CN106709903A - PM2.5 concentration prediction method based on image quality - Google Patents
PM2.5 concentration prediction method based on image quality Download PDFInfo
- Publication number
- CN106709903A CN106709903A CN201611031140.0A CN201611031140A CN106709903A CN 106709903 A CN106709903 A CN 106709903A CN 201611031140 A CN201611031140 A CN 201611031140A CN 106709903 A CN106709903 A CN 106709903A
- Authority
- CN
- China
- Prior art keywords
- image
- concentration
- window
- value
- model
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a PM2.5 concentration prediction method based on image quality and belongs to the image processing technology field. The method comprises the following steps of firstly, carrying out image registering on a collected data set (filtering images shot in a rainy day) and taking an image block satisfying a dark channel principle as a final training set; then, recovering a transmissivity image of the training set, using a sliding window method to extract a characteristic from the transmissivity iamge, carrying out image characteristic standardization and eliminating an influence of a relative humidity; and then, using a robustness regression analysis method to analyze a relation between the extracted characteristic of the training set and a real PM2.5 concentration, and then acquiring a PM2.5 concentration prediction model; and finally completing PM2.5 concentration prediction. An experiment result shows that an algorithm used for predicting the PM2.5 concentration in the invention is better than an existing algorithm; an influence of a relative humidity meteorology condition on atmosphere imaging can be overcome; and the method has an important meaning for PM2.5 concentration monitoring in a daily life.
Description
Technical field
The present invention relates to a kind of PM2.5 concentration predictions method, particularly a kind of PM2.5 concentration predictions based on picture quality
Method.
Background technology
PM2.5 is suspended in all particulates general name of the diameter less than or equal to 2.5 μm in air, and it is to weigh air quality
An important indicator, and have a great impact to health.At present, PM2.5 concentration is all some are big in detection air
Equipment, these equipment costs are higher and need periodic maintenance.By it has been observed that the figure photographed in the case of different air qualities
, there is notable difference in quality in picture.Therefore want to design a kind of picture quality and PM2.5 concentration dependence model building methods.
The existing PM2.5 concentration prediction methods based on image procossing are mainly to be extracted some and can reflect the feature of picture quality.It is near several
Mainly occur in that the following two kinds is based on the PM2.5 concentration prediction methods of image procossing year：
(1) method based on Image Visual Feature.Such method extracts gradient, the color character of image.Use shot image
Sky color difference estimation PM2.5 concentration, the method influenceed by weather, such as：Cloudy day is gloomy, increased the mistake of estimation
Difference.Influence of the relative humidity to picture quality is not accounted for the Gradient Features of image.
(2) method based on image physical features.This process employs air Imaging physics model, and using dark primary elder generation
The transmittance figure that method of estimation recovers image is tested, eigenmatrix is extracted to transmittance figure with sliding window strategy, and using sane
The relational model that property regression analysis are set up between eigenmatrix and true PM2.5 concentration.But the method is not accounted for relatively
The influence that humidity is imaged to air.Relative humidity can influence the extinction capability of particulate in air PM2.5, and relative humidity is bigger,
The moisture content that PM2.5 absorbs in air is more, and the scattering power to atmosphere light is stronger, is imaged fuzzyyer.
Therefore, the existing PM2.5 Forecasting Methodologies based on image procossing do not account for the shadow that relative humidity is imaged to air
Ring, cause the PM2.5 concentration sealing degrees of accuracy relatively low.
The content of the invention
It is an object of the invention to provide a kind of picture quality and the new method of PM2.5 concentration dependence model constructions.
The technical solution for realizing the purpose of the present invention is：A kind of picture quality and PM2.5 concentration dependence model structures
The new method built, comprises the following steps：
Step 1, collection natural image of fixed place and time, and image is pre-processed；
Step 2, using extracting based on dark channel prior theory defogging algorithm the transmittance figure of collecting image；
Step 3, using sliding window method to the transmissivity image zooming-out eigenmatrix that is obtained in step 2；
Step 4, the eigenmatrix that step 3 is obtained is standardized, eliminate relative humidity influences on it；
Step 5, with robustness regression analysis model obtain PM2.5 concentration prediction models；
Step 6, the model obtained with step 5 are predicted to PM2.5 concentration.
Compared with prior art, its remarkable advantage is the present invention：1) present invention considers and overcomes relative humidity to air
The influence of imaging, improves the precision of prediction of PM2.5 estimations.2) this method has only taken off the block for meeting dark principle in image
To set up forecast model；It is simple and effective to eliminate useless piece in image, the modeling time is greatly shortened, while decreasing
Memory headroom needed for modeling.
The present invention is described in further detail below in conjunction with the accompanying drawings.
Brief description of the drawings
Fig. 1 is the flow chart that the present invention estimates PM2.5 concentration new methods based on image procossing.
Fig. 2 is to standardize the flow chart that relative humidity influences on image transmission rate feature.
Fig. 3 is the natural image being collected into smart mobile phone.
Fig. 4 is the useful image block taken from original image.
Fig. 5 is natural image transmittance figure.
Fig. 6 is particulate in air scattering moisture absorption growth factor empirical equation f (RH) schematic diagram.
Fig. 7 is area preference precedence diagram.
Fig. 8 is the training pattern that the inventive method is obtained.
Fig. 9 is the training pattern that the method based on image physical features is obtained.
Figure 10 is the inventive method PM2.5 concentration estimations and actual value comparison diagram.
Figure 11 is method PM2.5 concentration estimations and actual value comparison diagram based on image physical features.
Specific embodiment
It is of the invention that PM2.5 concentration new methods are estimated based on image procossing with reference to Fig. 1, comprise the following steps：
Step 1, collection natural image of fixed place and time, and image is pre-processed；Specially：
Step 1-1, the image for screening out rainy day shooting in data set；
Step 1-2, image registration is done to remaining data collection, selection piece image does as benchmark image to remaining image matches somebody with somebody
Standard, with it is punctual be Generalized Dual Bootstrap-ICP algorithms, transformation model selects similitude
Similarity；
Step 1-3, a certain piece of image in data set is taken as final training set, remove image garbage, figure
As block includes the scenery for meeting dark principle.
Step 2, using extracting based on dark channel prior theory defogging algorithm the transmittance figure of collecting image；Specially：
Step 2-1：Mini-value filtering is done respectively to tri- passages of image R, G, B in training set, window size p is：
1) .p=m*m
2) .m=floor (max ([3, w*kenlRatio, h*kenlRatio]))
Wherein m is window diameter, and w is the width of image, and h is the height of image, and kenlRatio is a ratio, value
Between 0.01 to 0.05；After finishing mini-value filtering to three passages, pixel minimum luminance value conduct in three passages is chosen
Dark channel diagram corresponding pixel points brightness value, so as to recover dark channel diagram；
Step 2-2：Obtain the atmosphere illumination intensity A of each image：First, before being taken according to the size of brightness from dark
0.1% pixel；Secondly, in these positions, the corresponding point with maximum brightness is found in original foggy image I
Value, as A values；
Step 2-3：Build air Imaging physics model:I (x)=J (x) t (x)+A (1-t (x)), wherein x are that pixel is sat
Mark, I is the foggy image for observing, J is clearly fog free images, and A is global atmosphere illumination intensity, and t is used for describing light leading to
Cross medium and be transmitted to the part not scattered during imaging device, transmissivity；
Step 2-4：Building dark channel prior theoretical model is：Its
Middle x is pixel point coordinates, and c represents any passage, and y is the field pixel point coordinates of pixel x, and J is clearly fog free images,
JduckX () is dark channel diagram, the model shows, in the regional area of most non-skies, certain some pixel always has
Very low value, tends to 0；
Step 2-5：Formula is derived by according to above-mentioned formulaWherein x is picture
Vegetarian refreshments coordinate, c represents any passage, and y is the field pixel point coordinates of pixel x, and I is the foggy image for observing, t (x)
It is transmittance figure to be asked；Thick transmittance figure is recovered by the formula, then carries out refining thick transmission plot with guiding wave filter,
Transmittance figure after refinement is required transmittance figure.
Transmittance figure picture conforms generally near big and far smaller rule, because the particulate matter in air is equally distributed, close to obtaining
The distance propagated in atmosphere of the light that sends of object it is short, the light into camera is more, and correspondence transmittance values are just big.Conversely, at a distance
Object transmittance values just it is small.
Step 3, using sliding window method to the transmissivity image zooming-out eigenmatrix that is obtained in step 2；Specially：
Step 3-1, sliding window are dimensioned toH and w be image height and
Width, ws is window size, and moving step length is set toStep is the step-length of window sliding；
Step 3-2, sliding window is progressively moved along the transverse and longitudinal direction of transmittance figure picture, and calculation window brightness is flat
The logarithm of average and then obtains eigenmatrix as the characteristic value of window, piece image one eigenmatrix of correspondence.
Step 4, the eigenmatrix that step 3 is obtained is standardized, eliminate relative humidity influences on it；Specifically
For：
Step 4-1, determine that different regions particulate in air scatters moisture absorption growth factor empirical equation f (RH)=1+a*
(RH/100)bIn two values of parameter a, b, wherein RH is relative humidity, and the value of parameter a, b is as shown in table 1：
Table 1
| Particulate in air type | a | b |
| Urban type | 2.06 | 3.60 |
| Ocean/city mixed type | 3.26 | 3.85 |
| Ocean type | 4.92 | 5.04 |
Step 4-2, f (RH) the standardized feature matrix with determination, i.e., with each characteristic value in eigenmatrix divided by f
(RH) influence of the relative humidity to characteristic value, is eliminated.
Step 5, with robustness regression analysis model obtain PM2.5 concentration prediction models；Specially：
Step 5-1, characteristic value and true PM2.5 concentration with each window of robustness regression analysis analyzing and training collection
Between relation, and calculate the correlation between characteristic value and true PM2.5 concentration；
Step 5-2, selection with true PM2.5 concentration dependences highest window as optimal window, and by the window pair
The relational model answered carries out PM2.5 concentration sealings as final training pattern, and model includes four parts, relation mould altogether
Type, optimal window coordinates, window size ws, moving step length step.
Step 6, the model obtained with step 5 are predicted to PM2.5 concentration.The forecast model can only be used for predicting training
The PM2.5 concentration corresponding to image that collection collection is there and then photographed, specially：
Step 6-1, the eigenmatrix that testing image is extracted according to above-mentioned steps；
Step 6-2, it is transmitted to forecast model using the optimal window character pair value of testing image as input, forecast model is returned
One predicted value is testing image shooting PM2.5 concentration there and then.
It is of the invention to consider and overcome the influence that relative humidity is imaged to air, improve the precision of prediction of PM2.5 estimations.
This method has only taken and has met the block of dark principle in image and set up forecast model；It is simple and effective to eliminate nothing in image
With block, the modeling time is greatly shortened, while decreasing memory headroom needed for modeling.
With reference to embodiment, the present invention will be further described in detail：
The natural image that the system invention is photographed using smart mobile phone is schemed using image processing meanses as input by input
As prognostic chart picture shoots PM2.5 concentration there and then.In order to predict that input picture shoots the concentration of PM2.5 there and then,
Image, the width of amount of images 15 are gathered firstly the need of going input picture to shoot ground in input picture shooting time smart mobile phone daily
More than.Then, with dynamic construction forecast model of the present invention, the PM2.5 that finally i.e. predictable input picture shoots there and then is dense
Degree.
The flow of the present embodiment is as shown in figure 1, the natural color of fixed place and time collected by smart mobile phone imaging device
Image size is 4160 × 3120, and image totally 20 width, shooting image is as shown in Figure 3.The image of rainy day shooting is screened out first, so
Image registration is done to remaining 15 width image afterwards.Image registration is Generalized Dual Bootstrap-ICP algorithms,
That transformation model selection is similitude Similarity when registering.Then image block is taken, Fig. 4 gives this example and really uses
Image block, size be 400 × 300.
15 width images are done with registration, is scratched after figure pre-process, using based on dark channel prior theory defogging algorithm recovery 15
The transmittance figure of width image, as shown in figure 5, concretely comprising the following steps：
The first step：Mini-value filtering is done respectively to 15 tri- passages of width image R, G, B, window size is 20*20.Choose three
Pixel minimum luminance value is used as dark channel diagram corresponding pixel points brightness value in individual passage.
Second step：Obtain the atmosphere illumination intensity A of each image：First, before being taken according to the size of brightness from dark
0.1% pixel；Secondly, in these positions, the corresponding point with maximum brightness is found in original foggy image I
Value, as A values.
3rd step：According to formulaRecover thick transmittance figure, then with directiveness
Wave filter carries out refining thick transmission plot, and the transmittance figure after refinement is required transmittance figure.
After recovering transmittance figure, using sliding window method to transmissivity image zooming-out eigenmatrix.Concretely comprise the following steps：
The first step：Sliding window is dimensioned to 15*15 in this example, and window moving step length is 2.
Second step：Sliding window is progressively moved along the transverse and longitudinal direction of transmittance figure picture, and calculation window brightness is flat
The logarithm of average and then obtains eigenmatrix as the characteristic value of window.
The eigenmatrix of 15 width images is extracted, eigenmatrix has been standardized, eliminated relative humidity to its shadow
Ring.Fig. 6 gives particulate in air scattering moisture absorption growth factor empirical equation f (RH) schematic diagram.Generally, f (RH) with
The increase of RH and increase, in RH<The value of 40%, f (RH) close to 1, when illustrating that envionmental humidity is relatively low, particulate in air
Particle diameter does not increase significantly.In addition, the f (RH) of different regions has obvious difference under the conditions of same humidity.This example
Middle a=1.24, b=6.27. according to Fig. 2 to the step of the eigenmatrix that obtains is standardized, eliminate the influence of relative humidity.
Then modeled with robustness regression analysis and obtain PM2.5 concentration prediction models.The width of this instance graph image set 15, adopts
Estimate that each image shoots PM2.5 concentration there and then with a method of remaininging, i.e., every time with 14 width image therein as training set
Modeling, a remaining width is tested as test set, is carried out altogether 15 times.Concretely comprise the following steps：
The first step：With robustness regression analysis (matlab from tape function robustfit functions) 14 width images of analysis
Relation between the characteristic value (each characteristic value in eigenmatrix) of each window and true PM2.5 concentration builds prediction mould
Type, and the correlation between characteristic value and true PM2.5 concentration is calculated, area preference precedence diagram is obtained, as shown in fig. 7, each
Window one model dependency of correspondence, color is deeper, and to represent correlation higher.The most deep window of color is designated as optimal window, optimal
Window coordinates are x=23, y=90.
Second step：By optimal window x=23, the corresponding relational models of y=90 (as shown in Figure 8) is used as final training mould
Type carries out PM2.5 concentration sealings.
PM2.5 concentration predictions are finally carried out to test set image with the forecast model for obtaining, specially test set is optimal
Window (x=23, y=90) character pair value brings forecast model into, and forecast model returns to a PM2.5 concentration prediction value.
Fig. 9 is the PM2.5 concentration prediction models that the existing method for extracting image physical features is obtained.Figure 10 is this example
The PM2.5 concentration estimations and actual value comparison diagram for obtaining.Figure 11 is the method PM2.5 concentration sealings based on image physical features
Value and actual value comparison diagram.Table 2 is that the inventive method and existing methods compare.
Table 2
| Before standardization | After standardization | Improve | |
| R | 0.8741 | 0.9056 | 0.0315 |
| MAE(ug/m3) | 7.0219 | 5.7105 | 1.3114 |
Knowable to Figure 10, Figure 11, table 2：The PM2.5 concentration prediction models that the present invention is obtained can accurately estimate intelligence
The PM2.5 concentration of image shot by cell phone there and then, and the influence that relative humidity is imaged to air is overcome, PM2.5 concentration is estimated
Correlation R and absolute average error MAE between evaluation and actual value are better than existing method, and this method has only taken off figure
Meet the block of dark principle as in set up forecast model.It is simple and effective to eliminate useless piece in image, greatly shorten
Modeling time.The shortening of time and rise to daily life PM2.5 concentration predictions and the monitoring of estimated accuracy are provided conveniently.
Claims (7)
1. a kind of PM2.5 concentration prediction methods based on picture quality, it is characterised in that comprise the following steps：
Step 1, collection natural image of fixed place and time, and image is pre-processed；
Step 2, using the transmittance figure for recovering based on dark channel prior theory defogging algorithm collection image；
Step 3, using sliding window method to the transmissivity image zooming-out eigenmatrix that is obtained in step 2；
Step 4, the eigenmatrix that step 3 is obtained is standardized, eliminate relative humidity influences on it；
Step 5, with robustness regression analysis model obtain PM2.5 concentration prediction models；
Step 6, the model obtained with step 5 are predicted to PM2.5 concentration.
2. PM2.5 concentration prediction methods based on picture quality according to claim 1, it is characterised in that right in step 1
Image does pretreatment and is specially：
Step 1-1, the image for screening out rainy day shooting in data set；
Step 1-2, image registration is done to remaining data collection, selection piece image does registration as benchmark image to remaining image,
With it is punctual be Generalized Dual Bootstrap-ICP algorithms, transformation model selects similitude Similarity；
Step 1-3, a certain piece of image in data set is taken as final training set, remove image garbage, image block
Comprising the scenery for meeting dark principle.
3. PM2.5 concentration prediction methods based on picture quality according to claim 1, it is characterised in that adopted in step 2
Recover to collect the transmittance figure of image with based on dark channel prior theory defogging algorithm, specially：
Step 2-1：Mini-value filtering is done respectively to tri- passages of image R, G, B in training set, window size p is：
1) .p=m*m
2) .m=floor (max ([3, w*kenlRatio, h*kenlRatio]))
Wherein m is window diameter, and w is the width of image, and h is the height of image, and kenlRatio is a ratio, and value exists
Between 0.01 to 0.05；After finishing mini-value filtering to three passages, pixel minimum luminance value is used as dark in choosing three passages
Passage figure corresponding pixel points brightness value, so as to recover dark channel diagram；
Step 2-2：Obtain the atmosphere illumination intensity A of each image：First, before being taken according to the size of brightness from dark
0.1% pixel；Secondly, in these positions, the corresponding point with maximum brightness is found in original foggy image I
Value, as A values；
Step 2-3：Build air Imaging physics model:I (x)=J (x) t (x)+A (1-t (x)), wherein x are pixel point coordinates, I
It is the foggy image for observing, J is clearly fog free images, and A is global atmosphere illumination intensity, and t is used for describing light by being situated between
Matter is transmitted to the part not scattered during imaging device, transmissivity；
Step 2-4：Building dark channel prior theoretical model is：Wherein x is
Pixel point coordinates, c represents any passage, and y is the field pixel point coordinates of pixel x, and J is clearly fog free images, Jdark
X () is dark channel diagram, the model shows, in the regional area of most non-skies, certain some pixel always has very low
Value, tend to 0；
Step 2-5：Formula is derived by according to above-mentioned formulaWherein x is pixel
Coordinate, c represents any passage, and y is the field pixel point coordinates of pixel x, and I is the foggy image for observing, t (x) is to treat
The transmittance figure asked；Thick transmittance figure is recovered by the formula, then carries out refining thick transmission plot with guiding wave filter, refined
Transmittance figure afterwards is required transmittance figure.
4. PM2.5 concentration prediction methods based on picture quality according to claim 1, it is characterised in that step 3 is utilized
Sliding window method is specially to transmissivity image zooming-out eigenmatrix：
Step 3-1, sliding window are dimensioned toH and w are the height and width of image
Degree, ws is window size, and moving step length is set toStep is the step-length of window sliding；
Step 3-2, sliding window is progressively moved along the transverse and longitudinal direction of transmittance figure picture, and calculation window brightness average value
Logarithm as the characteristic value of window, and then obtain eigenmatrix, piece image one eigenmatrix of correspondence.
5. PM2.5 concentration prediction methods based on picture quality according to claim 1, it is characterised in that step 4 pair spy
Levy matrix to be standardized, specially：
Step 4-1, determine that different regions particulate in air scatters moisture absorption growth factor empirical equation f (RH)=1+a* (RH/
100)bIn two values of parameter a, b, wherein RH is relative humidity, and the value of parameter a, b is as shown in table 1：
Table 1
Step 4-2, f (RH) the standardized feature matrix with determination, i.e., with each characteristic value in eigenmatrix divided by f (RH),
Eliminate influence of the relative humidity to characteristic value.
6. PM2.5 concentration prediction methods based on picture quality according to claim 1, it is characterised in that step 5 is with steady
Strong property regression analysis modeling obtains PM2.5 concentration prediction models and is specially：
Step 5-1, between the characteristic value and true PM2.5 concentration of robustness regression analysis analyzing and training collection each window
Relation, and calculate the correlation between characteristic value and true PM2.5 concentration；
Step 5-2, selection are as optimal window and the window is corresponding with true PM2.5 concentration dependences highest window
Relational model carries out PM2.5 concentration sealings as final training pattern, and model includes four parts altogether, relational model, most
Excellent window coordinates, window size ws, moving step length step.
7. PM2.5 concentration prediction methods based on picture quality according to claim 1, it is characterised in that step 6 with
To model PM2.5 concentration is predicted, the forecast model can only be used for predicting that training set collection is there and then photographed
PM2.5 concentration corresponding to image, specially：
Step 6-1, the eigenmatrix that testing image is extracted according to above-mentioned steps；
Step 6-2, it is transmitted to forecast model using the optimal window character pair value of testing image as input, forecast model returns to one
Predicted value is testing image shooting PM2.5 concentration there and then.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611031140.0A CN106709903B (en) | 2016-11-22 | 2016-11-22 | PM2.5 concentration prediction method based on image quality |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611031140.0A CN106709903B (en) | 2016-11-22 | 2016-11-22 | PM2.5 concentration prediction method based on image quality |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106709903A true CN106709903A (en) | 2017-05-24 |
| CN106709903B CN106709903B (en) | 2020-06-19 |
Family
ID=58941184
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611031140.0A Active CN106709903B (en) | 2016-11-22 | 2016-11-22 | PM2.5 concentration prediction method based on image quality |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106709903B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108693087A (en) * | 2018-04-13 | 2018-10-23 | 中国科学院城市环境研究所 | A kind of air quality monitoring method based on image understanding |
| CN108761571A (en) * | 2018-04-03 | 2018-11-06 | 北方民族大学 | Atmospheric visibility prediction technique based on neural network and system |
| CN111598156A (en) * | 2020-05-14 | 2020-08-28 | 北京工业大学 | PM based on multi-source heterogeneous data fusion2.5Prediction model |
| WO2020232710A1 (en) * | 2019-05-23 | 2020-11-26 | 深圳大学 | Haze image quality evaluation method and system, storage medium, and electronic device |
| CN112580600A (en) * | 2020-12-29 | 2021-03-30 | 神华黄骅港务有限责任公司 | Dust concentration detection method and device, computer equipment and storage medium |
| CN113433037A (en) * | 2021-05-13 | 2021-09-24 | 北京理工大学 | Mobile phone type PM2.5Observation device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103903273A (en) * | 2014-04-17 | 2014-07-02 | 北京邮电大学 | PM2.5 grade fast-evaluating system based on mobile phone terminal |
| CN104217404A (en) * | 2014-08-27 | 2014-12-17 | 华南农业大学 | Video image sharpness processing method in fog and haze day and device thereof |
| CN105735416A (en) * | 2016-04-12 | 2016-07-06 | 顾美红 | Remotely controllable air purification water maker |
-
2016
- 2016-11-22 CN CN201611031140.0A patent/CN106709903B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103903273A (en) * | 2014-04-17 | 2014-07-02 | 北京邮电大学 | PM2.5 grade fast-evaluating system based on mobile phone terminal |
| CN104217404A (en) * | 2014-08-27 | 2014-12-17 | 华南农业大学 | Video image sharpness processing method in fog and haze day and device thereof |
| CN105735416A (en) * | 2016-04-12 | 2016-07-06 | 顾美红 | Remotely controllable air purification water maker |
Non-Patent Citations (2)
| Title |
|---|
| XIAOYANG LIU: "《PM2.5 Monitoring using Images from Smartphones in Participatory Sensing》", 《THE FIRST INTERNATIONAL WORKSHOP ON SMART CITIES AND URBAN INFORMATICS 2015》 * |
| XINGANG LIU: "《Influences of relative humidity and particle chemical composition on aerosol scattering properties during the 2006 PRD campaign》", 《ATMOSPHERIC ENVIRONMENT》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108761571A (en) * | 2018-04-03 | 2018-11-06 | 北方民族大学 | Atmospheric visibility prediction technique based on neural network and system |
| CN108693087A (en) * | 2018-04-13 | 2018-10-23 | 中国科学院城市环境研究所 | A kind of air quality monitoring method based on image understanding |
| WO2020232710A1 (en) * | 2019-05-23 | 2020-11-26 | 深圳大学 | Haze image quality evaluation method and system, storage medium, and electronic device |
| CN111598156A (en) * | 2020-05-14 | 2020-08-28 | 北京工业大学 | PM based on multi-source heterogeneous data fusion2.5Prediction model |
| CN112580600A (en) * | 2020-12-29 | 2021-03-30 | 神华黄骅港务有限责任公司 | Dust concentration detection method and device, computer equipment and storage medium |
| CN113433037A (en) * | 2021-05-13 | 2021-09-24 | 北京理工大学 | Mobile phone type PM2.5Observation device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106709903B (en) | 2020-06-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106709903A (en) | PM2.5 concentration prediction method based on image quality | |
| CN107665492B (en) | Colorectal panoramic digital pathological image tissue segmentation method based on depth network | |
| Li et al. | Using user generated online photos to estimate and monitor air pollution in major cities | |
| Raikwar et al. | Lower bound on transmission using non-linear bounding function in single image dehazing | |
| CN102297660B (en) | Measuring method of shield tunnel lining segment seam open width and apparatus thereof | |
| CN109977790A (en) | A kind of video smoke detection and recognition methods based on transfer learning | |
| CN106650812B (en) | A kind of urban water-body extracting method of satellite remote-sensing image | |
| CN111126136A (en) | Smoke concentration quantification method based on image recognition | |
| CN109559310A (en) | Power transmission and transformation inspection image quality evaluating method and system based on conspicuousness detection | |
| Bo et al. | Particle pollution estimation from images using convolutional neural network and weather features | |
| CN106779054B (en) | A kind of PM2.5 methods of estimation based on Misty Image | |
| CN111339948B (en) | Automatic identification method for newly-added buildings of high-resolution remote sensing images | |
| CN110503637B (en) | Road crack automatic detection method based on convolutional neural network | |
| CN116543241B (en) | Detection method and device for leakage gas cloud, storage medium and electronic equipment | |
| CN113469097B (en) | Multi-camera real-time detection method for water surface floaters based on SSD network | |
| CN109325426A (en) | A kind of black smoke vehicle detection method based on three orthogonal plane space-time characteristics | |
| CN117011756A (en) | Video rainfall inversion method based on migration learning method | |
| CN110765900B (en) | Automatic detection illegal building method and system based on DSSD | |
| CN117152617A (en) | Urban flood identification method and system | |
| CN110059544B (en) | Pedestrian detection method and system based on road scene | |
| CN110930393A (en) | Chip material pipe counting method, device and system based on machine vision | |
| Gao et al. | Intelligent crack damage detection system in shield tunnel using combination of retinanet and optimal adaptive selection | |
| CN103278505B (en) | Blast furnace fly ash constituent analysis method based on multi-feature analysis | |
| CN116109813A (en) | Anchor hole drilling identification method, system, electronic equipment and medium | |
| AU2021102276A4 (en) | A Detection Method for Quantity of Residual Plastic Films Based on Heuristic Iterative Algorithm and Its Early Warning System |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |