CN117110217B - Three-dimensional water quality monitoring method and system - Google Patents
Three-dimensional water quality monitoring method and system Download PDFInfo
- Publication number
- CN117110217B CN117110217B CN202311374660.1A CN202311374660A CN117110217B CN 117110217 B CN117110217 B CN 117110217B CN 202311374660 A CN202311374660 A CN 202311374660A CN 117110217 B CN117110217 B CN 117110217B
- Authority
- CN
- China
- Prior art keywords
- water
- water quality
- representing
- target element
- module
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 459
- 238000012544 monitoring process Methods 0.000 title claims abstract description 46
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000009826 distribution Methods 0.000 claims abstract description 79
- 238000000605 extraction Methods 0.000 claims abstract description 52
- 230000011218 segmentation Effects 0.000 claims abstract description 33
- 230000002776 aggregation Effects 0.000 claims abstract description 30
- 238000004220 aggregation Methods 0.000 claims abstract description 30
- 238000001228 spectrum Methods 0.000 claims abstract description 30
- 230000008447 perception Effects 0.000 claims abstract description 9
- 230000003595 spectral effect Effects 0.000 claims description 91
- 239000013598 vector Substances 0.000 claims description 33
- 238000012795 verification Methods 0.000 claims description 22
- 238000012549 training Methods 0.000 claims description 21
- 238000004364 calculation method Methods 0.000 claims description 19
- 238000013507 mapping Methods 0.000 claims description 18
- 238000005070 sampling Methods 0.000 claims description 17
- 238000007637 random forest analysis Methods 0.000 claims description 12
- 238000002310 reflectometry Methods 0.000 claims description 12
- 241001113556 Elodea Species 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 238000012163 sequencing technique Methods 0.000 claims description 8
- 208000034699 Vitreous floaters Diseases 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000003860 storage Methods 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 238000012352 Spearman correlation analysis Methods 0.000 claims description 3
- 244000025254 Cannabis sativa Species 0.000 claims description 2
- 230000035772 mutation Effects 0.000 claims 1
- 241000196324 Embryophyta Species 0.000 description 8
- 238000010586 diagram Methods 0.000 description 5
- 230000001537 neural effect Effects 0.000 description 5
- 238000012937 correction Methods 0.000 description 4
- 238000003062 neural network model Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- XKMRRTOUMJRJIA-UHFFFAOYSA-N ammonia nh3 Chemical compound N.N XKMRRTOUMJRJIA-UHFFFAOYSA-N 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000013508 migration Methods 0.000 description 2
- 230000005012 migration Effects 0.000 description 2
- 239000001301 oxygen Substances 0.000 description 2
- 229910052760 oxygen Inorganic materials 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 241001465754 Metazoa Species 0.000 description 1
- 208000027066 STING-associated vasculopathy with onset in infancy Diseases 0.000 description 1
- 238000009360 aquaculture Methods 0.000 description 1
- 244000144974 aquaculture Species 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000003556 assay Methods 0.000 description 1
- 230000003750 conditioning effect Effects 0.000 description 1
- 239000000356 contaminant Substances 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000013486 operation strategy Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000013441 quality evaluation Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/7715—Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/88—Image or video recognition using optical means, e.g. reference filters, holographic masks, frequency domain filters or spatial domain filters
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/05—Underwater scenes
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Pathology (AREA)
- Artificial Intelligence (AREA)
- Biochemistry (AREA)
- Analytical Chemistry (AREA)
- Chemical & Material Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Immunology (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a three-dimensional water quality monitoring method and system, and belongs to the technical field of water quality monitoring. According to the invention, medium-small area discrete distribution water information is adaptively captured through the water discrete feature extraction module, large-area continuous water semantics are fully perceived through the water space distribution perception module in terms of space distribution, and a water feature aggregation module is constructed to establish a dependency relationship between key semantic information obtained by the water space distribution perception module and key semantic information respectively, so that different area information is obtained by carrying out area segmentation on a water quality image, then inversion combination of the spectrum vegetation indexes is established through the spectrum vegetation indexes of each area information and water quality target element data, a water quality target element inversion model is constructed through the inversion combination of the spectrum vegetation indexes, and water quality is continuously monitored in large area through the water quality target element inversion model, so that sudden water quality problems can be timely found and targeted, the water quality space distribution condition is effectively reflected, and the method has strong practicability and wide applicability.
Description
Technical Field
The invention relates to the technical field of water quality monitoring, in particular to a three-dimensional water quality monitoring method and system.
Background
As the global population grows and economies develop, water resources face significant challenges. In this context, for the agricultural planting and aquaculture industry, the quality of water directly affects the growth and yield of animals and plants and may lead to diseases, increased mortality and slower growth rates of farmed organisms. High throughput monitoring of water quality and real-time regulation of water quality resources is particularly necessary under multiple pressures that ensure quality, yield enhancement and sustainable resource utilization.
Traditional water quality monitoring methods often rely on manual sampling and then analysis in the laboratory. Although the method has higher accuracy, the method is time-consuming and expensive, cannot realize large-area real-time or continuous monitoring, can only obtain local point location data information, lacks water quality spatial distribution status information of water bodies, and is difficult to discover and cope with sudden water quality problems in time. Although there are many techniques for water quality monitoring, they tend to cover only a certain portion of the water body. For example, only the water surface is monitored or only certain specific contaminants are of interest. Therefore, there is an urgent need for an efficient, real-time and all-round water quality monitoring and conditioning system.
Through searching, chinese patent application, application number 202210801133.3, publication day 2022, 10 and 14, discloses a water quality monitoring method based on BP neural network. The method comprises the following steps: s1, collecting water quality sample data; s2, performing feature marking on the water quality sample data to obtain marked sample data; s3, inputting the marked sample data into a BP neural network model for network model training to obtain a trained neural network model; s4, acquiring water quality real-time data, inputting the water quality real-time data into the trained neural network model, and outputting a water quality evaluation result. The method can realize intelligent real-time monitoring of water quality, but the method can not realize large-area real-time or continuous water quality monitoring, and is difficult to reflect the spatial distribution condition of water quality.
Disclosure of Invention
1. Technical problem to be solved
Aiming at the problems that the traditional artificial water quality sampling monitoring is time-consuming and expensive, large-area real-time or continuous monitoring cannot be realized, the water quality spatial distribution condition is difficult to reflect and the like in the prior art, the invention provides a three-dimensional water quality monitoring method and system.
2. Technical proposal
The aim of the invention is achieved by the following technical scheme.
A three-dimensional water quality monitoring method comprises the following steps:
collecting water quality sample data, processing the water quality sample data to obtain water quality target element data, and dividing the water quality target element data into training set data and verification set data;
acquiring a water quality image data set and inputting a water quality image;
carrying out region segmentation on the water quality image through a semantic segmentation model to obtain different region information, calculating a spectral vegetation index of each region information, obtaining an inversion combination of the spectral vegetation index by utilizing the spectral vegetation index and water quality target element data, and constructing a water quality target element inversion model through the inversion combination of the spectral vegetation index;
And training the water quality target element inversion model through training set data to obtain a trained water quality target element inversion model, inputting verification set data into the trained water quality target element inversion model for verification, and outputting a water quality monitoring result.
