A water quality parameter classification inversion method and system based on multidimensional hyperspectral images

Through the cyclic generation of adversarial networks and k-mean clustering method, the inconsistency of input dimensions caused by water diversity in hyperspectral imaging technology is solved, and the accurate inversion of water quality parameters is achieved, ensuring the input dimension of neural networks in the training and application stages is achieved. consistency.

CN118655099BActive Publication Date: 2024-11-05陕西省环境监测中心站
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411133538.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-19
Publication Date
2024-11-05
Estimated Expiration
2044-08-19

AI Technical Summary

Technical Problem

In water quality monitoring, hyperspectral imaging technology leads to inconsistent information loss and changes in the dimensionality reduction process due to the diversity of water bodies, resulting in inconsistent input dimensions of neural networks during training and application stages, affecting the accuracy of water quality parameters.

Method used

Using the cyclic generation adversarial network and k-mean clustering method, the hyperspectral images are segmented into different classes, and the polymerized spectral vector is obtained by polymerizing the spectral values, and input it into the inversion model to obtain water quality parameters, ensuring the input in the training and application stages. The dimensions are consistent.

Benefits of technology

It effectively reduces spectral feature redundancy, ensures the accuracy of deep learning networks in the application stage, and improves the accuracy and robustness of water quality parameter inversion.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118655099B_ABST
    Figure CN118655099B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for water quality parameter classification inversion based on multidimensional hyperspectral images, and relates to the technical field of water quality monitoring. It includes: constructing a data set based on spectral vectors corresponding to water quality parameters; using the data set to train a water quality parameter inversion model; dividing the pixel points in the hyperspectral image of the water body to be detected into different classes; aggregating the spectral values ​​of the same type of pixels and inputting them into the inversion model to obtain water quality parameters. In order to solve the problem that the input dimensions of the inversion model in the training stage and the application stage are consistent due to dimensionality reduction, thereby causing inaccurate water quality parameters. The present invention aggregates water bodies with similar water quality characteristics through clustering; aggregates the spectral values ​​of similar water quality characteristics, reduces the redundancy of spectral features of the same type of water bodies, and does not change the number or order of spectral bands of each pixel point, thereby ensuring the consistency of the input dimensions of the inversion model in the training stage and the application stage, thereby ensuring the accuracy of the inversion of water quality parameters.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of water quality monitoring, and in particular to a water quality parameter classification inversion method and system based on multi-dimensional hyperspectral images. Background Art

[0002] In the field of water quality monitoring, hyperspectral imaging technology, with its wide spectral coverage, can capture the characteristic absorption, reflection or fluorescence of different chemical components in water bodies in different spectral bands, thereby realizing the simultaneous monitoring of various types of water quality parameters. By analyzing the collected spectral characteristics of different types and concentrations of substances contained in the water body, an inversion model of water quality parameters can be constructed to realize the evaluation of various water quality indicators such as suspended sediment content and organic pollution.

[0003] At present, when monitoring water quality based on hyperspectral imaging technology, first, use hyperspectral cameras and other equipment to scan the target water area to obtain the reflectance spectral signals of the water body in different spectral bands, and at the same time obtain key water quality parameters such as total nitrogen, total phosphorus and suspended solids concentration in the water body; then, use the collected data to train the neural network to establish a mapping relationship between the spectral data and the concentration of water quality parameters; finally, input the hyperspectral image of the water body to be monitored into the trained neural network to obtain the corresponding water body parameters. Due to the high correlation in the hyperspectral image of the water body, this correlation will cause redundancy of spectral features, so dimensionality reduction processing is often required before the hyperspectral image is input into the neural network.

[0004] However, due to the diversity of water quality in nature, the collected water hyperspectral images have different redundant characteristics, which leads to different information loss and changes in each water hyperspectral image during the dimensionality reduction process, making it impossible to ensure that the input dimensions of the neural network are consistent during the training and application stages. The performance of the neural network is often highly dependent on the data used during its training. If there is a difference between the input dimension of the neural network in the training stage and the input dimension in the application stage, this may cause the neural network to be unable to accurately process the input water hyperspectral image, resulting in inaccurate water quality parameters. Summary of the invention

[0005] Based on this, in order to solve the above technical problems, the present invention provides a water quality parameter classification inversion method and system based on multi-dimensional hyperspectral images.

[0006] The present invention provides a water quality parameter classification inversion method based on multidimensional hyperspectral images, comprising:

[0007] Obtain multiple groups of water quality index parameters corresponding to multiple different water samples, each group of water quality index parameters is used to describe the water quality characteristics of the corresponding water sample; collect each water sample in LSpectral values ​​of different spectral bands are obtained to obtain multiple groups of water quality index parameters corresponding to each other. L dimensional spectral vector, where L is a positive integer; based on water quality index parameters and L dimensional spectral vector to construct the dataset.

[0008] A deep learning network is constructed, and the deep learning network is trained using a data set to obtain an inversion model for inverting water quality parameters. The deep learning network adopts a cyclic generative adversarial network; during the training process of the cyclic generative adversarial network, the original L The dimensional spectral vector is converted into pseudo water quality parameters, and then the pseudo water quality parameters are converted back L dimensional spectral vector; the original L dimensional spectral vector and the pseudo water quality parameters converted back L The gap between dimensional spectral vectors enables recurrent generative adversarial network learning L The mapping relationship between dimensional spectral vector and water quality parameters.

[0009] Collect water to be tested L dimensional hyperspectral image and use k-means clustering method to L The pixels in the 2D hyperspectral image are divided into different classes. K-means clustering calculates the distance between the spectral values ​​of different pixels and clusters the pixels whose spectral value distance is less than the set threshold into one class. The water area corresponding to the same class of pixels is L The spectral values ​​of the same type of pixels belonging to the same spectral band are aggregated to obtain the corresponding L dimensional aggregated spectral vector; the corresponding L The dimensional aggregated spectral vectors are respectively input into the inversion model to obtain the water quality parameters of the water body to be detected; wherein the spectral values ​​of the same type of pixel points belonging to the same spectral band are aggregated including:

[0010] Get the spectral values ​​of similar pixels belonging to the same spectral band, and calculate the average spectral values ​​of similar pixels in the same spectral band:

[0011]

[0012] in, n k Indicates k The number of class pixels, Indicates i spectral bands, Indicates k The first pixel in the j pixels in the spectral band The spectral value of Indicates k Similar pixels in the spectral band The above formula is used to aggregate similar pixels through averaging operation; L The spectral bands are averaged to obtain the corresponding pixel points. L dimensional aggregate spectral vector:

[0013]

[0014] in, ~ Indicates k Class pixel points from spectral bands ~ Spectral bands The average value of the spectrum.

[0015] Aggregating the spectral values ​​of similar pixels belonging to the same spectral band also includes:

[0016] Get the spectral values ​​of similar pixels belonging to the same spectral band, extract the mode or median of the spectral values ​​of similar pixels in the same spectral band; L The mode or median extraction operation is performed on each spectral band to obtain the corresponding values ​​of various pixel points. L dimensional aggregated spectral vector.

[0017] The present invention provides a water quality parameter classification inversion system based on multidimensional hyperspectral images, comprising:

[0018] The data acquisition module is used to obtain multiple groups of water quality index parameters corresponding to multiple different water samples. Each group of water quality index parameters is used to describe the water quality characteristics of the corresponding water sample. L Spectral values ​​of different spectral bands are obtained to obtain multiple groups of water quality index parameters corresponding to each other. L dimensional spectral vector, where L is a positive integer; based on water quality index parameters and L dimensional spectral vector to construct the dataset.

[0019] The model building module is used to build a deep learning network for inverting water quality parameters, and use the data set to train the deep learning network to obtain a water quality parameter inversion model; the model building module uses a cyclic generative adversarial network, and the cyclic generative adversarial network training process converts the original L The dimensional spectral vector is converted into pseudo water quality parameters, and then the pseudo water quality parameters are converted back L dimensional spectral vector; the original L dimensional spectral vector and the pseudo water quality parameters converted back LThe gap between dimensional spectral vectors enables recurrent generative adversarial network learning L The mapping relationship between dimensional spectral vector and water quality parameters.

[0020] Water quality parameter inversion module, used to collect the water quality parameters of the water body to be tested. L dimensional hyperspectral image and use k-means clustering method to L The pixels in the 2D hyperspectral image are divided into different classes. K-means clustering calculates the distance between the spectral values ​​of different pixels and clusters the pixels whose spectral value distance is less than the set threshold into one class. The water area corresponding to the same class of pixels is L The spectral values ​​of the same type of pixels belonging to the same spectral band are aggregated to obtain the corresponding L dimensional aggregated spectral vector; the corresponding L The dimensional aggregated spectral vectors are respectively input into the inversion model to obtain the water quality parameters of the water body to be detected; wherein, the spectral values ​​of the same type of pixel points belonging to the same spectral band are aggregated including:

[0021] Get the spectral values ​​of similar pixels belonging to the same spectral band, and calculate the average spectral values ​​of similar pixels in the same spectral band:

[0022]

[0023] in, n k Indicates k The number of class pixels, Indicates i spectral bands, Indicates k The first pixel in the j pixels in the spectral band The spectral value of Indicates k Similar pixels in the spectral band The above formula is used to aggregate similar pixels through averaging operation; L The spectral bands are averaged to obtain the corresponding pixel points. L dimensional aggregate spectral vector:

[0024]

[0025] in, ~ Indicates k Class pixel points from spectral bands ~ Spectral bands The average value of the spectrum.

[0026] Aggregating the spectral values ​​of similar pixels belonging to the same spectral band also includes:

[0027] Get the spectral values ​​of similar pixels belonging to the same spectral band, extract the mode or median of the spectral values ​​of similar pixels in the same spectral band; L The mode or median extraction operation is performed on each spectral band to obtain the corresponding values ​​of various pixel points. L dimensional aggregated spectral vector.

[0028] At least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects:

[0029] The present invention divides the hyperspectral image of the water body to be detected into different classes so as to group together water areas with similar water quality characteristics; the spectral values ​​of similar pixel points belonging to the same spectral band are aggregated, and the aggregation result can reflect the concentrated characteristics of the pixel points of this category in the same spectral band. This aggregation method not only reduces the redundancy of spectral features of similar water bodies, but also does not change the number or order of spectral bands of each pixel point, thereby ensuring the consistency of the input dimensions of the deep learning network in the training stage and the application stage, thereby ensuring the accuracy of the deep learning network in inverting water quality parameters in the application stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0031] Figure 1 A schematic diagram of the process flow of the water quality parameter classification inversion method based on multidimensional hyperspectral images provided by the present invention;

[0032] Figure 2 A schematic diagram of a process for obtaining water quality parameters of a water body to be detected provided by the present invention;

[0033] Figure 3 The present invention provides L Schematic diagram of 3D hyperspectral image;

[0034] Figure 4 The flow chart of the k-means clustering algorithm provided by the present invention;

[0035] Figure 5 Schematic diagram of the framework of the water quality parameter classification inversion system based on multi-dimensional hyperspectral images provided by the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in combination with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] Hyperspectral images usually contain information from multiple bands, each of which corresponds to the reflection or emission characteristics of water bodies in different spectral bands. The biggest feature of hyperspectral images is their multi-band nature, which usually contains dozens or even hundreds of bands, covering multiple spectral regions such as visible light, near infrared, and short-wave infrared. Each pixel in the image corresponds to the spectral response value of the water body at a specific wavelength. Due to the combined influence of multiple factors such as dissolved matter, suspended matter, and bottom sediment in the water body, there may be differences in parameters such as reflectivity and transmittance in different bands. Based on this difference, water quality parameters can be inverted.

[0038] However, due to the high spectral resolution of hyperspectral images, there is often a high degree of correlation between adjacent bands, and this correlation is particularly significant in similar water bodies. This redundant information not only increases the difficulty and computational cost of data processing, but may also mask or interfere with key information useful for predicting water quality parameters, thereby reducing prediction accuracy. The traditional dimensionality reduction method is based on principal component analysis. This dimensionality reduction method projects the multidimensional spectral values ​​of each pixel point in the hyperspectral image and selects the first k dimensions with the largest variance in the projection direction as new dimensionality reduction data. However, due to the different redundant characteristics of water body hyperspectral images, each hyperspectral image may have different information losses and changes in the dimensionality reduction process, which makes it impossible to ensure that the input dimensions of the neural network are consistent during the training and application stages.

[0039] Based on this, this embodiment proposes a water quality parameter classification inversion method based on multidimensional hyperspectral images, which aims to reduce the dimension while making the input dimension of the network consistent in the training stage and the application stage. In this way, the water quality parameters of the water body can be inverted using hyperspectral remote sensing technology, providing a theoretical basis for the monitoring of water bodies such as lakes, reservoirs and rivers.

[0040] Figure 1 This is a schematic diagram of the process flow of the water quality parameter classification inversion method based on multidimensional hyperspectral images in this embodiment. Figure 1 The method is described in detail and specifically comprises the following steps:

[0041] S1: Obtain multiple groups of water quality index parameters corresponding to multiple different water samples, each group of water quality index parameters is used to describe the water quality characteristics of the corresponding water sample; collect each water sample at LSpectral values ​​of different spectral bands are obtained to obtain multiple groups of water quality index parameters corresponding to each other. L dimensional spectral vector, where L is a positive integer; based on water quality index parameters and L dimensional spectral vector to construct the dataset.

[0042] The water quality index parameter is a multidimensional vector, in which each dimension represents the value of a water quality index parameter. The water quality index parameter vector includes at least one of the water quality indicators such as water temperature, pH value, turbidity, total hardness, ammonia nitrogen, nitrate, total phosphorus, heavy metal ions, total coliform bacteria and chlorophyll. Each vector is composed of the specific values ​​of each index parameter, and different combinations of parameter values ​​represent different water quality characteristics. An example of a water quality parameter and the corresponding value range is shown in Table 1.

[0043] Table 1 A water quality parameter and its corresponding value range

[0044]

[0045] The hyperspectral vector contains the reflectance values ​​of water bodies in multiple different spectral bands, which usually cover visible light, near infrared and even part of the short-wave infrared region. Taking chlorophyll a as an example, chlorophyll a has obvious absorption peaks in the blue-violet light band (about 420~500nm) and the red light band (especially around 675nm), that is, the spectral value is large. As the concentration of chlorophyll a increases, this reflection peak may become more significant.

[0046] The reflectivity of water to light in different spectral bands is measured under different water quality conditions. For each specific combination of water quality parameters, a corresponding hyperspectral vector is measured, thereby establishing a one-to-one corresponding data set between the water quality parameter vector and the hyperspectral vector.

[0047] S2: Construct a deep learning network for inverting water quality parameters, use the data set to train the deep learning network, and obtain a water quality parameter inversion model.

[0048] S201: Build a deep learning network.

[0049] Cycle Generative Adversarial Networks (CycleGAN) consists of four main parts, namely two generators G and F, and two discriminators Dx and Dy. Generator G is responsible for converting data from the original domain (spectral vector) to the target domain (water quality parameters), and generator F performs the opposite operation of G, that is, converting data from the target domain back to the original domain. Discriminator Dx is responsible for judging whether the input data belongs to the original domain, and discriminator Dy is responsible for judging whether the input data belongs to the target domain.

[0050] S202: Train a deep learning network.

[0051] CycleGAN achieves data transformation and maintains the consistency of transformation by optimizing multiple loss functions. These loss functions include: Adversarial Loss, which is used to train the discriminator and generator so that the generator can generate more realistic data and the discriminator can judge the authenticity of the data more accurately. Cycle Consistency Loss is used to ensure that the data can maintain consistency when it is transformed back to the original domain. For example, for the input data x, it is transformed by G to obtain y', and then transformed back to x' by F. The cycle consistency loss requires that x' is as close to x as possible. Identity Loss is used to encourage the generator to keep the parts of the data that are not related to the transformation unchanged during the transformation process. For example, when the input data already belongs to the target domain, the generator should output the same data as the input as much as possible.

[0052] During the training process, CycleGAN gradually improves the network performance by alternately optimizing the loss functions of the generator and the discriminator. Specifically, in each iteration, the generator is first fixed to optimize the discriminator, and then the discriminator is fixed to optimize the generator. Through continuous iteration, a CycleGAN model that can achieve high-quality data conversion is finally obtained. The trained CycleGAN can achieve bidirectional conversion between water quality parameter vectors (target domain) and spectral vectors (original domain).

[0053] S3: Collect water to be tested L dimensional hyperspectral image, and L The pixels in the dimensional hyperspectral image are divided into different classes, and the water area corresponding to the same class of pixels is L The spectral values ​​of the same type of pixels belonging to the same spectral band are aggregated to obtain the corresponding L dimensional aggregated spectral vector; the corresponding L The dimensional aggregated spectral vectors are input into the inversion model respectively to obtain the water quality parameters of the water body to be tested. Figure 2 In order to obtain the flow chart of water quality parameters of the water body to be tested, the following is a specific combination of Figure 2 Detailed description of the steps to obtain the water quality parameters of the water body to be tested:

[0054] S301: Collect water to be tested L dimensional hyperspectral image.

[0055] A drone equipped with a hyperspectral imaging spectrometer is used to obtain water surface images, and preprocessing including mirror transformation and black and white frame calibration is performed. Considering that after the drone flies to a certain altitude, the hyperspectral image data obtained by the hyperspectral imager may be affected by factors such as atmosphere and water vapor. In order to eliminate the influence of these factors, a calibrated gray cloth is placed in the shooting area before the drone takes off. When the hyperspectral image is acquired, the gray cloth is placed in the hyperspectral image acquisition area. By comparing the reflectance values ​​of the water area and the gray cloth area, the influence of factors such as atmosphere and water vapor is eliminated. Figure 3 For collection L Schematic diagram of a dimensional hyperspectral image, composed of L It is composed of a two-dimensional matrix of size M×N, where M×N represents the size of the hyperspectral image in two-dimensional space, M is the number of pixels in the horizontal direction, and N is the number of pixels in the vertical direction. Each two-dimensional matrix corresponds to a spectral band, and the value of each pixel in the two-dimensional matrix represents the spectral response of the pixel in the corresponding spectral band.

[0056] S302: L The pixels in the 3D hyperspectral image are segmented into different classes.

[0057] Water bodies in nature are not homogeneous. Water bodies of different water qualities will show different reflectance spectral characteristics in hyperspectral images, which will increase the computational complexity and distract attention because the algorithm needs to process multiple spectral features at the same time. Therefore, before inversion, the k-means clustering method is used to cluster the L This problem can be avoided by dividing the pixels in the 2D hyperspectral image into different classes. K-means clustering calculates the distance between the spectral values ​​of different pixels and clusters the pixels whose spectral value distance is less than the set threshold into one class. Specifically:

[0058] First, the hyperspectral image is converted into a sample matrix, where each row represents a pixel point. L dimensional vector, each column represents a band; and the data is standardized or normalized by dividing the pixel value of each band by the maximum value or standard deviation of the band to ensure that the value range of different bands has the same impact on the clustering results. L dimensional vector as the initial clustering center, these centers will represent the initial k classes. Then, for each pixel in the image, that is, each Ldimensional vector, calculate its Euclidean distance, cosine distance or Mahalanobis distance with each cluster center, and assign each pixel to the cluster center closest to it, that is, classify it into the corresponding class. Then, for each class, calculate the mean vector of all pixels in it, and use the mean vector as the new cluster center. Finally, repeat the iterative optimization until the convergence condition is met, such as the change of the cluster center is less than a certain threshold or the preset maximum number of iterations is reached. Figure 4 This is the flow chart of the k-means clustering algorithm, including:

[0059] I. Calculate the distance from each non-initial classification pixel to the initial classification pixel based on the Euclidean distance.

[0060] II. Classify each non-initial classified pixel into the class to which the initial segmentation point with the smallest distance belongs.

[0061] III. For the classified pixels, calculate the centroid of each class of pixels and use the centroid as the new initial classification pixel of each class.

[0062] IV. Repeat steps I to III, and stop the calculation when the position change of the initial classified pixel is less than the set threshold.

[0063] After clustering is completed, each pixel will be assigned a class label to indicate the category it belongs to. The hyperspectral image of the water body to be detected is divided into different categories. Water bodies of the same category have similar water quality index parameters. This method reduces the intra-class differences and increases the inter-class differences, which helps to process the same type of data more focused and can more accurately invert different water quality states.

[0064] S303: Aggregate the spectral values ​​of the same type of pixels belonging to the same spectral band to obtain the corresponding L dimensional aggregated spectral vector.

[0065] By calculating the spectral values ​​of the same type of water bodies and aggregating them, the common characteristics of the same type of water bodies are fully utilized to obtain the high spectral vectors of each class, thereby improving the accuracy and robustness of the prediction. Specifically:

[0066] Get the spectral values ​​of the same type of pixels belonging to the same spectral band, and get the average, mode or median of the spectral values ​​of the same type of pixels in the same spectral band; L Each spectral band is subjected to average mode or median extraction operation to obtain the corresponding values ​​of various pixel points. L dimensional aggregated spectral vector. L Take the averaging operation for each spectral band as an example:

[0067]

[0068]

[0069] in, n k Indicates k The number of class pixels, Indicates i spectral bands, Indicates k The first pixel in the j pixels in the spectral band The spectral value of Indicates k Similar pixels in the spectral band The spectral average value of ; the above formula is used to aggregate similar pixels through the averaging operation; ~ Indicates k Class pixel points from spectral bands ~ Spectral bands The average value of the spectrum.

[0070] S304: Corresponding to each type of pixel L The dimensional aggregated spectral vectors are input into the inversion model respectively to obtain the water quality parameters of the water body to be tested.

[0071] based on Figure 1 The water quality parameter classification inversion method based on multidimensional hyperspectral images shown in the figure divides the hyperspectral image of the water body to be detected into different classes to group water body areas with similar water quality characteristics together; the spectral values ​​of similar pixels belonging to the same spectral band are aggregated, and the aggregation result can reflect the concentrated characteristics of the pixels of this category in the same spectral band. This aggregation method not only reduces the redundancy of spectral features of similar water bodies, but also does not change the number or order of spectral bands of each pixel, thereby ensuring the consistency of the input dimensions of the deep learning network in the training stage and the application stage, thereby ensuring the accuracy of the deep learning network in inverting water quality parameters in the application stage.

[0072] The above is a water quality parameter classification inversion method based on multidimensional hyperspectral images provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding water quality parameter classification inversion system based on multidimensional hyperspectral images. Figure 5 This is a schematic diagram of the framework of the water quality parameter classification inversion system based on multi-dimensional hyperspectral images, including:

[0073] The data acquisition module is used to obtain multiple groups of water quality index parameters corresponding to multiple different water samples. Each group of water quality index parameters is used to describe the water quality characteristics of the corresponding water sample. L Spectral values ​​of different spectral bands are obtained to obtain multiple groups of water quality index parameters corresponding to each other. L dimensional spectral vector, where L is a positive integer; based on water quality index parameters and L dimensional spectral vector to construct the dataset.

[0074] The model building module is used to build a deep learning network for inverting water quality parameters, and use the data set to train the deep learning network to obtain a water quality parameter inversion model.

[0075] Water quality parameter inversion module, used to collect the water quality parameters of the water body to be tested. L dimensional hyperspectral image, and L The pixels in the dimensional hyperspectral image are divided into different classes, and the water area corresponding to the same class of pixels is L The spectral values ​​of the same type of pixels belonging to the same spectral band are aggregated to obtain the corresponding L dimensional aggregated spectral vector; the corresponding L The dimensional aggregated spectral vectors are input into the inversion model respectively to obtain the water quality parameters of the water body to be tested.

[0076] For the specific definition of the water quality parameter classification inversion system based on multidimensional hyperspectral images, please refer to the definition of the water quality parameter classification inversion method based on multidimensional hyperspectral images in the above text, which will not be repeated here. Each module in the above-mentioned water quality parameter classification inversion system based on multidimensional hyperspectral images can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0077] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0078] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.

Claims

1. A water quality parameter classification inversion method based on multidimensional hyperspectral images, characterized in that: include: Acquire multiple groups of water quality index parameters corresponding to multiple different water samples, each group of water quality index parameters is used to describe the water quality characteristics of the corresponding water sample; collect spectral values ​​of each water sample in Λ different spectral bands, and obtain multiple groups of Λ-dimensional spectral vectors corresponding to the multiple groups of water quality index parameters, where Λ is a positive integer; construct a data set based on the water quality index parameters and the Λ-dimensional spectral vectors; Construct a deep learning network for inverting water quality parameters, use the data set to train the deep learning network, and obtain a water quality parameter inversion model; Collect a Λ-dimensional hyperspectral image of the water body to be detected, and divide the pixels in the Λ-dimensional hyperspectral image into different classes. The water body areas corresponding to the same class of pixels have similar spectral values ​​in Λ different spectral bands; and aggregate the spectral values ​​of the same class of pixels belonging to the same spectral band to obtain the Λ-dimensional aggregated spectral vectors corresponding to each class of pixels; input the Λ-dimensional aggregated spectral vectors corresponding to each class of pixels into the inversion model to obtain the water quality parameters of the water body to be detected; Aggregating the spectral values ​​of the same type of pixels belonging to the same spectral band to obtain the Λ-dimensional aggregated spectral vectors corresponding to the various types of pixels specifically includes: Get the spectral values ​​of similar pixels belonging to the same spectral band, and calculate the average spectral values ​​of similar pixels in the same spectral band: Among them, n k represents the number of pixels in the kth category, λ i represents the i-th spectral band, Indicates the jth pixel in the kth class of pixels in the spectral band λ i The spectral value of Indicates that the k-th pixel point is in the spectral band λ i The spectral average value of ; the above formula is used to aggregate similar pixels through the averaging operation; The Λ spectral bands of each type of pixel are averaged to obtain the Λ-dimensional aggregated spectral vector corresponding to each type of pixel: in, Indicates that the k-th pixel point is from spectral band λ1 to spectral band λ Λ The spectral average of or: Obtain the spectral values ​​of similar pixel points belonging to the same spectral band, and extract the mode or median of the spectral values ​​of similar pixel points in the same spectral band; The mode or median extraction operation is performed on the Λ spectral bands of each type of pixel point to obtain the Λ-dimensional aggregated spectral vector corresponding to each type of pixel point.

2. The water quality parameter classification inversion method based on multidimensional hyperspectral images according to claim 1, characterized in that: The pixel points in the Λ-dimensional hyperspectral image are divided into different classes by using a k-means clustering method; the k-means clustering calculates the distance between the spectral values ​​of different pixels and clusters the pixels whose spectral value distance is less than a set threshold into one class.

3. The water quality parameter classification inversion method based on multidimensional hyperspectral images according to claim 1, characterized in that: The deep learning network adopts a cyclic generative adversarial network; During the training process of the cyclic generative adversarial network, the original Λ-dimensional spectral vector is converted into pseudo water quality parameters, and then the pseudo water quality parameters are converted back into the Λ-dimensional spectral vector; the gap between the original Λ-dimensional spectral vector and the Λ-dimensional spectral vector converted back from the pseudo water quality parameters is minimized by the cycle consistency loss function, so that the cyclic generative adversarial network learns the mapping relationship between the Λ-dimensional spectral vector and the water quality parameters.

4. A water quality parameter classification inversion system based on multidimensional hyperspectral images, characterized in that: include: A data acquisition module is used to obtain multiple groups of water quality index parameters corresponding to multiple different water samples, each group of water quality index parameters is used to describe the water quality characteristics of the corresponding water sample; collect the spectral values ​​of each water sample in Λ different spectral bands, and obtain multiple groups of Λ-dimensional spectral vectors corresponding to the multiple groups of water quality index parameters, where Λ is a positive integer; and construct a data set based on the water quality index parameters and the Λ-dimensional spectral vectors; A model building module is used to build a deep learning network for inverting water quality parameters, and train the deep learning network using a data set to obtain a water quality parameter inversion model; The water quality parameter inversion module is used to collect the Λ-dimensional hyperspectral image of the water body to be detected, and divide the pixels in the Λ-dimensional hyperspectral image into different classes. The water body areas corresponding to the same class of pixels have similar spectral values ​​in Λ different spectral bands; and the spectral values ​​of the same class of pixels belonging to the same spectral band are aggregated to obtain the Λ-dimensional aggregated spectral vectors corresponding to each class of pixels; the Λ-dimensional aggregated spectral vectors corresponding to each class of pixels are respectively input into the inversion model to obtain the water quality parameters of the water body to be detected; Aggregating the spectral values ​​of the same type of pixels belonging to the same spectral band to obtain the Λ-dimensional aggregated spectral vectors corresponding to the various types of pixels specifically includes: Get the spectral values ​​of similar pixels belonging to the same spectral band, and calculate the average spectral values ​​of similar pixels in the same spectral band: Among them, n k represents the number of pixels in the kth category, λ i represents the i-th spectral band, Indicates the jth pixel in the kth class of pixels in the spectral band λ i The spectral value of Indicates that the k-th pixel point is in the spectral band λ i The spectral average value of ; the above formula is used to aggregate similar pixels through the averaging operation; The Λ spectral bands of each type of pixel are averaged to obtain the Λ-dimensional aggregated spectral vector corresponding to each type of pixel: in, Indicates that the k-th pixel point is from spectral band λ1 to spectral band λ Λ The spectral average of or: Obtain the spectral values ​​of similar pixel points belonging to the same spectral band, and extract the mode or median of the spectral values ​​of similar pixel points in the same spectral band; The mode or median extraction operation is performed on the Λ spectral bands of each type of pixel point to obtain the Λ-dimensional aggregated spectral vector corresponding to each type of pixel point.

5. The water quality parameter classification inversion system based on multidimensional hyperspectral images as claimed in claim 4, characterized in that: The water quality parameter inversion module is specifically used for: The k-means clustering method is used to segment the pixels in the Λ-dimensional hyperspectral image into different classes; the k-means clustering method calculates the distance between the spectral values ​​of different pixels and clusters the pixels whose spectral value distance is less than a set threshold into one class.

6. The water quality parameter classification inversion system based on multidimensional hyperspectral images according to claim 4, characterized in that: The model building module is specifically used for: Adopting a recurrent generative adversarial network; During the training process of the cyclic generative adversarial network, the original Λ-dimensional spectral vector is converted into pseudo water quality parameters, and then the pseudo water quality parameters are converted back into the Λ-dimensional spectral vector; the gap between the original Λ-dimensional spectral vector and the Λ-dimensional spectral vector converted back from the pseudo water quality parameters is minimized by the cycle consistency loss function, so that the cyclic generative adversarial network learns the mapping relationship between the Λ-dimensional spectral vector and the water quality parameters.