CN112231983B - Brain-like cross-modal identification and parallel processing method for pollution precursor emission source of remote sensing space-time big data - Google Patents
Brain-like cross-modal identification and parallel processing method for pollution precursor emission source of remote sensing space-time big data Download PDFInfo
- Publication number
- CN112231983B CN112231983B CN202011146347.9A CN202011146347A CN112231983B CN 112231983 B CN112231983 B CN 112231983B CN 202011146347 A CN202011146347 A CN 202011146347A CN 112231983 B CN112231983 B CN 112231983B
- Authority
- CN
- China
- Prior art keywords
- espp
- modal
- cross
- remote sensing
- image
- 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
- 239000002243 precursor Substances 0.000 title claims abstract description 40
- 238000003672 processing method Methods 0.000 title claims abstract description 10
- 238000000034 method Methods 0.000 claims abstract description 83
- 238000004364 calculation method Methods 0.000 claims abstract description 38
- 238000000605 extraction Methods 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims abstract description 13
- 210000004556 brain Anatomy 0.000 claims abstract description 11
- 238000012545 processing Methods 0.000 claims abstract description 11
- 230000019771 cognition Effects 0.000 claims abstract description 7
- 239000005436 troposphere Substances 0.000 claims abstract description 7
- 238000013135 deep learning Methods 0.000 claims abstract description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 38
- 239000013618 particulate matter Substances 0.000 claims description 25
- 230000008569 process Effects 0.000 claims description 20
- 230000001149 cognitive effect Effects 0.000 claims description 19
- 239000003344 environmental pollutant Substances 0.000 claims description 14
- 231100000719 pollutant Toxicity 0.000 claims description 14
- 230000004913 activation Effects 0.000 claims description 13
- 230000002787 reinforcement Effects 0.000 claims description 13
- 230000001133 acceleration Effects 0.000 claims description 10
- 230000009471 action Effects 0.000 claims description 9
- 239000000356 contaminant Substances 0.000 claims description 9
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 230000007246 mechanism Effects 0.000 claims description 6
- 238000004929 transmission Raman spectroscopy Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 5
- 238000001228 spectrum Methods 0.000 claims description 5
- 239000013256 coordination polymer Substances 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 230000000007 visual effect Effects 0.000 claims description 4
- 238000004847 absorption spectroscopy Methods 0.000 claims description 3
- 230000001413 cellular effect Effects 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000000354 decomposition reaction Methods 0.000 claims description 3
- 230000000873 masking effect Effects 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 229910052757 nitrogen Inorganic materials 0.000 claims description 3
- 230000010332 selective attention Effects 0.000 claims description 3
- XWPCYYOZOJKYKQ-UHFFFAOYSA-N 1-(2-chloroethyl)-3-[2-[2-[[2-chloroethyl(nitroso)carbamoyl]amino]ethyldisulfanyl]ethyl]-1-nitrosourea Chemical compound ClCCN(N=O)C(=O)NCCSSCCNC(=O)N(N=O)CCCl XWPCYYOZOJKYKQ-UHFFFAOYSA-N 0.000 claims 5
- 230000009897 systematic effect Effects 0.000 abstract description 3
- 239000000443 aerosol Substances 0.000 description 16
- 238000010521 absorption reaction Methods 0.000 description 15
- MWUXSHHQAYIFBG-UHFFFAOYSA-N nitrogen oxide Inorganic materials O=[N] MWUXSHHQAYIFBG-UHFFFAOYSA-N 0.000 description 15
- RAHZWNYVWXNFOC-UHFFFAOYSA-N Sulphur dioxide Chemical group O=S=O RAHZWNYVWXNFOC-UHFFFAOYSA-N 0.000 description 14
- 239000007789 gas Substances 0.000 description 10
- 238000002310 reflectometry Methods 0.000 description 10
- 230000003287 optical effect Effects 0.000 description 8
- 238000001658 differential optical absorption spectrophotometry Methods 0.000 description 7
- 230000008033 biological extinction Effects 0.000 description 6
- 239000002245 particle Substances 0.000 description 6
- 238000012937 correction Methods 0.000 description 5
- 238000012544 monitoring process Methods 0.000 description 5
- 230000005855 radiation Effects 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 230000006399 behavior Effects 0.000 description 4
- 241000894007 species Species 0.000 description 4
- 239000000126 substance Substances 0.000 description 4
- 238000012795 verification Methods 0.000 description 4
- 238000000862 absorption spectrum Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 238000009792 diffusion process Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000007613 environmental effect Effects 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- QGZKDVFQNNGYKY-UHFFFAOYSA-N Ammonia Chemical compound N QGZKDVFQNNGYKY-UHFFFAOYSA-N 0.000 description 2
- 238000003915 air pollution Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 239000002689 soil Substances 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 239000012855 volatile organic compound Substances 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 238000003911 water pollution Methods 0.000 description 2
- MGWGWNFMUOTEHG-UHFFFAOYSA-N 4-(3,5-dimethylphenyl)-1,3-thiazol-2-amine Chemical compound CC1=CC(C)=CC(C=2N=C(N)SC=2)=C1 MGWGWNFMUOTEHG-UHFFFAOYSA-N 0.000 description 1
- 241000254158 Lampyridae Species 0.000 description 1
- 241000131894 Lampyris noctiluca Species 0.000 description 1
- 240000007594 Oryza sativa Species 0.000 description 1
- 235000007164 Oryza sativa Nutrition 0.000 description 1
- CBENFWSGALASAD-UHFFFAOYSA-N Ozone Chemical compound [O-][O+]=O CBENFWSGALASAD-UHFFFAOYSA-N 0.000 description 1
- 239000000809 air pollutant Substances 0.000 description 1
- 231100001243 air pollutant Toxicity 0.000 description 1
- 229910021529 ammonia Inorganic materials 0.000 description 1
- 239000005427 atmospheric aerosol Substances 0.000 description 1
- 230000003935 attention Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000031018 biological processes and functions Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000000653 nervous system Anatomy 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- JCXJVPUVTGWSNB-UHFFFAOYSA-N nitrogen dioxide Inorganic materials O=[N]=O JCXJVPUVTGWSNB-UHFFFAOYSA-N 0.000 description 1
- 238000006552 photochemical reaction Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 235000009566 rice Nutrition 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 238000003900 soil pollution Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000002211 ultraviolet spectrum Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention provides a brain-like cross-modal identification and parallel brain-like intelligent processing method for a pollution precursor Emission Source (ESPP) of remote sensing space-time big data. The method specifically comprises the following steps: respectively inverting NO based on satellite remote sensing inversion 2 、O 3 、SO 2 And NH 3 The concentration of the troposphere column of the atmospheric pollution trace gas is equal, and PM near the ground is calculated at the same time 2.5 Concentration; target semantic tag C based on satellite remote sensing inversion result, hyperspectral image, synthetic aperture radar image and ESPP p Training based on brain heuristic calculation, and establishing a cross-modal neuro-cognitive calculation model facing ESPP; deep learning based extraction of cross-modal, multi-level and multi-scale ESPP features in remote sensing space-time big data; object-oriented remote sensing cross-modal ESPP representation and reasoning method based on probability cognition framework is used for carrying out multi-scale topic clustering, cross-modal ESPP cognition processing and ESPP recognition incremental learning on ESPP features. The invention can effectively realize ESPP identification based on brain-like calculation and provide a systematic solution for tracing the atmospheric pollution.
Description
Technical Field
The invention relates to the field of atmosphere remote sensing and environmental pollution, in particular to a brain-like cross-modal identification method for a pollutant precursor emission source facing remote sensing space-time big data and a parallel processing method thereof.
Background
The good atmospheric environment is the natural environment foundation for realizing economic development. However, air pollution weather such as sand storm and haze is frequent in a global large scale, and the human health and the ecological balance of the earth are seriously affected. The main cause of the formation of air-polluted weather such as haze is sulfur dioxide (SO) 2 ) Nitrogen Oxides (NO) x ) (mainly comprising nitric oxide NO and nitrogen dioxide NO 2 ) Ammonia (NH) 3 ) Ozone (O) 3 ) And Volatile Organic Compounds (VOCs) s ) Secondary aerosol of solid particles is generated by photochemical reaction of the primary pollutants, and the solid particles are further aggregated to form PM in a moist and steady weather environment 2.5 And (5) pollution. The secondary aerosols include Secondary Organic Aerosols (SOA) and Secondary Inorganic Aerosols (SIA).
Different from the verification and identification of pollution sources such as water body, soil and the like, the pollutant has atmospheric diffusion, high-altitude vertical transportation and long-distance transmission due to the fluid mechanical properties such as convection and diffusion of the atmosphere, and the discharge process of the air pollutant is generally difficult to verify in real time on site. Under the drive of economic benefits, some enterprises have illegal actions such as theft and discharge, and often, it is difficult to acquire on-site evidence of real-time discharge actions and discharge amounts of pollutants. For soil pollution tracing and water pollution tracing with relatively fixed pollution source positions, the pollution positions are easy to find. Because of the space-time flow characteristics of the atmosphere, the positioning and identification of the atmospheric pollution sources are very difficult, and the atmospheric environment management often lacks scientific basis for law enforcement. The pollution tracing is generally carried out by compiling a discharge source list, and the reconstruction and updating of the corresponding precursor discharge list, the selection of a source analysis technical method and analysis and evaluation of analysis results are very complex system engineering. The traditional method generally adopts a quantum behavior particle swarm optimization algorithm (Tian Na and the like, application of a quantum particle swarm with a disturbance operator in water pollution source identification, system simulation academy 2015), a firefly swarm algorithm (Chen and the like, A New Air Pollution Source Identification Method Based on Remotely Sensed Aerosol and Improved Glowworm Swarm optimization, IEEE Journal of Selected Topics in Applied Earth Observations and Remote sensing, 2017), inversion, numerical simulation (Yang Yifan and the like, numerical simulation of the inversion problem of sudden atmospheric pollution source position identification, environmental science, academy 2013) and the like, and the identification efficiency is relatively low, so that the urgent need of scientific and effective pollution source identification is present.
Disclosure of Invention
In order to effectively identify a pollutant precursor emission source, the invention provides a brain-like cross-mode identification method for the pollutant precursor emission source facing remote sensing space-time big data and a parallel processing method thereof.
The invention provides a brain-like cross-modal identification method for a pollution precursor emission source of remote sensing space-time big data, which comprises the following steps:
step 1: atmospheric pollution precursor troposphere column concentration and PM (particulate matter) based on satellite remote sensing 2.5 Inversion method for respectively inverting atmospheric NO 2 、O 3 、SO 2 、NH 3 And the tropospheric column concentration of the AOD while inverting and calculating the near-surface PM 2.5 Concentration;
step 2: according to NO 2 、O 3 、SO 2 、NH 3 And AOD tropospheric column concentration, PM 2.5 Target semantic tags C of concentration, hyperspectral image, synthetic aperture radar image and ESPP p Training based on brain heuristic calculation to form an ESPP-oriented cross-modal neuro-cognitive calculation model; wherein ESPP refers to a source of contaminant precursor emissions; the saidESPP-oriented cross-modal neurocognitive computing model is marked as a CNCC model;
step 3: the deep learning-based extraction method for the cross-modal, multi-level and multi-scale ESPP features in the remote sensing space-time big data specifically comprises the following steps: performing significance calculation on the remote sensing image by using the concentration of the atmospheric pollution precursor, extracting the cross-modal characteristics of ESPP by using a multi-level and multi-scale characteristic extraction algorithm, and performing reinforcement learning on the cross-modal characteristics of ESPP;
step 4: object-oriented remote sensing cross-modal ESPP representation and reasoning method based on probability cognition framework is used for carrying out multi-scale topic clustering, cross-modal ESPP cognition processing and ESPP recognition incremental learning on ESPP features.
Further, in step 1, the atmospheric NO is converted by using a differential absorption spectrum algorithm 2 And NH 3 Is used for generating O by adopting a band residual error algorithm 3 And SO 2 Is a tropospheric column concentration; using AOD-PM 2.5 Tropospheric column concentration and near-surface PM for AOD inverse of hybrid correlation model 2.5 Concentration.
Further, in step 2, the process of training the CNCC model based on brain heuristic calculation is:
CNCC:{RS,VCD X }→C P
wherein C is P Is a target semantic tag for training a CNCC model;
in step 2, the method further comprises a CNCC model identification process based on brain-like calculation, specifically comprising the following steps:
C' P ＝CNCC(RS,VCD X )
wherein C' P Is a target semantic label identified by adopting a CNCC model, RS represents an available remote sensing image,VS denotes a visible light image, SAR denotes a synthetic aperture radar image, UV denotes an ultraviolet image, IR denotes an infrared image, HS denotes a hyperspectral image, VCD X Representing a concentration profile of trace gas species.
Further, in step 3:
the method for performing significance calculation on the remote sensing image by using the concentration of the atmospheric pollution precursor specifically comprises the following steps:
according to a visual selective attention mechanism and a spectrum residual error method, a significance calculation method of a remote sensing image is designed by utilizing remote sensing big data and troposphere column concentration distribution inverted by an atmospheric pollution precursor, and the method specifically comprises the following steps: using saliency maps SC RS And the original remote sensing image I RS Masking to obtain a significant region image Si in which ESPP may be present RS The method comprises the following steps:
wherein RS represents an available remote sensing image, VS represents a visible light image, SAR represents a synthetic aperture radar image, UV represents an ultraviolet image, IR represents an infrared image, HS represents a hyperspectral image, and VCD X Representing a concentration profile of trace gas species;
the adoption of a multi-level and multi-scale feature extraction algorithm to extract the cross-modal features of ESPP comprises the following specific steps:
ESPP activation feature Fi is extracted based on target feature alliance model and deep neural network RS The method comprises the following steps:
wherein, TFSM represents a target feature subjective model, DNN represents a deep neural network;
the ESPP features are subjected to reinforcement learning, and specifically:
activation feature Fi for sensing neuron N of DNN RS Extracting by adjusting the network parameters and weights in the action space ACT in action aNovel activation feature Fi RS ' the reward for obtaining ESPP semantics is r; the ESPP characteristic reinforcement learning is realized through a Markov decision process, namely:
wherein Fi is RS Belonging to the state space S, a epsilon ACT, r epsilon RW, where RW is the return function, and MDP represents the Markov decision process.
Further, in step 4:
the multi-scale theme clustering is carried out on ESPP features, and specifically comprises the following steps:
multi-scale topic clustering is carried out by adopting a hierarchical Dirichlet topic model, so that the activation characteristic Fi of ESPP is realized RS Attribute semantics Tf to ESPP RS Is a conversion of (1), namely:
wherein HLDA represents a hierarchical Dirichlet topic model; RS represents a usable remote sensing image, VS represents a visible light image, SAR represents a synthetic aperture radar image, UV represents an ultraviolet image, IR represents an infrared image, HS represents a hyperspectral image, VCD X Representing a concentration profile of atmospheric trace species;
constructing a cross-modal knowledge graph CKG of a typical ESPP, namely:
wherein TS and CP are the type and confidence probability of ESPP respectively;
the cross-modal ESPP cognitive processing is performed on ESPP characteristics, and specifically comprises the following steps:
ESPP-based spatio-temporal context C TS And a cross-modal knowledge graph CKG, and establishing attribute semantics Tf of ESPP according to the mobile intelligent body, the cellular automaton and the cognitive computing theory RS And type TSCTC is mapped non-linearly, i.e.:
the incremental learning for ESPP identification on ESPP features is specifically as follows:
when ESPP modal information DeltaTf is added RS When the method is used, under the condition of keeping the cross-modal knowledge graph CKG of the original ESPP, CKG attributes are directly learned and updated without reconstruction, and the latest cross-modal knowledge graph CKG' is obtained through incremental learning and mapping TIL, namely:
the invention provides a parallel processing method of a brain-like cross-modal identification method of a pollution precursor emission source facing remote sensing space-time big data, which comprises the following steps: the whole recognition process of the brain-like cross-modal recognition method for the pollutants facing the remote sensing space-time big data is used as a BCR task, and the parallel processing method comprises the following steps:
step A1, task decomposition for ESPP parallel identification;
step A2, heterogeneous parallel acceleration for a BCR task;
a3, multi-core parallel acceleration for the BCR task;
and step A4, multi-machine parallel acceleration for the BCR task.
Further, the step A1 specifically includes:
constructing a multi-resolution pyramid by utilizing the preprocessed remote sensing image, and slicing the image layer by layer to form a data block D k Decomposing the BCR task into M workflows, each workflow consisting of N serial task TRSs i Composition, each serial task TRS i Comprising Q parallelizable subtasks TRP j The method comprises the following steps:
further, the step A2 specifically includes: for each data block D k There are multiple loops of computation inside, programmed with GPU.
Further, step A3 includes: the step of performing saliency calculation on the remote sensing image by using the concentration of the atmospheric pollution precursor in the step 3, the step of extracting the cross-modal characteristics of the ESPP by using a multi-level and multi-scale characteristic extraction algorithm, and the step of performing multi-scale topic clustering on the ESPP characteristics in the step 4 adopt an OpenMP parallel programming algorithm.
Further, step A4 includes: the step of reinforcement learning of the ESPP features in the step 3, the step of cross-modal ESPP cognitive processing of the ESPP features in the step 4, and the step of incremental learning of ESPP recognition of the ESPP features adopt a Map-Reduce programming algorithm of MPI. .
The invention has the beneficial effects that:
(1) The invention simulates a nervous system and a cognitive scientific model, and establishes a brain-like CNCC model for ESPP recognition of remote sensing space-time big data based on a deep learning and cognitive learning theory. The CNCC model adopting brain-like calculation realizes the recognition of the cross-mode and has higher ESPP recognition precision.
(2) Aiming at the cross-modal ESPP positioning, classifying and identifying technology of the haze remote sensing space-time big data, the invention designs a multi-level, multi-scale and object-oriented BCR algorithm of a pixel-object-ESPP-scene based on a brain-like CNCC model, and can effectively solve the technical bottleneck problem of BCR system application of the remote sensing big data.
(3) In order to solve the engineering application of the haze remote sensing space-time big data BCR, the invention provides a high-efficiency parallel algorithm of the BCR, and the BCR parallel processing system for the remote sensing space-time big data has very high recognition speed and can establish an application platform for monitoring the atmosphere remote sensing environment with high performance by utilizing a plurality of parallel computing mechanisms such as multiple computers, multiple cores, mixed isomerism and the like.
Drawings
FIG. 1 is one of the flow diagrams of a brain-like cross-modal identification method of a pollution precursor emission source facing remote sensing space-time big data provided by an embodiment of the invention;
FIG. 2 is a flow chart of inversion of the concentration of a tropospheric column of an atmospheric pollution precursor and PM2.5 based on satellite remote sensing according to an embodiment of the present invention;
FIG. 3 is a flow chart of extracting cross-modal, multi-level and multi-scale ESPP features in remote sensing space-time big data based on deep learning according to the embodiment of the invention;
FIG. 4 is a schematic diagram of a probabilistic cognitive framework-based object-oriented remote sensing cross-modal ESPP representation and reasoning process provided by an embodiment of the invention;
FIG. 5 is a second schematic flow chart of a brain-like cross-modal identification method of a pollution precursor emission source facing remote sensing space-time big data provided by the embodiment of the invention;
fig. 6 is a flow chart of a parallel processing method of a brain-like cross-modal identification method of a pollution precursor emission source facing remote sensing space-time big data, which is provided by the embodiment of the invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions in the embodiments of the present invention will be clearly described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Atmospheric pollution source identification generally has two-layer implications. One is the identification of equipment, objects or sites that produce the contaminants and release the contaminants to the atmosphere. And secondly, component source analysis for forming an emission list of atmospheric pollution. The identification of the emission source (EmissionSource of Pollutant Precursor, ESPP) of the contaminant precursor studied in the present invention refers to the identification of the device, object or location of the emission of atmospheric contaminants that can be observed by remote sensing of spatiotemporal big data.
Aiming at space-time change of regional atmosphere environment, various atmosphere remote sensing means such as visible light, infrared, ultraviolet, hyperspectral, microwave, laser, sound wave and the like exist at present. The multi-mode information of atmospheric pollution such as an optical image, hyperspectral fingerprint, precursor concentration, ground real sampling and the like can be obtained by comprehensively utilizing remote sensing technologies such as optics, electromagnetism and the like. The remote sensing space-time big data information is fused, and whether the monitoring target belongs to ESPP or not is effectively identified and verified, so that the method is a precondition for atmospheric environment treatment. By adopting Brain-isolated Cross-modal Recognition (BCR) technology, the physical and chemical characteristics of multisource Cross modes of an observed object can be fused, so that the space-time changes of atmospheric pollutants and ground pollution sources can be analyzed in time, and ESPP can be effectively screened.
RS-STBD: remote Sensing Spatio-Temporal Big Data, remote sensing space-time big data;
fig. 1 is a schematic flow chart of a brain-like cross-mode identification method for a pollution precursor emission source facing remote sensing space-time big data, which is provided by the embodiment of the invention. As shown in fig. 1, the method comprises the steps of:
s11, atmospheric pollution precursor troposphere column concentration and PM based on satellite remote sensing 2.5 Inversion method for respectively inverting atmospheric NO 2 、O 3 、SO 2 、NH 3 And the tropospheric column concentration of the AOD while inverting and calculating the near-surface PM 2.5 Concentration;
specifically, in this step, as shown in FIG. 2, the atmospheric NO is converted using a differential absorption spectroscopy (DOAS) algorithm 2 And NH 3 Is inverted to O using a band residual error (BRD) algorithm 3 And SO 2 Adopts AOD-PM to the tropospheric column concentration 2.5 Tropospheric column concentration and near-surface PM for AOD inverse of hybrid correlation model 2.5 Concentration.
As an embodiment, the atmospheric NO is inverted using a differential absorption spectroscopy (DOAS) algorithm 2 And NH 3 The tropospheric column concentration of (2) is specifically: by NO 2 And NH 3 In the strong absorption characteristic of ultraviolet band, differential absorption spectrum algorithm is adopted to realize atmospheric NO 2 And NH 3 Is a function of the inversion of (a). In essence DOAS inversion is an extraction process of the absorption spectrum characteristics of a particular gas.
The optical thickness τ of the optical path is the intensity I of the incident radiation according to the Beer-Lambert law 0 (lambda) and the measured outgoing radiation intensity I (lambda, sigma) ratio:
I(λ)＝I 0 (λ)e -τ(λ) (1)
DOAS consists of a portion where the atmospheric extinction changes rapidly with wavelength during radiation transmission (differential optical thickness) and a portion where the atmospheric extinction changes slowly with wavelength:
wherein sigma i ' the (lambda, T) is the differential absorption cross section of the ith absorption gas at temperature T and represents the portion that changes rapidly with wavelength. SCD (SCD) i (lambda) is the column concentration of the ith absorption gas. The atmospheric extinction effect which slowly varies with wavelength is expressed by using a low-order polynomial P (lambda),representing the portion of the extinction of the atmosphere that slowly varies with wavelength, σ Ray SCD Ray (lambda) and sigma Mie SCD Mie (lambda) represents atmospheric molecular Rayleigh scattering and aerosol rice scattering, respectively.
In a specific wave band of gas absorption, the concentration of the absorption gas column on the inclined path can be obtained according to a linear least square fitting method. Finally, contaminating precursor NO 2 And NH 3 And respectively carrying out authenticity verification on the inversion result, the OMI product and the foundation MAX-DOAS observation result.
As one implementation, the O is inverted using a Band Residual (BRD) algorithm 3 And SO 2 The tropospheric column concentration of (2) is specifically: o in the selected ultraviolet spectral range 3 And SO 2 Wave crest and wave trough combined wave crest and wave trough of gas absorption form wavelength pairs, satellite zenith observation data of the wavelength pairs are adopted to calculate residual errors, and atmosphere O is extracted maximally 3 And SO 2 Effective information for realizing atmosphere O 3 And SO 2 And (5) inversion of column concentration.
According to the extinction ratio Beer Lambert law I (λ) =i 0 (λ)e -τ(λ) The BRD calculation method is as follows:
wherein N is j Is the N value difference of the j-band pair, AMF is the geometric atmospheric quality factor, θ o Is the zenith angle of the sun, θ is the zenith angle observed by the satellite, g j Is a correction factor for AMF, a j Is O 3 The j-th pair of differential absorption coefficients, r j Is SO 2 Is the j-th pair of differential absorption coefficients, Ω 0 Is O 3 Is the actual column concentration Θ 0 Is SO 2 S (λ) is the systematic error.
Due to O in the atmosphere 3 The content is far higher than SO 2 High in SO 2 In the absorption inversion spectrum of (2), O is present 3 Is absorbed in the ultraviolet spectrum of (2), O must be considered 3 Ultraviolet absorption to SO of (C) 2 Inversion effects. Assuming that the ultraviolet radiation is absorbed entirely by O 3 Is caused by absorption of (A) by O 3 Absorption wave band, and inversion calculation to obtain O of atmospheric initial estimation 3 Column concentration. But actually inverted O 3 Column concentration is SO-containing 2 Is the integrated column concentration of (1), namely:
wherein Ω T Representing the inverted O 3 Column concentration, Ω 0 Indicating true O in the atmosphere 3 Column concentration Θ 0 Representing the true SO in the atmosphere 2 Column concentration. Then O 3 The absorption of (c) can be expressed simply as:
wherein Ω T The representation considers O only 3 Is absorbed by ultraviolet light of (2)Inversion of the obtained O 3 Column concentration, at the inversion wavelength of BRD, simulating N-pair value according to radiation transmission model, and calculating residual Res j Obtaining:
SO is then SO 2 Column concentration mean Θ mean From band pair P 1 ,P 2 And P 3 The calculation can be as follows:
finally SO is carried out 2 And carrying out authenticity verification on the column concentration inversion result and the OMI transit observation data mean value and the foundation MAX-DOAS observation result respectively.
As an embodiment, AOD-PM is used 2.5 Tropospheric column concentration and near-surface PM for AOD inverse of hybrid correlation model 2.5 The concentration is specifically as follows:
near-stratum PM 2.5 Inversion involves selecting an aerosol model and determining the surface reflectivity, constructing an aerosol thickness (AOD-PM) based on the correlation of the AOD with near-formation particles 2.5 And (5) inverting the model. The essence of AOD remote sensing inversion is the process of removing earth surface reflection noise in satellite signals and obtaining atmospheric aerosol information. The land aerosol remote sensing can realize inversion by utilizing the characteristic that the visible light wave band is more sensitive to aerosol scattering. According to the season and the feature environmental characteristics, the AOD respectively adopts a Deep Blue (DB) algorithm or a dark pixel (DT) method to invert the land aerosol product.
Under dark background conditions such as a vegetation dense agricultural planting area or a dark color soil surface, the surface reflectivity of a red-blue channel is determined by adopting a DT method through the inherent linear relation of the surface reflectivity of the blue-red-near infrared channel, a proper aerosol model is selected, and the AOD is obtained through inversion of a lookup table.
Aiming at the problem that the inversion of the DT algorithm is ineffective under the condition of high earth surface reflectivity in winter, the blue light channel earth surface reflectivity has relatively darker characteristics based on the areas of high earth surface reflectivity such as cities, deserts and the like, and the DB algorithm is adopted to invert the AOD. Assuming that the earth surface reflectivity of most ground objects is kept unchanged in a short time, selecting the lowest reflectivity period of satellite observation in a time interval, determining the earth surface reflectivity in the time interval, establishing a lookup table according to the satellite apparent reflectivity observation based on an aerosol model and an aerosol profile simulated by a chemical conveying mode, and realizing the inversion of the AOD.
Near-stratum PM 2.5 Inversion with AOD-PM 2.5 The mixed correlation model adopts vertical correction and humidity correction, considers influence factors such as seasons, meteorological conditions, regions and the like, and improves the estimation capability of the regional atmospheric particulate concentration. Consider AOD-PM 2.5 Is to build a simple linear regression model as follows:
PM 2.5 ＝α 1 +β 1 AOD (8)
for aerosol soluble SNA, the moisture absorption increases faster under the condition of higher humidity, but the particle radius increases more rapidly than the mass, and the extinction capability increases rapidly under the condition of smaller mass change, so the following method is adopted for high humidity correction:
wherein PBLH is boundary layer height, f (RH) is humidity correction factor, alpha 1 、α 2 Respectively the fixed intercept, beta 1 、β 2 Is a fixed slope. Finally, PM is observed by utilizing foundation 2.5 And carrying out authenticity verification and error evaluation on the result.
S12, constructing an ESPP-oriented cross-modal neuro-cognitive computing model based on brain heuristic computation, wherein the method specifically comprises the following steps of: according to NO 2 、O 3 、SO 2 、NH 3 And AOD tropospheric column concentration, PM 2.5 Target semantic tags C of concentration, hyperspectral image, synthetic aperture radar image and ESPP p Training based on brain heuristic calculation to form ESPP-oriented cross-modal neuro-cognitive calculation modelThe method comprises the steps of carrying out a first treatment on the surface of the Wherein ESPP refers to a source of contaminant precursor emissions;
specifically, in the invention, the ESPP-oriented Cross-modal neurocognitive computing model is simply called a CNCC (Cross-modal Neural Cognitive Computing) model. This step is actually a training process of CNCC model based on brain heuristic calculation.
It can be understood that in practical application, the construction of the CNCC model may include, in addition to the training process described above, a CNCC model test process based on brain heuristic calculation, where the CNCC model test process based on brain heuristic calculation extracts the feature target semantic tag C by using data of different modalities and a CNCC model obtained by training p I.e. the type of ESPP, thereby verifying the identification effect of ESPP.
As an embodiment, CNCC model training based on brain heuristic calculation and CNCC model recognition based on brain-like calculation are implemented by the following two processes, respectively:
wherein C is P Is a target semantic tag, C 'for training a CNCC model' P Is a target semantic tag identified using a CNCC model.
S13, extracting cross-modal, multi-level and multi-scale ESPP features in remote sensing space-time big data based on deep learning, wherein the ESPP features specifically comprise: performing significance calculation on the remote sensing image by using the concentration of the atmospheric pollution precursor, extracting the cross-modal characteristics of ESPP by using a multi-level and multi-scale characteristic extraction algorithm, and performing reinforcement learning on the ESPP characteristics;
as an implementation manner, the significance calculation is performed on the remote sensing image by using the concentration of the atmospheric pollution precursor, specifically:
method for attention mechanism and spectrum residual error according to visual selectivityTropospheric column concentration profile inversion using remote sensing big data and atmospheric contaminant precursors (e.g., NO 2 、O 3 、SO 2 And NH 3 The concentration of the equal trace gas in the troposphere column), the significance calculation method for designing the remote sensing image specifically comprises the following steps: using saliency maps SC RS And the original remote sensing image I RS Masking to obtain a significant region image Si in which ESPP may be present RS The method comprises the following steps:
wherein RS represents an available remote sensing image, VS represents a visible light image, SAR represents a synthetic aperture radar image, UV represents an ultraviolet image, IR represents an infrared image, HS represents a hyperspectral image, and VCD X Representing a concentration profile of trace gas species;
specifically, by adopting a visual selective attention mechanism and a spectrum residual error method, the calculated amount can be reduced, thereby realizing reasonable distribution of calculation resources. By using saliency map SC RS And the original remote sensing image I RS By performing mask processing, the search range of pollution sources in the high-resolution large-format remote sensing image can be reduced, and the detection and identification efficiency of ESPP can be improved.
As an implementation manner, the method for extracting the cross-modal feature of the ESPP by adopting a multi-level and multi-scale feature extraction algorithm specifically comprises the following steps:
ESPP activation feature Fi is extracted based on target feature alliance model and deep neural network RS The method comprises the following steps:
wherein, TFSM represents a target feature subjective model, DNN represents a deep neural network; as an implementation manner, the reinforcement learning on the ESPP features specifically includes:
activation feature Fi for sensing neuron N of DNN RS Through network parameters and rightsThe adjustment action a of the action space ACT is repeated, and a new activation feature Fi is extracted RS ' the reward for obtaining ESPP semantics is r; the ESPP characteristic reinforcement learning is realized through a Markov decision process, namely:
wherein Fi is RS Belonging to a state space S, a epsilon ACT, r epsilon RW, wherein RW is a return function, and MDP represents a Markov decision process;
s14, performing multi-scale theme clustering, cross-modal ESPP cognitive processing and ESPP recognition incremental learning on ESPP features by using an object-oriented remote sensing cross-modal ESPP representation and reasoning method based on a probability cognitive framework; as shown in fig. 4.
Specifically, in this step, the object-oriented remote sensing cross-modal ESPP representation and reasoning method based on the probabilistic cognitive framework includes: analyzing the group interpretation thought of the manual interpretation of the remote sensing image, researching the collaborative cognitive mechanism of semantic association, establishing a cross-modal ESPP interpretation and semantic reasoning framework for simulating the manual interpretation based on the statistical learning theory, and realizing the representation and recognition of the object-oriented cross-modal ESPP aiming at the remote sensing big data.
As an implementation manner, the multi-scale topic clustering of the ESPP features specifically includes:
multi-scale topic clustering is carried out by adopting a hierarchical Dirichlet topic model, so that the activation characteristic Fi of ESPP is realized RS Attribute semantics Tf to ESPP RS Is a conversion of (1), namely:
wherein HLDA represents a hierarchical Dirichlet topic model;
constructing a cross-modal knowledge graph CKG of a typical ESPP, namely:
wherein TS and CP are the type and confidence probability of ESPP respectively;
as an implementation mode, the embodiment of the invention researches an ESPP cognitive model of manual interpretation and semantic association to realize the cross-modal cognition and understanding of ESPP in a remote sensing scene. Release sources for generating atmospheric pollution can be divided into two major types, namely mobile pollution sources (such as automobiles) and fixed sources (such as factories), and physical, chemical and biological processes of release, migration, diffusion and degradation of the atmospheric pollution are considered, wherein the cross-modal ESPP cognitive treatment is performed on ESPP characteristics, specifically:
ESPP-based spatio-temporal context C TS And a cross-modal knowledge graph CKG, and establishing attribute semantics Tf of ESPP according to the mobile intelligent body, the cellular automaton and the cognitive computing theory RS Nonlinear mapping CTC with type TS, i.e.:
the telemetry image, while typically large data, is often sparse, small sample data of the available data at a given time-space. For this reason, it is necessary to solve the problem when ESPP modality information DeltaTf is newly added RS When the knowledge graph CKG of the original ESPP is maintained, the attribute can be directly learned and updated without reconstruction. As an implementation manner, the incremental learning of ESPP identification on ESPP features specifically includes:
when ESPP modal information DeltaTf is added RS When the method is used, under the condition of keeping the cross-modal knowledge graph CKG of the original ESPP, CKG attributes are directly learned and updated without reconstruction, and the latest cross-modal knowledge graph CKG' is obtained through incremental learning and mapping TIL, namely:
the brain-like cross-modal identification method for the pollution precursor emission source facing the remote sensing space-time big data provided by the embodiment of the invention can effectively realize ESPP identification based on brain-like calculation. The embodiment of the invention combines the multi-source cross-mode physical and chemical characteristics of ESPP, can timely analyze the space-time variation of the atmospheric pollutants and the ground pollution sources, and provides a brain-like intelligent systematic solution for monitoring the atmospheric emission sources and controlling haze.
Example 2
As shown in the figure, the embodiment of the invention provides a brain-like cross-mode identification method for a pollution precursor emission source facing remote sensing space-time big data, which comprises the following steps of:
s21, ESPP-oriented target positioning
According to BRD and DOAS algorithm, inverting the two-time-phase atmospheric satellite monitoring data, and calculating PM of the observation area under the two time phases 2.5 Concentration profile VCD PM And concentration profile VCD of contaminant precursor HP . Locating the high concentration zone HCD of the precursor within the threshold delta and gamma range HP And PM 2.5 High concentration zone VCD PM ：
Wherein S is a spatial region and T is a time interval; this procedure revealed that: HCD must be located in the spatio-temporal range for a possible ESPP HP The pollution discharge behavior occurs in the water tank.
S22, ESPP-oriented feature extraction
Remote sensing image I by deep neural network DNN RS Processing, extracting activation feature Fi RS ESPP-characterized MEF constituting multiple sources<Fi VS ,Fi SAR >. Wherein Fi is VS Fi, an optical image feature of ESPP SAR Is a SAR image feature of ESPP.
In the step, ESPP is identified according to the high-resolution optical image and the SAR image with haze penetrability, the ESPP pollution emission behavior is verified by utilizing optical atmosphere remote sensing space-time big data, and the pollution emission amount is estimated quantitatively.
S23 ESPP-oriented feature clustering
Bayesian-based cognitive computational framework, defined by MEF<Fi VS ,Fi SAR >Clustering construction ESPP cross-mode theme feature TFC, and construction ESPP knowledge graph CKG<TFC,TS,CP>. Simultaneous extraction of ESPP spatiotemporal context CTS<CT TS ,CS TS >Providing a context judgment basis for identifying ESPP emission behavior. Wherein CT is TS Is ESPP type TS time-varying information, CS TS Is the space variation information of ESPP type TS.
S24, ESPP-oriented target identification
According to a characteristic matrix TFC of a cross-modal subject of ESPP, under the space-time context CTS and knowledge graph CKG priori of ESPP, adopting statistical learning to judge the cross-modal type TS of ESPP i I.e.
And further identifying and tracking the identified ESPP type and change by using reinforcement learning and incremental learning.
S25, calculating and verifying pollution discharge amount
Inversion and estimation of pollutant emissions using aerosol optical thickness (AOD) and reference to ground-observed actual PM 2.5 And the pollution information and the foundation AOD data are compared and verified, so that a decision basis is provided for the follow-up action of the atmospheric environmental protection treatment.
Example 3
Considering the timeliness of atmospheric environment monitoring, in the embodiment of the invention, the BCR algorithm speed of ESPP is improved by adopting a mixed heterogeneous parallel strategy, and the remote sensing BCR application with high identification rate and high speed is realized. As shown in fig. 6, the specific idea of parallel BCR study is as follows:
s31 task decomposition for ESPP parallel identification
Because the task of the remote sensing image BCR has decomposability and a tensor structure of data, the preprocessed remote sensing image is utilized to construct the multi-resolution pyramid. Wherein the low resolution image is used for pollution source positioning and the high resolution image is used for pollution source identificationAnd checking. Slicing the image layer by layer to form data blocks D k . According to task properties, the BCR task can be decomposed into M workflows, each consisting of N serial task TRSs i Composition, each serial task TRS i Comprising Q parallelizable subtasks TRP j The method comprises the following steps:
s32, heterogeneous parallel acceleration facing BCR algorithm
For each data block D k Multiple-cycle calculation exists inside, GPU programming acceleration is adopted, and heterogeneous parallelism is utilized to improve the speed of a BCR algorithm.
S33, multi-core parallel acceleration for BCR algorithm
Aiming at significance calculation of remote sensing images, multi-level cross-mode ESPP feature extraction, multi-scale theme clustering of ESPP features and other algorithms with task parallelism, openMP parallel programming is adopted, and multi-core parallel is utilized to improve the speed of a BCR algorithm.
S34, multi-machine parallel acceleration facing BCR algorithm
Aiming at parallel tasks with less interactive communication, such as cross-modal ESPP cognitive calculation, ESPP characteristic reinforcement learning, ESPP recognition increment learning and the like, a Map-Reduce programming algorithm of MPI is adopted, and the speed of a BCR algorithm is improved by utilizing multiple machines in parallel.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (7)
1. The brain-like cross-modal identification method for the pollutant precursor emission source of the remote sensing space-time big data is characterized by comprising the following steps of:
step 1: atmospheric pollution precursor troposphere column concentration and PM (particulate matter) based on satellite remote sensing 2.5 Inversion method for respectively inverting atmospheric NO 2 、O 3 、SO 2 、NH 3 And the tropospheric column concentration of the AOD while inverting and calculating the near-surface PM 2.5 Concentration;
step 2: according to NO 2 、O 3 、SO 2 、NH 3 And AOD tropospheric column concentration, PM 2.5 Target semantic tags C of concentration, hyperspectral image, synthetic aperture radar image and ESPP p Training based on brain heuristic calculation to form an ESPP-oriented cross-modal neuro-cognitive calculation model; wherein ESPP refers to a source of contaminant precursor emissions; the ESPP-oriented cross-modal neurocognitive computing model is marked as a CNCC model;
in step 2, the process of training the CNCC model based on brain heuristic calculation is:
wherein C is P Is a target semantic tag for training a CNCC model;
in step 2, the method further comprises a CNCC model identification process based on brain-like calculation, specifically comprising the following steps:
wherein C' P Is a target semantic tag identified by adopting a CNCC model, RS represents an available remote sensing image, VS represents a visible light image, SAR represents a synthetic aperture radar image, UV represents an ultraviolet image, IR represents an infrared image, HS represents a hyperspectral image and VCD X Representing a concentration profile of trace gas species;
step 3: the deep learning-based extraction method for the cross-modal, multi-level and multi-scale ESPP features in the remote sensing space-time big data specifically comprises the following steps: performing significance calculation on the remote sensing image by using the concentration of the atmospheric pollution precursor, extracting the cross-modal characteristics of ESPP by using a multi-level and multi-scale characteristic extraction algorithm, and performing reinforcement learning on the cross-modal characteristics of ESPP;
the method for performing significance calculation on the remote sensing image by using the concentration of the atmospheric pollution precursor specifically comprises the following steps:
according to a visual selective attention mechanism and a spectrum residual error method, a significance calculation method of a remote sensing image is designed by utilizing remote sensing big data and troposphere column concentration distribution inverted by an atmospheric pollution precursor, and the method specifically comprises the following steps: using saliency maps SC RS And the original remote sensing image I RS Masking to obtain a significant region image Si in which ESPP may be present RS The method comprises the following steps:
wherein RS represents an available remote sensing image, VS represents a visible light image, SAR represents a synthetic aperture radar image, UV represents an ultraviolet image, IR represents an infrared image, HS represents a hyperspectral image, and VCD X Representing a concentration profile of trace gas species;
the adoption of a multi-level and multi-scale feature extraction algorithm to extract the cross-modal features of ESPP comprises the following specific steps:
ESPP activation feature Fi is extracted based on target feature alliance model and deep neural network RS The method comprises the following steps:
wherein, TFSM represents a target feature subjective model, DNN represents a deep neural network;
the ESPP features are subjected to reinforcement learning, and specifically:
activation feature Fi for sensing neuron N of DNN RS Extracting new activation characteristics Fi through an adjustment action a of network parameters and weights in an action space ACT RS ' obtain rewards of ESPP semanticsR is; the ESPP characteristic reinforcement learning is realized through a Markov decision process, namely:
wherein Fi is RS Belonging to a state space S, a epsilon ACT, r epsilon RW, wherein RW is a return function, and MDP represents a Markov decision process;
step 4: object-oriented remote sensing cross-modal ESPP representation and reasoning method based on probability cognition framework carries out multi-scale theme clustering, cross-modal ESPP cognition processing and ESPP recognition incremental learning on ESPP features;
the multi-scale theme clustering is carried out on ESPP features, and specifically comprises the following steps:
multi-scale topic clustering is carried out by adopting a hierarchical Dirichlet topic model, so that the activation characteristic Fi of ESPP is realized RS Attribute semantics Tf to ESPP RS Is a conversion of (1), namely:
wherein HLDA represents a hierarchical Dirichlet topic model; RS represents a usable remote sensing image, VS represents a visible light image, SAR represents a synthetic aperture radar image, UV represents an ultraviolet image, IR represents an infrared image, HS represents a hyperspectral image, VCD X Representing a concentration profile of trace gas species;
constructing a cross-modal knowledge graph CKG of a typical ESPP, namely:
wherein TS and CP are the type and confidence probability of ESPP respectively;
the cross-modal ESPP cognitive processing is performed on ESPP characteristics, and specifically comprises the following steps:
ESPP-based spatio-temporal context C TS And a cross-modal knowledge graph CKG, and establishing attribute semantics Tf of ESPP according to the mobile intelligent body, the cellular automaton and the cognitive computing theory RS Nonlinear mapping CTC with type TS, i.e.:
the incremental learning for ESPP identification on ESPP features is specifically as follows:
when ESPP modal information DeltaTf is added RS When the method is used, under the condition of keeping the cross-modal knowledge graph CKG of the original ESPP, CKG attributes are directly learned and updated without reconstruction, and the latest cross-modal knowledge graph CKG' is obtained through incremental learning and mapping TIL, namely:
2. the method of claim 1, wherein in step 1, the atmospheric NO is inverted using a differential absorption spectroscopy algorithm 2 And NH 3 Is used for generating O by adopting a band residual error algorithm 3 And SO 2 Is a tropospheric column concentration; using AOD-PM 2.5 Tropospheric column concentration and near-surface PM for AOD inverse of hybrid correlation model 2.5 Concentration.
3. The parallel processing method of the brain-like cross-modal identification method for a pollutant emission source of remote sensing spatiotemporal big data according to claim 1 or 2, wherein the whole identification process of the brain-like cross-modal identification method for a pollutant of remote sensing spatiotemporal big data is used as a BCR task, and the parallel processing method comprises:
step A1, task decomposition for ESPP parallel identification;
step A2, heterogeneous parallel acceleration for a BCR task;
a3, multi-core parallel acceleration for the BCR task;
and step A4, multi-machine parallel acceleration for the BCR task.
4. A method according to claim 3, wherein step A1 is specifically:
constructing a multi-resolution pyramid by utilizing the preprocessed remote sensing image, and slicing the image layer by layer to form a data block D k Decomposing the BCR task into M workflows, each workflow consisting of N serial task TRSs i Composition, each serial task TRS i Comprising Q parallelizable subtasks TRP j The method comprises the following steps:
5. the method according to claim 4, wherein step A2 is specifically: for each data block D k There are multiple loops of computation inside, programmed with GPU.
6. A method according to claim 3, wherein step A3 comprises: the step of performing saliency calculation on the remote sensing image by using the concentration of the atmospheric pollution precursor in the step 3, the step of extracting the cross-modal characteristics of the ESPP by using a multi-level and multi-scale characteristic extraction algorithm, and the step of performing multi-scale topic clustering on the ESPP characteristics in the step 4 adopt an OpenMP parallel programming algorithm.
7. The method according to claim 3 or 6, wherein step A4 comprises: the step of reinforcement learning of the ESPP features in the step 3, the step of cross-modal ESPP cognitive processing of the ESPP features in the step 4, and the step of incremental learning of ESPP recognition of the ESPP features adopt a Map-Reduce programming algorithm of MPI.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011146347.9A CN112231983B (en) | 2020-10-23 | 2020-10-23 | Brain-like cross-modal identification and parallel processing method for pollution precursor emission source of remote sensing space-time big data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011146347.9A CN112231983B (en) | 2020-10-23 | 2020-10-23 | Brain-like cross-modal identification and parallel processing method for pollution precursor emission source of remote sensing space-time big data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112231983A CN112231983A (en) | 2021-01-15 |
| CN112231983B true CN112231983B (en) | 2024-02-02 |
Family
ID=74110863
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011146347.9A Active CN112231983B (en) | 2020-10-23 | 2020-10-23 | Brain-like cross-modal identification and parallel processing method for pollution precursor emission source of remote sensing space-time big data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112231983B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113129215B (en) * | 2021-04-29 | 2022-11-22 | 中国科学院合肥物质科学研究院 | FTIR passive telemetering scanning imaging high-resolution reconstruction method for polluted gas |
| CN113344215B (en) * | 2021-06-01 | 2022-12-30 | 山东大学 | Extensible cognitive development method and system supporting new mode online learning |
| CN113535810B (en) * | 2021-06-25 | 2024-02-27 | 杨粤湘 | Mining method, device, equipment and medium for traffic violation objects |
| CN114492210B (en) * | 2022-04-13 | 2022-07-19 | 潍坊绘圆地理信息有限公司 | Hyperspectral satellite borne data intelligent interpretation system and implementation method thereof |
| CN115112586B (en) * | 2022-07-01 | 2024-06-07 | 行星数据科技(苏州)有限公司 | Pasture methane emission estimation method under multi-source data fusion |
| CN117607057A (en) * | 2023-08-24 | 2024-02-27 | 中国科学技术大学 | Full-period sky-ground integrated hyperspectral greenhouse/polluted gas three-dimensional telemetry and tracing and prediction method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108519340A (en) * | 2018-03-28 | 2018-09-11 | 刘诚 | A kind of satellite remote sensing contamination gas scapus concentration spatial and temporal distributions inversion algorithm |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105608697B (en) * | 2015-12-24 | 2018-04-13 | 电子科技大学 | Air pollution source discrimination based on remote sensing aerosol and firefly group's algorithm |
-
2020
- 2020-10-23 CN CN202011146347.9A patent/CN112231983B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108519340A (en) * | 2018-03-28 | 2018-09-11 | 刘诚 | A kind of satellite remote sensing contamination gas scapus concentration spatial and temporal distributions inversion algorithm |
Non-Patent Citations (2)
| Title |
|---|
| 宿兴涛 ; 许丽人 ; 张志标 ; 吴洪星 ; .基于卫星观测的我国北方地区紫外吸收性气溶胶的时空分布研究.气候与环境研究.2015,(01),全文. * |
| 马鹏飞 ; 厉青 ; 陈辉 ; 张丽娟 ; 张玉环 ; 王桥 ; 周春艳 ; 毛慧琴 ; 陈翠红 ; 王中挺 ; .京津冀及周边地区大气污染防治重点关注区域遥感综合分析.遥感技术与应用.2019,(02),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112231983A (en) | 2021-01-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112231983B (en) | Brain-like cross-modal identification and parallel processing method for pollution precursor emission source of remote sensing space-time big data | |
| Zhou et al. | Multi-output support vector machine for regional multi-step-ahead PM2. 5 forecasting | |
| CN106650825B (en) | Motor vehicle exhaust emission data fusion system | |
| Qiu et al. | Atmospheric dispersion prediction and source estimation of hazardous gas using artificial neural network, particle swarm optimization and expectation maximization | |
| Guo et al. | Air pollution forecasting using artificial and wavelet neural networks with meteorological conditions | |
| He et al. | Numerical model-based artificial neural network model and its application for quantifying impact factors of urban air quality | |
| Mirzaei et al. | Evaluation of linear, nonlinear, and hybrid models for predicting PM2. 5 based on a GTWR model and MODIS AOD data | |
| Zhang et al. | Spatiotemporal causal convolutional network for forecasting hourly PM2. 5 concentrations in Beijing, China | |
| Mo et al. | A novel framework for daily forecasting of ozone mass concentrations based on cycle reservoir with regular jumps neural networks | |
| CN112906941B (en) | Prediction method and system for dynamic correlation air quality time series | |
| Ma et al. | Estimation of the vertical distribution of particle matter (PM 2.5) concentration and its transport flux from lidar measurements based on machine learning algorithms | |
| Elbaz et al. | Spatiotemporal air quality forecasting and health risk assessment over smart city of NEOM | |
| Ballestrín et al. | Modeling solar extinction using artificial neural networks. Application to solar tower plants | |
| Trieu et al. | Evaluation of summertime surface ozone in Kanto area of Japan using a semi-regional model and observation | |
| Yu et al. | A fast forecasting method for PM2. 5 concentrations based on footprint modeling and emission optimization | |
| Braghiere et al. | Characterization of the radiative impact of aerosols on CO 2 and energy fluxes in the Amazon deforestation arch using artificial neural networks | |
| Umar et al. | A novel multi-model data-driven ensemble approach for the prediction of particulate matter concentration | |
| CN111723525A (en) | PM2.5 inversion method based on multi-source data and neural network model | |
| Kang et al. | High-emitting vehicle identification by on-road emission remote sensing with scarce positive labels | |
| Feng et al. | Performance evaluation of photographic measurement in the machine-learning prediction of ground PM2. 5 concentration | |
| Guo et al. | An enhanced deep learning approach to assessing inland lake water quality and its response to climate and anthropogenic factors | |
| Liu et al. | A comparison of statistical and machine-learning approaches for spatiotemporal modeling of nitrogen dioxide across Switzerland | |
| Han et al. | Estimation of high-resolution PM2. 5 concentrations based on gap-filling aerosol optical depth using gradient boosting model | |
| Li et al. | Improving air quality assessment using physics-inspired deep graph learning | |
| Do et al. | Assessing surface water pollution in Hanoi, Vietnam, using remote sensing and machine learning algorithms |
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 |