Further, the semantic segmentation model comprises a water discrete feature extraction module, a water space distribution perception module and a water feature aggregation module; the water discrete feature extraction module and the water space distribution sensing module respectively acquire semantic information of the water quality image, and the water feature aggregation module processes the semantic information acquired by the water discrete feature extraction module and the water space distribution sensing module to acquire different area information in the water quality image.
Further, a calculation formula for acquiring key semantic information of the water quality image through the water discrete feature extraction module is as follows:
wherein,represents the discrete semantic information of the water body,MSArepresenting a multi-headed self-attention operation,α 1 、α 2 、α 3 all representing different high-dimensional mapping matrices, < >>Representing the input of the water discrete feature extraction module,xrepresenting the abscissa of the discrete feature anchor points,yrepresenting the ordinate of the discrete feature anchor point,Krepresenting the number of discrete feature anchor points, L K Representing optimized discrete feature anchor points, +.>Representing the sampling operation of the optimized discrete feature anchor points,θan adaptive offset sub-network is shown,Qrepresenting a high-dimensional space,F Q representing a mapped matrix via high dimensionsα 1 Mapping to obtain the water body first-order discrete features.
Further, the calculation formula for acquiring the semantic information of the water quality image through the water space distribution sensing module is as follows:
wherein,representing the spatial distribution semantic information of the water body,Y R representing the extraction result of the water space distribution sensing module on the horizontal axis component,Y C representing the extraction result of the spatial distribution sensing module of the water body on the longitudinal axis component, the back-up represents element-by-element addition,MSArepresenting a multi-headed self-attention operation,C 1 、C 2 、C 3 all representing different convolution operations, +.>The output of the water space distribution sensing module is expressed as the initial row vector of the water divided by the input triples>The output of the water distribution sensing module is expressed as the initial column vector of the water divided by the input triplets>Representing a learnable offset superimposed on the initial line vector,/->Representing a learnable offset variable superimposed on the initial column vector.
Further, the water body characteristic aggregation module is used for coupling and decoupling semantic information acquired by the water body discrete characteristic extraction module and the water body space distribution sensing module, and a calculation formula for acquiring different area information in the water quality image is as follows:
Wherein,representing the output of the water body characteristic aggregation module to the water body discrete characteristic extraction module,/for>The output of the water body characteristic aggregation module to the water body space distribution sensing module is represented,ithe natural number is represented by a number of natural numbers,O MLP a multi-layer perceptron is shown,softmaxrepresenting normalization operations->Representing the output of the ith water discrete feature extraction module,/->Representing the output of the ith water body space distribution sensing module,/->、/>、/>、/>、/>、/>Respectively, six different mapping matrices, H denotes a transpose operation,Q1、Q2、K1、K2、V1、V2respectively representing six different high-dimensional spaces corresponding to six different mapping matrices,Crepresenting the number of channels entered.
Further, calculating a spearman correlation coefficient of the spectral vegetation index and the water quality target element data, setting a threshold lambda, and selecting the spectral vegetation index with the spearman correlation coefficient larger than the threshold lambda; the calculation formula of the spearman correlation coefficient of the spectral vegetation index and the water quality target element data is as follows:
wherein,SRCrepresenting the spearman correlation coefficient,Nsample size representing spectral vegetation index and water quality target element data, i represents the current sample number,R i representing the value corresponding to the current sample number of the spectral vegetation index,represents the average value of the spectral vegetation index, S i A value corresponding to the current number of the water quality target element data, < >>The average value of the water quality target element data is shown.
Further, the spectrum vegetation indexes with the spearman correlation coefficient larger than the threshold lambda are subjected to sequencing combination, and a random forest model is established by utilizing the spectrum vegetation indexes after sequencing combination and water quality target element data to obtain inversion combination of the spectrum vegetation indexes.
Further, different regional information features in inversion combination of the spectrum vegetation indexes are extracted, different regional information features are fused to obtain a multi-scale network architecture, the fused multi-scale architecture is used as a parent network, the trained parent network is subjected to performance sequencing and then inheritance and variation are used as child networks, knowledge of the parent network is migrated to the child networks, the child networks are subjected to performance sequencing, and individual selection is performed on the parent network and the child networks, so that an optimal water quality target element inversion model is obtained.
A stereoscopic water quality monitoring system, comprising:
the acquisition module acquires water quality sample data, processes the water quality sample data to obtain water quality target element data, and divides the water quality target element data into training set data and verification set data;
The input module acquires a water quality image data set and inputs a water quality image;
the model construction module is used for carrying out region segmentation on the water quality image through the semantic segmentation model to obtain different region information, calculating a spectral vegetation index of each region information, obtaining an inversion combination of the spectral vegetation index by utilizing the spectral vegetation index and the water quality target element data, and constructing a water quality target element inversion model through the inversion combination of the spectral vegetation index;
the monitoring module trains the water quality target element inversion model through the training set data to obtain a trained water quality target element inversion model; inputting the verification set data into the trained water quality target element inversion model for verification, and outputting a water quality monitoring result.
A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the method described above.
3. Advantageous effects
Compared with the prior art, the invention has the advantages that:
according to the three-dimensional water quality monitoring method and system, the water quality image is accurately segmented through the semantic segmentation model, so that the information characteristics of different areas are obtained, errors caused by similar reflectivity can be effectively avoided, and accordingly sudden water quality problems can be timely found and targeted. Meanwhile, the inversion combination of the spectrum vegetation index is obtained through the spectrum vegetation index and the water quality target element data, the water quality target element inversion model is constructed through the inversion combination of the spectrum vegetation index, and the water quality can be continuously monitored in large area in real time by utilizing the water quality target element inversion model, so that the water quality space distribution condition can be effectively reflected. In addition, compared with the traditional manual water quality sampling method for monitoring water quality, the three-dimensional water quality monitoring method and system provided by the invention are low in cost, wide in monitoring range, and strong in practicability and wide in applicability.
Drawings
FIG. 1 is a flow chart of a water quality monitoring method according to an embodiment of the invention;
FIG. 2 is a schematic diagram of a semantic segmentation model according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a water discrete feature extraction module according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a water space distribution sensing module according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a water feature aggregation module according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of a water quality target element inversion model according to an embodiment of the invention.
Detailed Description
The invention will now be described in detail with reference to the drawings and the accompanying specific examples.
Examples
The three-dimensional water quality monitoring method provided by the embodiment comprises the following steps: collecting water quality sample data, processing the water quality sample data to obtain water quality target element data, and dividing the water quality target element data into training set data and verification set data; acquiring a water quality image data set and inputting a water quality image; carrying out region segmentation on the water quality image through a semantic segmentation model to obtain different region information, calculating a spectral vegetation index of each region information, obtaining an inversion combination of the spectral vegetation index by utilizing the spectral vegetation index and water quality target element data, and constructing a water quality target element inversion model through the inversion combination of the spectral vegetation index; and training the water quality target element inversion model through training set data to obtain a trained water quality target element inversion model, inputting verification set data into the trained water quality target element inversion model for verification, and outputting a water quality monitoring result.
In this embodiment, first, water quality sample data is collected, water quality sample data is processed to obtain water quality target element data, and the water quality target element data is divided into training set data and verification set data. Specifically, in this embodiment, a plurality of water quality sampling points are uniformly distributed on a satellite image in a research area, longitude and latitude coordinates of the water quality sampling points are determined to collect water quality sample data, a water quality target element analysis chemical assay is performed on the water quality sample data, so as to obtain water quality target element data, and the water quality target element data is divided into training set data and verification set data. In this embodiment, the water quality target element data includes PH, suspended matter detection, ammonia nitrogen amount, dissolved oxygen concentration and turbidity.
Further, as shown in fig. 1, a water quality image dataset is acquired, and a water quality image is input. In the embodiment, water quality sampling points are shot above a research area to obtain a water quality image dataset, and then the obtained water quality image dataset is preprocessed to obtain a water quality image with true reflectivity of the earth surface. In this embodiment, the unmanned aerial vehicle may be used to carry a multispectral camera to shoot the water quality sampling point above the research area, so as to obtain a water quality image dataset. In this embodiment, the pretreatment includes operations such as radiation calibration, orthographic correction, and image stitching on the water quality image. Specifically, the radiometric calibration is carried out by shooting three calibration plate images with fixed reflectivity, inputting the reflectivity of the calibration plate is completed by means of the existing DJI Terra software, so that calibration of multispectral five-band images is completed, then space three-solution and correction are carried out on the DJI Terra software by means of homonymous point matching through positioning information recorded by the existing unmanned aerial vehicle aircraft, an orthophoto is generated, finally image splicing is completed by means of the DJI Terra software, and therefore, a water quality image with the ground surface true reflectivity is obtained through operations such as radiometric calibration, orthographic correction, image splicing and the like. In this embodiment, operations such as radiation calibration, orthographic correction and image stitching are all performed on a water quality image. In this embodiment, the multispectral camera used has five bands, which are: the red wave band is 650+/-16 nm and is marked as R1; the green wave band is 560+/-16 nm and is marked as R2; blue band 450+ -16 nm, denoted as R3; the near red wave band 840+/-16 nm is marked as R4; the red band is 730+/-16 nm and is marked as R5; the ground surface true reflectivities of the water quality sampling points corresponding to the spectral reflectivities of the wave bands R1, R2, R3, R4 and R5 one by one are as follows: r1', R2', R3', R4', R5'.
Further, the water quality image is subjected to regional segmentation through the semantic segmentation model to obtain different regional information, the spectral vegetation index of each regional information is calculated, the inversion combination of the spectral vegetation index is obtained through the spectral vegetation index and the water quality target element data, and the water quality target element inversion model is constructed through the inversion combination of the spectral vegetation index.
As shown in fig. 2, the water quality image is input into a semantic segmentation model, which performs region segmentation on the water quality image. In this embodiment, the semantic segmentation model is used to segment the water quality image, including the water surface, the float grass, the moss and the sundries in the water quality image, so that the semantic segmentation model is used to accurately segment the water quality image, thereby obtaining specific information of different areas and avoiding errors caused by similar reflectivity. In this embodiment, the semantic segmentation model includes a water discrete feature extraction module, a water spatial distribution perception module, and a water feature aggregation module. The water discrete feature extraction module and the water space distribution sensing module respectively acquire key semantic information of the water quality image, and the water feature aggregation module couples and decouples the key semantic information acquired by the water discrete feature extraction module and the water space distribution sensing module to acquire different area information in the water quality image.
Specifically, as shown in fig. 3, the water discrete feature extraction module can adaptively capture features of water surfaces, aquatic weeds, moss floaters and sundries which are discretely distributed in a small area. In the water discrete feature extraction module, the input water quality image is subjected to high-dimensional mapping matrix to obtain first-order discrete features of the water body and clustered to generate a plurality of discrete feature positioning points, and further in the embodiment, the discrete feature positioning points are optimized through the self-adaptive migration sub-network, water quality information obtained by sampling in the neighborhood of each optimized discrete feature positioning point is subjected to high-dimensional mapping matrix to obtain water discrete combination features, and therefore multi-head self-attention calculation is performed on the obtained water discrete combination features to obtain water discrete semantic information. In this embodiment, the calculation formula for acquiring the key semantic information of the water quality image by the water discrete feature extraction module is as follows:
wherein,represents the discrete semantic information of the water body,MSArepresenting a multi-headed self-attention operation,α 1 、α 2 、α 3 all representing different high-dimensional mapping matrices, < >>Representing the input of the water discrete feature extraction module,xrepresenting the abscissa of the discrete feature anchor points,yrepresenting the ordinate of the discrete feature anchor point,Krepresenting the number of discrete feature anchor points, L K Representing optimized discrete feature anchor points, +.>Representing the sampling operation of the optimized discrete feature anchor points,θan adaptive offset sub-network is shown,Qrepresenting a high-dimensional space,F Q representing a mapped matrix via high dimensionsα 1 Mapping to obtain the water body first-order discrete features. In this embodiment, the input of the water discrete feature extraction module may be a water quality image or a feature output by the water feature aggregation module in the previous stage.
As shown in fig. 4, the water body space distribution sensing module can fully sense targets such as large-area continuous water surfaces, aquatic weeds, moss floaters, sundries and the like at a space distribution angle. This practice isIn the embodiment, in the water space distribution sensing module, for the input water quality image, the actual water quality area density characteristic and the calculation power resource can be combined according to the super parameter N R 、N C 、N I Respectively setting the number of rows, the number of columns and the distance, and the water quality image is based on the triplet [ N ] R 、N C 、N I ]Self-attention operation strategies are distributed, and automatic zero-filling operation is carried out to adapt to different triplet settings. In this embodiment, the number of rows is 3, the number of columns is 3, and the pitch is 2, namely, [3, 2 ]]For example, the divided row vectors and column vectors are aggregated to obtain an initial water body row vector and an initial water body column vector respectively. In this embodiment, the initial line vector of the water body is expressed as: The initial column vector of the water body is expressed as:. Respectively superposing the water body initial row vector and the water body initial column vector with the learnable offset variable +.>、/>Therefore, the water body offset row vector and the water body offset column vector are obtained by dynamically adjusting the distribution sensing strategy of the water body space distribution sensing module according to the water source environment. In this embodiment, the water offset row vector is expressed as:the water offset column vector is expressed as:at the same time, three different convolution operations are usedC 1 、C 2 、C 3 Generating a corresponding query vector Q, a key vector K and a value vector V to participate in self-attention operation, and finally adding operation results element by element to obtain an output result of water body space distribution perception, thereby realizing the water body space distribution perceptionAnd outputting the result to obtain the key semantic information in the water quality image. In this embodiment, the calculation formula for acquiring the semantic information of the water quality image by the water space distribution sensing module is as follows:
wherein,representing the spatial distribution semantic information of the water body,Y R representing the extraction result of the water space distribution sensing module on the horizontal axis component,Y C representing the extraction result of the spatial distribution sensing module of the water body on the longitudinal axis component, the back-up represents element-by-element addition,MSArepresenting a multi-headed self-attention operation, C 1 、C 2 、C 3 All representing different convolution operations, +.>The output of the water space distribution sensing module is expressed as the initial row vector of the water divided by the input triples>The output of the water distribution sensing module is expressed as the initial column vector of the water divided by the input triplets>Representing a learnable offset superimposed on the initial line vector,/->Representing a learnable offset variable superimposed on the initial column vector. In this embodiment, <' > in addition>The water quality image can be expressed as an input water quality image or the initial water line vector which is divided by the output characteristics of the water distribution sensing module according to the input triples by the water body characteristic aggregation module at the last stage>The method can be expressed as an input water quality image or a water body initial column vector which is divided by the output characteristics of the water body distribution sensing module by the water body characteristic aggregation module at the last stage according to the input triples.
As shown in fig. 5, the water feature aggregation module is configured to establish a dependency relationship with key semantic information obtained by the water discrete feature extraction module and the water spatial distribution sensing module, so as to enhance the processing capability of the semantic segmentation model on the diversity features. In the water body characteristic aggregation module, water body discrete semantic information and water body space distribution semantic information are mapped into six different high-dimensional spaces through two pairs of parameter matrixes respectively, so that the output of the water body discrete characteristic extraction module and the output of the water body space distribution sensing module are obtained through the mixing operation respectively. In this embodiment, the water feature aggregation module couples and decouples the key semantic information acquired by the water discrete feature extraction module and the water space distribution sensing module, so as to obtain the calculation formulas of different area information in the water quality image, wherein the calculation formulas are as follows:
Wherein,representing the output of the water body characteristic aggregation module to the water body discrete characteristic extraction module,/for>The output of the water body characteristic aggregation module to the water body space distribution sensing module is represented,ithe natural number is represented by a number of natural numbers,O MLP a multi-layer perceptron is shown,softmaxrepresenting normalization operations->Representing the output of the ith water discrete feature extraction module,/->Representing the output of the ith water body space distribution sensing module,/->、/>、/>、/>、/>、/>Respectively, six different mapping matrices, H denotes a transpose operation,Q1、Q2、K1、K2、V1、V2respectively representing six different high-dimensional spaces corresponding to six different mapping matrices,Crepresenting the number of channels entered.
After the water quality image is input into the semantic segmentation model, key semantic information in the water quality image is captured simultaneously through the water discrete feature extraction module and the water space distribution sensing module, the water discrete feature extraction module is output by the water feature aggregation module and is coupled with and decoupled from the water space distribution sensing module layer by layer, the resolution of the water quality image is restored by the encoder, and finally the pixel classification result is output, so that the distribution situation of water important monitoring objects such as water surfaces, water plants, moss floaters, sundries and the like is obtained.
Further, after the water quality image is subjected to regional segmentation to obtain different regional information through the semantic segmentation model, calculating a spectral vegetation index of each regional information, obtaining an inversion combination of the spectral vegetation index by utilizing the spectral vegetation index and the water quality target element data, and constructing a water quality target element inversion model through the inversion combination of the spectral vegetation index. In this embodiment, according to the distribution of water quality sampling points, the true reflectivity of the ground surface in four areas of the water surface, the float of the aquatic weed, the float of the moss and the sundries is extracted respectively, and the spectral vegetation index corresponding to each area is calculated. Specifically, spectral images corresponding to the bands R1, R2, R3, R4 and R5 of different areas after segmentation are imported into the existing ENVI5.3 software, an interested region is drawn by taking a water quality sampling point as a center and a radius 1m as a boundary by utilizing an ROI tool in the ENVI5.3 software, the reflectances of all pixel points in the interested region are extracted from the spectral images of the bands R1, R2, R3, R4 and R5, and an average value is calculated to serve as ground surface true reflectances R1', R2', R3', R4' and R5' of water quality sampling points, so that a spectral vegetation index is calculated. In this embodiment, the calculated spectral vegetation index includes: NDVI ((R4 ' -R1 ')/(R4 ' +r1 ')), GNDVI ((R4 ' -R2 ')/(R4 ' +r2 ')), EVI (2.5× (R4 ' -R1 ')/(R4 ' +6xa1-7.5×r3' +1)), LCI ((R4 ' -A5)/(R4 ' +r1 ')), SAVI (1.5× (R4 ' -R1 ')/(R4 ' +r1' +0.5)), GSAVI (1.5× (A4-R2 ')/(R4 ' +r2' +0.5)), OSAVI ((R4 ' -R1 ')/(R4 ' +r1' +0.5)), NDREI ((R4 ' -A5)/(R4 ' +r4 ' +a5)). In this embodiment, the spectral vegetation indexes calculated in the four areas of the water surface, the waterweed, the moss floater and the sundries are respectively recorded as: group a spectral vegetation index, group B spectral vegetation index, group C spectral vegetation index, group D spectral vegetation index. Further, the spectral vegetation indexes in different areas are respectively subjected to spearman correlation analysis with the water quality target element data, so that the spectral vegetation indexes with higher correlation with the water quality target element data are obtained. In the embodiment, calculating a spearman correlation coefficient of the spectral vegetation index and water quality target element data, setting a threshold lambda, and selecting the spectral vegetation index with the spearman correlation coefficient larger than the threshold lambda; the calculation formula of the spearman correlation coefficient of the spectral vegetation index and the water quality target element data is as follows:
Wherein,SRCrepresenting the spearman correlation coefficient,Nsample size representing spectral vegetation index and water quality target element data,i represents the number of the current sample,R i representing the value corresponding to the current sample number of the spectral vegetation index,represents the average value of the spectral vegetation index,S i a value corresponding to the current number of the water quality target element data, < >>The average value of the water quality target element data is shown. In this embodiment, the threshold λ is preferably set to 0.8, that is, when the spearman correlation coefficient is greater than 0.8, the spectral vegetation index is a high-correlation spectral vegetation index.
Further, the spectrum vegetation indexes with the spearman correlation coefficient larger than the threshold lambda are subjected to sequencing combination, and a random forest model is established by utilizing the spectrum vegetation indexes after sequencing combination and water quality target element data to obtain inversion combination of the spectrum vegetation indexes. Specifically, in this embodiment, the spectral vegetation indexes with the spearman correlation coefficient greater than 0.8 are selected for permutation and combination, and further, as an example, the permutation and combination is: NDVI+GNDVI, NDVI+GNDVI+LCI, GNDVI+LCI are used as inputs of a random forest model, the random forest model adopts a program in an existing Scikit-Learn library, only the parameter of the number of trees (n_evators) is regulated, and the performance of the random forest model is compared according to the determination coefficient R2, the root mean square error RMSE and the relative percentage difference RPD of the random forest model. Therefore, in this embodiment, spectral vegetation index combinations corresponding to the best random forest model in the four areas of the water surface, the aquatic weed, the moss floater and the sundries are selected and respectively marked as A1 group, a B1 group, a C1 group and a D1 group, and are used for constructing inversion combinations of the spectral vegetation indexes. It should be noted that the random forest model used in this embodiment is the prior art.
Further, as shown in fig. 6, different regional information features in the inversion combination of the spectral vegetation indexes are extracted, the different regional information features are fused to obtain a multi-scale network architecture, the multi-scale network architecture is optimized, and a water quality target element inversion model is constructed through the optimized multi-scale network architecture. In the embodiment, through inversion combination of four areas of water surface, aquatic weed, moss floating objects and sundries, different area information features are extracted by using a neural architecture searching method, the area ratio occupied by the four areas is used as a weight coefficient to fuse the different area information features, and a fusion result is encoded to obtain a multi-scale network architecture. In the embodiment, the optimal spectral vegetation index combination in the A1 group, the B1 group, the C1 group and the D1 group is used as the input of inversion combination, and a water quality target element inversion model is constructed based on a neural architecture searching method. The method comprises the steps of respectively constructing feature extraction networks in four areas of water surface, aquatic weed, moss floaters and sundries by using a neural architecture searching method, wherein the neural architecture searching method adopts an existing open-source neural architecture searching framework Auto-Keras, search time (max_three) parameters and iteration time (epochs) are adjusted in the training process, in the searching process, the existing R2 and RMSE are used as evaluation indexes, the feature extraction networks of the four areas are obtained after searching is completed, the area occupation ratio of the areas is used as a weight coefficient, and the networks of the four areas are fused to obtain a multi-scale network architecture. And further optimizing the multi-scale network architecture by utilizing knowledge migration and multi-objective optimization, and searching to obtain the optimal multi-scale network architecture serving as an optimal water quality objective element inversion model. Specifically, the fused multi-scale network architecture is used as a parent network Pt, performance ordering is carried out on the trained parent network Pt, the superior parent network Pt is inherited and mutated to be used as a child network Ft, knowledge of the parent network Pt is migrated into the child network Ft, performance ordering is carried out on the child network Ft, and finally individual selection is carried out on the parent network Pt and the child network Ft through a multi-objective fitness function, so that an optimal water quality target element inversion model is obtained. It should be noted that, in this embodiment, the multi-objective fitness function used is the prior art.
And finally, training the water quality target element inversion model through training set data to obtain a trained water quality target element inversion model, inputting verification set data into the trained water quality target element inversion model for verification, and outputting a water quality monitoring result. In the embodiment, the output of the water quality target element inversion model is five water quality target elements of PH value, suspended matter detection, ammonia nitrogen amount, dissolved oxygen concentration and turbidity to be monitored.
Therefore, according to the three-dimensional water quality monitoring method provided by the embodiment, the water discrete feature extraction module is used for adaptively capturing medium-small area discrete distribution water body information, the water space distribution sensing module is used for fully sensing large-area continuous water body semantics at a space distribution angle, the water feature aggregation module is constructed for establishing a dependency relationship between key semantic information obtained by the medium-small area discrete distribution water body information and the large-area continuous water body semantics, the water feature aggregation module is used for accurately dividing the water quality image by the semantic segmentation model based on the dependency relationship, so that information features of different areas are obtained, errors caused by similar reflectivity can be effectively avoided, and sudden water quality problems can be timely found and targeted. Meanwhile, the inversion combination of the spectrum vegetation index is obtained through the spectrum vegetation index and the water quality target element data, the water quality target element inversion model is constructed through the inversion combination of the spectrum vegetation index, and the water quality can be continuously monitored in large area in real time by utilizing the water quality target element inversion model, so that the water quality space distribution condition can be effectively reflected. Compared with the traditional manual water quality sampling method for monitoring water quality, the three-dimensional water quality monitoring method provided by the embodiment is low in cost, wide in monitoring range, and high in practicality and wide in applicability.
The embodiment also provides a three-dimensional water quality monitoring system, which comprises an acquisition module, an input module, a model construction module and a monitoring module. The acquisition module is used for acquiring water quality sample data, processing the water quality sample data to obtain water quality target element data, and dividing the water quality target element data into training set data and verification set data. The input module is used for acquiring a water quality image data set and inputting a water quality image. The model construction module is used for carrying out region segmentation on the water quality image through the semantic segmentation model to obtain different region information, calculating the spectral vegetation index of each region information, obtaining inversion combination of the spectral vegetation index by utilizing the spectral vegetation index and the water quality target element data, and constructing a water quality target element inversion model through the inversion combination of the spectral vegetation index. The monitoring module trains the water quality target element inversion model through the training set data to obtain a trained water quality target element inversion model; inputting the verification set data into the trained water quality target element inversion model for verification, and outputting a water quality monitoring result.
The present embodiment also provides a computer-readable storage medium. A computer readable storage medium having stored thereon a computer program which, when executed by a processor, performs a method of three-dimensional water quality monitoring as described in the present embodiment. Wherein a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device; program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
The foregoing has been described schematically the invention and embodiments thereof, which are not limiting, but are capable of other specific forms of implementing the invention without departing from its spirit or essential characteristics. The drawings are also intended to depict only one embodiment of the invention, and therefore the actual construction is not intended to limit the claims, any reference number in the claims not being intended to limit the claims. Therefore, if one of ordinary skill in the art is informed by this disclosure, a structural manner and an embodiment similar to the technical scheme are not creatively designed without departing from the gist of the present invention. In addition, the word "comprising" does not exclude other elements or steps, and the word "a" or "an" preceding an element does not exclude the inclusion of a plurality of such elements. The various elements recited in the product claims may also be embodied in software or hardware. The terms first, second, etc. are used to denote a name, but not any particular order.
Claims (4)
1. A three-dimensional water quality monitoring method comprises the following steps:
collecting water quality sample data, processing the water quality sample data to obtain water quality target element data, and dividing the water quality target element data into training set data and verification set data;
Acquiring a water quality image data set and inputting a water quality image;
the method comprises the steps of carrying out region segmentation on a water quality image through a semantic segmentation model, wherein the region segmentation is carried out on water surfaces, waterweeds, moss floaters and sundries in the water quality image to obtain different region information; the semantic segmentation model comprises a water discrete feature extraction module, a water space distribution sensing module and a water feature aggregation module; the water discrete feature extraction module and the water space distribution perception module respectively acquire semantic information of the water quality image, and a calculation formula for acquiring key semantic information of the water quality image through the water discrete feature extraction module is as follows:
wherein,represents the discrete semantic information of the water body,MSArepresenting a multi-headed self-attention operation,α 1 、α 2 、α 3 all representing different high-dimensional mapping matrices, < >>Representing the input of the water discrete feature extraction module,xrepresenting the abscissa of the discrete feature anchor points,yrepresenting the ordinate of the discrete feature anchor point,Krepresenting the number of discrete feature anchor points,L K representing the optimized discrete feature anchor points,representing the sampling operation of the optimized discrete feature anchor points,θan adaptive offset sub-network is shown,Qrepresenting a high-dimensional space,F Q representing a mapped matrix via high dimensions α 1 Mapping to obtain the primary discrete features of the water body; the calculation formula for acquiring the semantic information of the water quality image through the water space distribution perception module is as follows:
wherein,representing the spatial distribution semantic information of the water body,Y R representing the extraction result of the water space distribution sensing module on the horizontal axis component,Y C representing the extraction result of the spatial distribution sensing module of the water body on the longitudinal axis component, the back-up represents element-by-element addition,MSArepresenting a multi-headed self-attention operation,C 1 、C 2 、C 3 all representing different convolution operations, +.>The output of the water space distribution sensing module is expressed as the initial row vector of the water divided by the input triples>The output of the water distribution sensing module is expressed as the initial column vector of the water divided by the input triplets>Representing a learnable offset superimposed on the initial row vector,representing a learnable offset variable superimposed on the initial column vector; the water body characteristic aggregation module processes semantic information acquired by the water body discrete characteristic extraction module and the water body space distribution sensing module to obtain different area information in the water quality image, and the water body characteristic aggregation module couples the semantic information acquired by the water body discrete characteristic extraction module and the water body space distribution sensing module The calculation formulas for obtaining different area information in the water quality image by combining and decoupling are as follows:
wherein,representing the output of the water body characteristic aggregation module to the water body discrete characteristic extraction module,/for>The output of the water body characteristic aggregation module to the water body space distribution sensing module is represented,ithe natural number is represented by a number of natural numbers,O MLP a multi-layer perceptron is shown,softmaxrepresenting normalization operations->Represent the firstiThe output of the discrete water feature extraction module, +.>Represent the firstiOutput of the water space distribution sensing module +.>、/>、/>、/>、/>、/>Each representing six different mapping matrices,Hindicating the operation of the transpose,Q1、Q2、K1、K2、V1、V2respectively representing six different high-dimensional spaces corresponding to six different mapping matrices,Crepresenting the number of channels entered; calculating a spectral vegetation index of each region information, and obtaining an inversion combination of the spectral vegetation index by utilizing the spectral vegetation index and water quality target element data, wherein the inversion combination comprises the following steps: the method comprises the steps of respectively extracting the ground surface true reflectivity in four areas of the water surface, the waterweed, the moss floating matters and the sundries, calculating the spectral vegetation indexes corresponding to each area, and respectively marking the spectral vegetation indexes calculated in the four areas of the water surface, the waterweed, the moss floating matters and the sundries as: group a spectral vegetation index, group B spectral vegetation index, group C spectral vegetation index, group D spectral vegetation index;
Carrying out spearman correlation analysis on the spectral vegetation indexes in different areas and the water quality target element data respectively to obtain spectral vegetation indexes with higher correlation with the water quality target element data; calculating a spearman correlation coefficient of the spectral vegetation index and the water quality target element data, setting a threshold lambda, and selecting the spectral vegetation index with the spearman correlation coefficient larger than the threshold lambda, wherein the calculation formula of the spearman correlation coefficient of the spectral vegetation index and the water quality target element data is as follows:
wherein,SRCrepresenting the spearman correlation coefficient,Nsample size representing spectral vegetation index and water quality target element data,iindicating the number of the current sample,R i representing the value corresponding to the current sample number of the spectral vegetation index,represents the average value of the spectral vegetation index,S i a value corresponding to the current number of the water quality target element data, < >>An average value representing water quality target element data;
the spectrum vegetation indexes with the spearman correlation coefficient larger than the threshold lambda are ordered and combined, a random forest model is established by utilizing the spectrum vegetation indexes after the ordered combination and water quality target element data to obtain inversion combinations of the spectrum vegetation indexes, specifically, the spectrum vegetation index combinations corresponding to the best random forest model in four areas of water surface, waterweed, moss floating matters and sundries are selected and respectively marked as A1 group, B1 group, C1 group and D1 group, and are used for constructing inversion combinations of the spectrum vegetation indexes;
Constructing a water quality target element inversion model through inversion combination of spectrum vegetation indexes, which comprises the following steps: different regional information features in inversion combinations of the spectrum vegetation indexes are extracted, the different regional information features are fused to obtain a multi-scale network architecture, the multi-scale network architecture is optimized, and a water quality target element inversion model is constructed through the optimized multi-scale network architecture;
and training the water quality target element inversion model through training set data to obtain a trained water quality target element inversion model, inputting verification set data into the trained water quality target element inversion model for verification, and outputting a water quality monitoring result.
2. The method for three-dimensional water quality monitoring according to claim 1, wherein different regional information features in inversion combination of spectral vegetation indexes are extracted, the different regional information features are fused to obtain a multi-scale network architecture, the fused multi-scale architecture is used as a parent network, the trained parent network is subjected to performance sequencing and then inheritance and mutation are used as a child network, knowledge of the parent network is migrated into the child network, the child network is subjected to performance sequencing, and individual selection is performed on the parent network and the child network to obtain an optimal water quality target element inversion model.
3. A three-dimensional water quality monitoring system, comprising:
the acquisition module acquires water quality sample data, processes the water quality sample data to obtain water quality target element data, and divides the water quality target element data into training set data and verification set data;
the input module acquires a water quality image data set and inputs a water quality image;
the model construction module is used for carrying out region segmentation on the water quality image through the semantic segmentation model, and comprises the steps of carrying out region segmentation on the water surface, the float grass, the moss and the sundries in the water quality image to obtain different region information; the semantic segmentation model comprises a water discrete feature extraction module, a water space distribution sensing module and a water feature aggregation module; the water discrete feature extraction module and the water space distribution perception module respectively acquire semantic information of the water quality image, and a calculation formula for acquiring key semantic information of the water quality image through the water discrete feature extraction module is as follows:
wherein,represents the discrete semantic information of the water body,MSArepresenting a multi-headed self-attention operation,α 1 、α 2 、α 3 all representing different high-dimensional mapping matrices, < >>Representing the input of the water discrete feature extraction module,xrepresenting the abscissa of the discrete feature anchor points, yRepresenting the ordinate of the discrete feature anchor point,Krepresenting the number of discrete feature anchor points,L K representing the optimized discrete feature anchor points,representing the sampling operation of the optimized discrete feature anchor points,θan adaptive offset sub-network is shown,Qrepresenting a high-dimensional space,F Q representing a mapped matrix via high dimensionsα 1 Mapping to obtain the primary discrete features of the water body; the calculation formula for acquiring the semantic information of the water quality image through the water space distribution perception module is as follows:
wherein,representing the spatial distribution semantic information of the water body,Y R representing the extraction result of the water space distribution sensing module on the horizontal axis component,Y C representing the extraction result of the spatial distribution sensing module of the water body on the longitudinal axis component, the back-up represents element-by-element addition,MSArepresenting a multi-headed self-attention operation,C 1 、C 2 、C 3 all representing different convolution operations, +.>The output of the water space distribution sensing module is expressed as the initial row vector of the water divided by the input triples>The output of the water distribution sensing module is expressed as the initial column vector of the water divided by the input triplets>Representing a learnable offset superimposed on the initial row vector,representing a learnable offset variable superimposed on the initial column vector; the water body characteristic aggregation module processes semantic information acquired by the water body discrete characteristic extraction module and the water body space distribution sensing module to obtain different area information in a water quality image, and the water body characteristic aggregation module processes the water body discrete characteristic extraction module and the water body air space The semantic information acquired by the inter-distribution sensing module is coupled and decoupled, and a calculation formula for obtaining different area information in the water quality image is as follows:
wherein,representing the output of the water body characteristic aggregation module to the water body discrete characteristic extraction module,/for>The output of the water body characteristic aggregation module to the water body space distribution sensing module is represented,ithe natural number is represented by a number of natural numbers,O MLP a multi-layer perceptron is shown,softmaxrepresenting normalization operations->Represent the firstiThe output of the discrete water feature extraction module, +.>Represent the firstiOutput of the water space distribution sensing module +.>、/>、/>、/>、/>、/>Each representing six different mapping matrices,Hindicating the operation of the transpose,Q1、Q2、K1、K2、V1、V2respectively representing six different high-dimensional spaces corresponding to six different mapping matrices,Crepresenting the number of channels entered; calculating a spectral vegetation index of each region information, and obtaining an inversion combination of the spectral vegetation index by utilizing the spectral vegetation index and water quality target element data, wherein the inversion combination comprises the following steps: the method comprises the steps of respectively extracting the ground surface true reflectivity in four areas of the water surface, the waterweed, the moss floating matters and the sundries, calculating the spectral vegetation indexes corresponding to each area, and respectively marking the spectral vegetation indexes calculated in the four areas of the water surface, the waterweed, the moss floating matters and the sundries as: group a spectral vegetation index, group B spectral vegetation index, group C spectral vegetation index, group D spectral vegetation index;
Carrying out spearman correlation analysis on the spectral vegetation indexes in different areas and the water quality target element data respectively to obtain spectral vegetation indexes with higher correlation with the water quality target element data; calculating a spearman correlation coefficient of the spectral vegetation index and the water quality target element data, setting a threshold lambda, and selecting the spectral vegetation index with the spearman correlation coefficient larger than the threshold lambda, wherein the calculation formula of the spearman correlation coefficient of the spectral vegetation index and the water quality target element data is as follows:
wherein,SRCrepresenting the spearman correlation coefficient,Nsample size representing spectral vegetation index and water quality target element data,iindicating the number of the current sample,R i representing the value corresponding to the current sample number of the spectral vegetation index,represents the average value of the spectral vegetation index,S i a value corresponding to the current number of the water quality target element data, < >>An average value representing water quality target element data;
the spectrum vegetation indexes with the spearman correlation coefficient larger than the threshold lambda are ordered and combined, a random forest model is established by utilizing the spectrum vegetation indexes after the ordered combination and water quality target element data to obtain inversion combinations of the spectrum vegetation indexes, specifically, the spectrum vegetation index combinations corresponding to the best random forest model in four areas of water surface, waterweed, moss floating matters and sundries are selected and respectively marked as A1 group, B1 group, C1 group and D1 group, and are used for constructing inversion combinations of the spectrum vegetation indexes;
Constructing a water quality target element inversion model through inversion combination of spectrum vegetation indexes, which comprises the following steps: different regional information features in inversion combinations of the spectrum vegetation indexes are extracted, the different regional information features are fused to obtain a multi-scale network architecture, the multi-scale network architecture is optimized, and a water quality target element inversion model is constructed through the optimized multi-scale network architecture;
the monitoring module trains the water quality target element inversion model through the training set data to obtain a trained water quality target element inversion model, inputs the verification set data into the trained water quality target element inversion model for verification, and outputs a water quality monitoring result.
4. A computer readable storage medium having stored thereon a computer program, characterized in that the computer program when executed by a processor performs the method of any of the preceding claims 1-2.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311374660.1A CN117110217B (en) | 2023-10-23 | 2023-10-23 | Three-dimensional water quality monitoring method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311374660.1A CN117110217B (en) | 2023-10-23 | 2023-10-23 | Three-dimensional water quality monitoring method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117110217A CN117110217A (en) | 2023-11-24 |
CN117110217B true CN117110217B (en) | 2024-01-12 |
Family
ID=88809469
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311374660.1A Active CN117110217B (en) | 2023-10-23 | 2023-10-23 | Three-dimensional water quality monitoring method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117110217B (en) |
Citations (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106442422A (en) * | 2016-11-18 | 2017-02-22 | 中国水利水电科学研究院 | Method for indirect remote-sensing monitoring of water quality of macrophytic regions of macrophytic lakes |
KR20180000325A (en) * | 2017-07-03 | 2018-01-02 | (주)해동기술개발공사 | Water quality monitoring system and method based on drone |
CN108020511A (en) * | 2016-11-01 | 2018-05-11 | 中国科学院遥感与数字地球研究所 | A kind of shallow macrophytic lake water quality parameter remote-sensing monitoring method and device |
CN108764132A (en) * | 2018-05-28 | 2018-11-06 | 河海大学 | A kind of lake and marshland remote sensing images error detection method |
CN110320163A (en) * | 2019-06-26 | 2019-10-11 | 深圳市宇驰检测技术股份有限公司 | Online water quality monitoring method, system and the storage medium of unmanned plane |
CN111982831A (en) * | 2020-07-01 | 2020-11-24 | 深圳市迈珂斯环保科技有限公司 | Water quality online monitoring method, device and system based on hyperspectral technology |
CN112084842A (en) * | 2020-07-28 | 2020-12-15 | 北京工业大学 | Hydrological remote sensing image target identification method based on depth semantic model |
CN112184709A (en) * | 2020-11-04 | 2021-01-05 | 黄振海 | Water body turbidity prediction method and system based on artificial intelligence |
CN114279982A (en) * | 2021-12-14 | 2022-04-05 | 北斗导航位置服务(北京)有限公司 | Water body pollution information acquisition method and device |
CN114965300A (en) * | 2022-05-23 | 2022-08-30 | 中国科学院东北地理与农业生态研究所 | Lake turbidity mapping method for constructing BP-TURB based on optical water body type and BP neural network algorithm |
CN115240024A (en) * | 2022-06-16 | 2022-10-25 | 北京大学 | Method and system for segmenting extraterrestrial pictures by combining self-supervised learning and semi-supervised learning |
CN115266719A (en) * | 2022-07-29 | 2022-11-01 | 北京百度网讯科技有限公司 | Water quality monitoring method and device based on artificial intelligence |
CN115333678A (en) * | 2022-10-17 | 2022-11-11 | 江西省绿时代低碳环保发展有限公司 | Unmanned ship water quality monitoring data transmission method and system |
WO2023000159A1 (en) * | 2021-07-20 | 2023-01-26 | 海南长光卫星信息技术有限公司 | Semi-supervised classification method, apparatus and device for high-resolution remote sensing image, and medium |
TW202316101A (en) * | 2021-09-21 | 2023-04-16 | 日商東芝數字解決方案股份有限公司 | Water quality monitoring system |
CN116012726A (en) * | 2022-12-29 | 2023-04-25 | 中科宇图科技股份有限公司 | Deep learning-based black and odorous water body remote sensing monitoring and grading method |
CN116665057A (en) * | 2023-06-05 | 2023-08-29 | 中核大地勘察设计有限公司 | River channel water quality monitoring method and system based on image processing |
CN116699096A (en) * | 2023-08-08 | 2023-09-05 | 凯德技术长沙股份有限公司 | Water quality detection method and system based on deep learning |
-
2023
- 2023-10-23 CN CN202311374660.1A patent/CN117110217B/en active Active
Patent Citations (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108020511A (en) * | 2016-11-01 | 2018-05-11 | 中国科学院遥感与数字地球研究所 | A kind of shallow macrophytic lake water quality parameter remote-sensing monitoring method and device |
CN106442422A (en) * | 2016-11-18 | 2017-02-22 | 中国水利水电科学研究院 | Method for indirect remote-sensing monitoring of water quality of macrophytic regions of macrophytic lakes |
KR20180000325A (en) * | 2017-07-03 | 2018-01-02 | (주)해동기술개발공사 | Water quality monitoring system and method based on drone |
CN108764132A (en) * | 2018-05-28 | 2018-11-06 | 河海大学 | A kind of lake and marshland remote sensing images error detection method |
CN110320163A (en) * | 2019-06-26 | 2019-10-11 | 深圳市宇驰检测技术股份有限公司 | Online water quality monitoring method, system and the storage medium of unmanned plane |
CN111982831A (en) * | 2020-07-01 | 2020-11-24 | 深圳市迈珂斯环保科技有限公司 | Water quality online monitoring method, device and system based on hyperspectral technology |
CN112084842A (en) * | 2020-07-28 | 2020-12-15 | 北京工业大学 | Hydrological remote sensing image target identification method based on depth semantic model |
CN112184709A (en) * | 2020-11-04 | 2021-01-05 | 黄振海 | Water body turbidity prediction method and system based on artificial intelligence |
WO2023000159A1 (en) * | 2021-07-20 | 2023-01-26 | 海南长光卫星信息技术有限公司 | Semi-supervised classification method, apparatus and device for high-resolution remote sensing image, and medium |
TW202316101A (en) * | 2021-09-21 | 2023-04-16 | 日商東芝數字解決方案股份有限公司 | Water quality monitoring system |
CN114279982A (en) * | 2021-12-14 | 2022-04-05 | 北斗导航位置服务(北京)有限公司 | Water body pollution information acquisition method and device |
CN114965300A (en) * | 2022-05-23 | 2022-08-30 | 中国科学院东北地理与农业生态研究所 | Lake turbidity mapping method for constructing BP-TURB based on optical water body type and BP neural network algorithm |
CN115240024A (en) * | 2022-06-16 | 2022-10-25 | 北京大学 | Method and system for segmenting extraterrestrial pictures by combining self-supervised learning and semi-supervised learning |
CN115266719A (en) * | 2022-07-29 | 2022-11-01 | 北京百度网讯科技有限公司 | Water quality monitoring method and device based on artificial intelligence |
CN115333678A (en) * | 2022-10-17 | 2022-11-11 | 江西省绿时代低碳环保发展有限公司 | Unmanned ship water quality monitoring data transmission method and system |
CN116012726A (en) * | 2022-12-29 | 2023-04-25 | 中科宇图科技股份有限公司 | Deep learning-based black and odorous water body remote sensing monitoring and grading method |
CN116665057A (en) * | 2023-06-05 | 2023-08-29 | 中核大地勘察设计有限公司 | River channel water quality monitoring method and system based on image processing |
CN116699096A (en) * | 2023-08-08 | 2023-09-05 | 凯德技术长沙股份有限公司 | Water quality detection method and system based on deep learning |
Non-Patent Citations (5)
Title |
---|
ColorWater: A Diverse Dataset and Benchmark for Semantic Water Surface Understanding;Cuixiao Liang等;《2022 26th International Conference on Pattern Recognition》;第3743-3749页 * |
River water quality estimation based on convolutional neural network.《Proceedings, APSIPA Annual Summit and Conference 2018》.2018,第1305-1308页. * |
基于改进U-Net网络的高分遥感影像水体提取;何红术等;地球信息科学学报(10);第94-106页 * |
改进U-Net网络的遥感图像水质分割算法;赵晨曦等;《遥感信息》;第137-143页 * |
葡萄多模态目标检测和语义分割数据集;陈文骏等;《中国科学数据》;第1-16页 * |
Also Published As
Publication number | Publication date |
---|---|
CN117110217A (en) | 2023-11-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113486846B (en) | Method for detecting infected object from large-field-of-view image and non-transitory computer-readable storage medium for executing same | |
Li et al. | Automatic organ-level point cloud segmentation of maize shoots by integrating high-throughput data acquisition and deep learning | |
Avtar et al. | Unmanned aerial vehicle: Applications in agriculture and environment | |
Li et al. | A comparison of deep learning methods for airborne lidar point clouds classification | |
CN115860269B (en) | Crop yield prediction method based on triple attention mechanism | |
CN111222539A (en) | Method for optimizing and expanding supervision classification samples based on multi-source multi-temporal remote sensing image | |
CN117409339A (en) | Unmanned aerial vehicle crop state visual identification method for air-ground coordination | |
CN112881294A (en) | Unmanned aerial vehicle-based mangrove forest stand health degree evaluation method | |
Biglia et al. | 3D point cloud density-based segmentation for vine rows detection and localisation | |
Guo et al. | Inversion of maize leaf area index from UAV hyperspectral and multispectral imagery | |
CN110986884A (en) | Unmanned aerial vehicle-based aerial survey data preprocessing and vegetation rapid identification method | |
Bhadra et al. | End-to-end 3D CNN for plot-scale soybean yield prediction using multitemporal UAV-based RGB images | |
CN117110217B (en) | Three-dimensional water quality monitoring method and system | |
CN110909821B (en) | Method for carrying out high-space-time resolution vegetation index data fusion based on crop reference curve | |
Azhar et al. | A framework for multiscale intertidal sandflat mapping: A case study in the Whangateau estuary | |
Raptis et al. | Multimodal data collection system for UAV-based precision agriculture applications | |
CN113705454A (en) | Method for extracting forest land containing infrared spectrum remote sensing image | |
Nazeri | Evaluation of Multi-Platform LiDAR-Based Leaf Area Index Estimates Over Row Crops | |
Knoll et al. | The German Vision of Industry 4.0 Applied in Organic Farming | |
Feng | Quantifying the effect of environments on crop emergence, development and yield using sensing and deep learning techniques | |
CN116052141B (en) | Crop growth period identification method, device, equipment and medium | |
Pongpiyapaiboon et al. | Development of a digital phenotyping system using 3D model reconstruction for zoysiagrass | |
Pal et al. | Synthesizing Forestry Images Conditioned on Plant Phenotype Using a Generative Adversarial Network | |
CN114581348B (en) | Image fusion method based on plant community behaviors | |
CN114648500B (en) | Crop weed detection method and device based on unmanned aerial vehicle and deep learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |