123] Step S63: Solve for the interpolation weights of each blank grid region according to the improved Kriging equations, and calculate the resampled spectral data of the key target grid region according to the Kriging interpolation algorithm. The Kriging interpolation algorithm is as follows:

[0124]

[0125] in, For the resampled spectral data of the key target grid region, For the first Interpolation weights for the blank grid regions For the first Reconstructed spectral data for each of the blank grid regions, This represents the number of blank grid areas.

[0126] Those skilled in the art should understand that, in addition to using the first similarity weight to construct the improved Kriging interpolation algorithm, other types of interpolation algorithms can also be constructed using the first similarity weight. For example, improved inverse distance weight interpolation algorithms, improved polynomial interpolation algorithms, improved natural neighborhood methods, improved spline function methods, and improved trend surface methods can be constructed using the first similarity weight.

[0127] Alternatively, the non-critical target grid region can be resampled based on the reconstructed spectral data of each blank grid region, or the method in steps S61 to S63 can be used. See the above for details, which will not be elaborated here.

[0128] In one implementation, after step S11 and before step S12, the following steps are also included:

[0129] Step S71: Select the starting target grid region and the ending target grid region from multiple target grid regions respectively.

[0130] The starting target grid area is: The target grid area is The starting target grid area can be determined based on the sampling starting point location, and a target grid area that is closer to the sampling starting point location can be selected; the ending target grid area can be determined based on the sampling ending point location, and a target grid area that is closer to the sampling ending point can be selected.

[0131] Step S72: Based on the starting target grid region and the ending target grid region, traverse all target grid regions and obtain multiple sampling paths respectively.

[0132] Among them, sampling path The number of target grid regions is , They are respectively except and outside Given a target grid region, iterate through the arrangement of all target grid regions to obtain multiple sampling paths.

[0133] Step S73: Traverse all sampling paths and calculate the total distance for each sampling path. The total distance is the sum of the distances between any two adjacent target grid regions in the sampling path.

[0134] Among them, sampling path Its total distance for .

[0135] Step S74: Based on the total distance of the sampling paths, obtain a target sampling path from multiple sampling paths.

[0136] Among these, the path with the smallest total distance can be selected as the target sampling path. For example, the target sampling path can be as follows: Figure 2 As shown.

[0137] Accordingly, in step S12, spectral data of each target grid region is collected sequentially according to the target sampling path.

[0138] One embodiment of this application provides a spectroscopic pollutant concentration fitting device. Please refer to [link to relevant documentation]. Figure 5 As shown, the spectroscopic pollutant concentration fitting device 200 includes: a grid division module 21, a data acquisition module 22, a reconstruction module 23, a resampling module 24, a fitting module 25, and an inversion module 26.

[0139] The grid division module 21 is used to divide the target monitoring area into multiple grid regions, and to select multiple target grid regions from the multiple grid regions based on the similarity of pollutant impact between each pair of adjacent grid regions, wherein at least some of the multiple target grid regions are key target grid regions; the data acquisition module 22 is used to collect spectral data of the target grid regions and acquire pollutant concentration data of the key target grid regions; the reconstruction module 23 is used to obtain a first similarity index between the blank grid region and the target grid region based on the similarity of pollutant impact and spatial similarity between the blank grid region and the target grid region, and to obtain the reconstruction index of the blank grid region based on the spectral data of the target grid region and the first similarity index between the blank grid region and the target grid region. The system comprises: a reconstructed spectral data set, wherein the blank grid region refers to other grid regions in the plurality of grid regions besides the target grid region; a resampling module 24, used to resample the key target grid region based on the reconstructed spectral data of the blank grid region to obtain resampled spectral data of the key target grid region; a fitting module 25, used to establish a first fitting model based on the spectral data and pollutant concentration data of the key target grid region, establish a second fitting model based on the resampled spectral data and pollutant concentration data of the key target grid region, and obtain a fusion fitting model based on the first fitting model and the second fitting model; and an inversion module 26, used to obtain pollutant concentration data of other grid regions in the grid regions besides the key target grid region based on the fusion fitting model.

[0140] Figure 6This is a schematic diagram of the structure of an unmanned intelligent device according to an embodiment of this application. Figure 5 As shown, the unmanned intelligent device 60 includes a processor 61 and a memory 62 coupled to the processor 61.

[0141] The memory 62 stores program instructions for implementing the spectrum-based pollutant concentration fitting method of any of the above embodiments.

[0142] The processor 61 is used to execute program instructions stored in the memory 62 to train a singing voice conversion or timbre conversion model.

[0143] The processor 61 can also be referred to as a CPU (Central Processing Unit). The processor 61 may be an integrated circuit chip with signal processing capabilities. The processor 61 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor.

[0144] For example, the unmanned intelligent device 60 can be a drone or an unmanned monitoring vessel.

[0145] See Figure 7 , Figure 7 This is a schematic diagram of the structure of a computer-readable storage medium according to an embodiment of this application. The computer-readable storage medium 70 of this embodiment stores program instructions 71 capable of implementing all the methods described above. These program instructions 71 can be stored in the storage medium in the form of a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or terminal devices such as computers, servers, mobile phones, and tablets.

[0146] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0147] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0148] The above description is merely an embodiment of this application. It should be noted that those skilled in the art can make improvements without departing from the inventive concept of this application, but these improvements all fall within the protection scope of this application.

Claims

1. A method for fitting pollutant concentrations based on spectra, characterized in that, include: The target monitoring area is divided into multiple grid areas. Based on the similarity of pollutant impact between any two adjacent grid areas, multiple target grid areas are selected from the multiple grid areas. At least some of the multiple target grid areas are key target grid areas. Spectral data of the target grid region is collected to obtain pollutant concentration data of the key target grid region; the target grid region has spectral data, and the key target region has both spectral data and pollutant concentration data. The first similarity index between the blank grid region and the target grid region is obtained based on the similarity of pollutant impact and spatial similarity between the blank grid region and the target grid region. The reconstructed spectral data of the blank grid region is obtained based on the spectral data of the target grid region and the first similarity index between the blank grid region and the target grid region. The blank grid region is any other grid region among the plurality of grid regions except the target grid region. Based on the reconstructed spectral data of the blank grid region, the key target grid region is resampled to obtain the resampled spectral data of the key target grid region; A first fitting model is established based on the spectral data and pollutant concentration data of the key target grid region; a second fitting model is established based on the resampled spectral data and pollutant concentration data of the key target grid region; and a fusion fitting model is obtained based on the first fitting model and the second fitting model. Based on the fusion fitting model, obtain pollutant concentration data for other grid areas in the grid region, excluding the key target grid area; Specifically, based on the spectral data of the target grid region and a first similarity index between the blank grid region and the target grid region, the reconstructed spectral data of the blank grid region is obtained, including: For each blank grid region, the first similarity index between the blank grid region and the target grid region is normalized to obtain the first similarity weight between the blank grid region and each target grid region. For each of the blank grid regions, an improved Kriging equation system is established based on the first similarity weight: in, For the first Interpolation weights for the target grid region The number of target grid regions, The blank grid area and the first The first similarity weight of each of the target grid regions The blank grid area and the first The distance between the target grid regions Distance The corresponding semi-mutation function value, For the first The target grid region and the first The first similarity weight of each of the target grid regions For the first The target grid region and the first The distance between the target grid regions Distance The corresponding semi-mutation function value, It is a Lagrange multiplier. ; The interpolation weights for each target grid region are solved using the improved Kriging equations. The reconstructed spectral data for the blank grid region is then calculated using the Kriging interpolation algorithm, which is as follows: in, The reconstructed spectral data for the blank grid region, For the first Interpolation weights for the target grid region For the first Spectral data of the target grid region, The number of target grid regions; The steps for calculating the similarity of pollutant impact between any two grid regions include: The pollutant concentration impact data for each of the grid regions is obtained, wherein the pollutant concentration impact data includes the values ​​of multiple pollutant concentration impact factors; Construct a pollutant influence feature vector for the grid region based on the values ​​of multiple corresponding pollutant concentration influence factors; Obtain the magnitude of each of the two pollutant impact feature vectors, and obtain the Euclidean distance between the two pollutant impact feature vectors; Based on the Euclidean distance and the magnitudes of the two pollutant influence feature vectors, the similarity of pollutant influence between the two grid regions is obtained. The steps for calculating the spatial similarity between any two grid regions include: The spatial similarity between the two grid regions is calculated based on a distance decay function, which is: in, For the first The grid region and the first The distance between grid regions The attenuation coefficient is... It is a natural exponential function. For the first The grid region and the first Spatial similarity of grid regions.

2. The method for fitting pollutant concentration based on spectrum according to claim 1, characterized in that, The step of resampling the key target grid region based on the reconstructed spectral data of the blank grid region to obtain the resampled spectral data of the key target grid region includes: The first similarity index between the key target grid region and the blank grid region is obtained based on the similarity of pollutant impact and spatial similarity between the key target grid region and the blank grid region; Based on the reconstructed spectral data of the blank grid region and the first similarity index between the key target grid region and the blank grid region, the resampled spectral data of the key target grid region is obtained.

3. The method for fitting pollutant concentration based on spectrum according to claim 2, characterized in that, Based on the reconstructed spectral data of the blank grid region and the first similarity index between the key target grid region and the blank grid region, the resampled spectral data of the key target grid region is obtained, including: For each of the key target grid regions, the first similarity index between the key target grid region and the blank grid region is normalized to obtain the first similarity weight between the key target grid region and each of the blank grid regions. For each of the key target grid regions, an improved Kriging equation system is established based on the first similarity weight: in, For the first Interpolation weights for the blank grid regions The number of blank grid areas. For the key target grid region and the first The first similarity weight of the blank grid regions For the key target grid region and the first The distance between the blank grid regions, Distance The corresponding semi-mutation function value, For the first The blank grid region mentioned above and the first The first similarity weight of the blank grid regions For the first The blank grid region mentioned above and the first The distance between the blank grid regions, Distance The corresponding semi-mutation function value, It is a Lagrange multiplier. ; The interpolation weights for each blank grid region are solved using the improved Kriging equations. The resampled spectral data for the key target grid region is then calculated using the Kriging interpolation algorithm, which is as follows: in, For the resampled spectral data of the key target grid region, For the first Interpolation weights for the blank grid regions For the first Reconstructed spectral data for each of the blank grid regions, This represents the number of blank grid areas.

4. The method for fitting pollutant concentration based on spectrum according to claim 1, characterized in that, Before acquiring the spectral data of the target grid region and obtaining the pollutant concentration data of the key target grid region, the method further includes: Select the starting target grid region and the ending target grid region from the plurality of target grid regions respectively; Based on the starting target grid region and the ending target grid region, traverse all the target grid regions and obtain multiple sampling paths respectively; Traverse all sampling paths and calculate the total distance for each sampling path. The total distance is the sum of the distances between any two adjacent target grid regions in the sampling path. Based on the total distance of the sampling paths, a target sampling path is obtained from the multiple sampling paths; Accordingly, the acquisition of spectral data of the target grid region includes: Spectral data for each target grid region are collected sequentially according to the target sampling path.

5. An unmanned intelligent device, characterized in that, The system includes a processor and a memory coupled to the processor, the memory storing program instructions executable by the processor; when the processor executes the program instructions stored in the memory, it implements the spectral-based pollutant concentration fitting method as described in any one of claims 1 to 4.

6. A spectroscopic pollutant concentration fitting device, used to implement the spectroscopic pollutant concentration fitting method as described in any one of claims 1 to 4, characterized in that, include: A grid division module is used to divide the target monitoring area into multiple grid regions, and to select multiple target grid regions from the multiple grid regions based on the similarity of pollutant impact between each pair of adjacent grid regions, wherein at least some of the multiple target grid regions are key target grid regions; The data acquisition module is used to collect spectral data of the target grid area and obtain pollutant concentration data of the key target grid area; The reconstruction module is used to obtain a first similarity index between the blank grid area and the target grid area based on the similarity of pollutant influence and spatial similarity between the blank grid area and the target grid area, and to obtain reconstructed spectral data of the blank grid area based on the spectral data of the target grid area and the first similarity index between the blank grid area and the target grid area, wherein the blank grid area is other grid areas besides the target grid area among the plurality of grid areas; The resampling module is used to resample the key target grid region based on the reconstructed spectral data of the blank grid region, and obtain the resampled spectral data of the key target grid region; The fitting module is used to establish a first fitting model based on the spectral data and pollutant concentration data of the key target grid area, establish a second fitting model based on the resampled spectral data and pollutant concentration data of the key target grid area, and obtain a fusion fitting model based on the first fitting model and the second fitting model. The inversion module is used to obtain pollutant concentration data of other grid areas in the grid region, excluding the key target grid area, based on the fusion fitting model.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions that, when executed by a processor, implement the spectral-based pollutant concentration fitting method as described in any one of claims 1 to 4.

Images