CN113269805B - Rainfall event guided remote sensing rainfall inversion training sample self-adaptive selection method - Google Patents
Rainfall event guided remote sensing rainfall inversion training sample self-adaptive selection method Download PDFInfo
- Publication number
- CN113269805B CN113269805B CN202110655233.5A CN202110655233A CN113269805B CN 113269805 B CN113269805 B CN 113269805B CN 202110655233 A CN202110655233 A CN 202110655233A CN 113269805 B CN113269805 B CN 113269805B
- Authority
- CN
- China
- Prior art keywords
- precipitation
- connected domain
- sample
- pixel
- samples
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- 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)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a remote sensing rainfall inversion training sample self-adaptive selection method guided by rainfall events. The method comprises the following steps: acquiring a precipitation binary diagram, and dividing the precipitation area into a plurality of precipitation connected areas by adopting a classification algorithm; for any precipitation connected domain, obtaining the sampling number of the precipitation connected domain according to the pixel number of the precipitation connected domain, the pixel number of the precipitation region and the preset precipitation sample number; sampling pixels of the precipitation connected domain according to the sampling number of the precipitation connected domain and the number of the pixels of the precipitation connected domain to obtain precipitation samples of the precipitation connected domain; randomly sampling pixels of the non-precipitation area to obtain a non-precipitation sample; and determining precipitation samples and non-precipitation samples of all precipitation connected domains as precipitation inversion training samples. The method can simultaneously meet the representativeness of the training sample and the universality of the inversion model so as to improve the inversion accuracy of the inversion model.
Description
Technical Field
The invention relates to the field of rainfall inversion, in particular to a remote sensing rainfall inversion training sample self-adaptive selection method and system guided by rainfall events.
Background
Precipitation plays an important role in the fields of hydrology, meteorology, ecology, agricultural research and the like, and particularly, one of the main driving forces of global scale matter energy exchange. Therefore, it is very important to use satellite data to perform rainfall inversion to obtain rainfall data with high precision and high space-time resolution. The precipitation inversion is a mathematical process for estimating the precipitation value by using satellite observation data, and the method usually adopts a model method, such as fitting formula parameter solving, machine learning and the like. No matter whether fitting formula parameter solving or machine learning is carried out, the model method needs to input an actually measured precipitation value and a characteristic value (usually cloud top brightness temperature) observed by a satellite as training data, after parameters are determined by a certain method, the trained model can be used for precipitation inversion, namely, a new observed value is input, and an inversion estimated value can be obtained. The selection of training samples in the process has a very direct influence on the final inversion result, that is, the strategy of sample selection can feel the final training result of the model.
An excellent sample selection strategy should satisfy both the characteristics of representative samples and universality of models. The representativeness means that various special cases can be covered in the sample selection process, and each case has enough samples, so that the model can not lack effective information in the training process. And universality means that the model has higher stability to different or new input, and is not limited to the types and situations provided by the sample set.
Existing sample selection strategies mainly include three major types: a random selection method, a numerical distribution control method, and a sample selection method based on similarity (consistency) determination.
The random selection method is to select all samples according to a certain random number principle or a certain interval step length under the condition of disordering sequence, so as to ensure the randomness of sample selection. The random selection method is simple and easy to implement, and the selected samples have certain universality under the condition of sufficient quantity, namely, the selected samples are used for model training, so that the obtained training model can deal with most conditions and can identify most features. However, certain representativeness is lacked, because random selection usually selects a large number of repeated characteristic samples or invalid characteristic samples, which results in that the representative capacity of the whole event is greatly reduced and greater data redundancy is generated under the condition that the sample set has the same number of samples.
The numerical value distribution control method is characterized in that a certain number of samples are selected from each numerical value area and added into the sample set according to a numerical value distribution rule in the whole event set in a numerical value area dividing mode. The sample set selected in the way can have stronger representativeness, and as the samples come from different specified value intervals, each value interval is guaranteed to have a certain number of samples, so that the sample set can represent most of the occurring conditions. However, the division of the numerical value interval and the allocation of the denominations are prior, so that the method is supervised sample selection, and human factors are inevitably added into a sample set, so that certain universality is reduced, and the finally trained model can only reflect the situation of the sample considered by human.
The sample selection method based on similarity (consistency) judgment is a more complex method, and the method generally judges through the similarity between samples and then refines the classification of a sample set through a continuous iteration method so that the final samples have higher consistency. The method is complex, the definition of the similarity is various, and the iteration method is more alternative. But in summary, the representativeness of the samples can be increased, namely, the occurrence and the characteristics of the whole event can be represented by a small number of sample sets, and the proportion of invalid samples and repeated samples is greatly reduced. However, due to the elimination of inconsistent samples, samples in complex situations are lost, overfitting is easy to occur in the model training process, and the universality of the model is finally reduced.
In summary, the existing sample selection strategy cannot satisfy both the sample representativeness and the model universality.
Disclosure of Invention
The invention aims to provide a remote sensing rainfall inversion training sample self-adaptive selection method guided by rainfall events, which can simultaneously meet the representativeness of training samples and the universality of an inversion model so as to improve the inversion accuracy of the inversion model.
In order to achieve the purpose, the invention provides the following scheme:
a rainfall event guided remote sensing rainfall inversion training sample self-adaptive selection method comprises the following steps:
acquiring a precipitation binary image, wherein the precipitation binary image comprises a precipitation area and a non-precipitation area;
adopting a classification algorithm to divide the precipitation area into a plurality of precipitation connected domains;
for any precipitation connected domain, obtaining the sampling number of the precipitation connected domain according to the pixel number of the precipitation connected domain, the pixel number of the precipitation area and the preset precipitation sample number;
sampling pixels of the precipitation connected domain according to the sampling number of the precipitation connected domain and the number of the pixels of the precipitation connected domain to obtain precipitation samples of the precipitation connected domain;
randomly sampling pixels of the non-precipitation area to obtain a non-precipitation sample;
and determining precipitation samples and non-precipitation samples of all precipitation connected domains as precipitation inversion training samples.
Optionally, the acquiring a precipitation binary image specifically includes:
acquiring a precipitation condition graph, wherein the pixel value of the precipitation condition graph is the precipitation amount, and the pixel position of the precipitation condition graph is the geographic position corresponding to the pixel;
And carrying out threshold segmentation on the precipitation condition graph to obtain a precipitation binary graph.
Optionally, before the performing connected domain division on the precipitation area by using the classification algorithm to obtain a plurality of precipitation connected domains, the method further includes:
and carrying out image mask processing on the precipitation binary image to obtain a processed precipitation area.
Optionally, the sampling number of the precipitation connected domain is obtained according to the number of pixels of the precipitation connected domain, the number of pixels of the precipitation region and the number of preset precipitation samples, and specifically is:
according to the formula Mi＝(Ni/Ntotal)*MtotalCalculating the sampling number of the ith precipitation connected domain, wherein MiNumber of samples, M, for the ith precipitation connected domaintotalPresetting the number of precipitation samples; n is a radical of hydrogeniIs the number of pixels of the ith precipitation connected domain, NtotalIs the number of pixels of the precipitation area.
Optionally, the pixels of the precipitation connected domain are sampled according to the sampling number of the precipitation connected domain and the number of pixels of the precipitation connected domain, so as to obtain precipitation samples of the precipitation connected domain, specifically:
according to the formula Si＝{Pk|k＝[(t-0.5)/Mi*Ni]，t∈N+，1≤t≤MiObtaining a precipitation sample of the ith precipitation connected domain, wherein SiPrecipitation sample for the ith precipitation connected field, PkThe kth precipitation pixel in the precipitation sample of the ith precipitation connected domain, t is the sample number of the precipitation connected domain, and M iNumber of samples, N, for the ith precipitation connected domainiThe number of pixels of the ith precipitation connected domain.
Optionally, the randomly sampling the pixels in the non-precipitation area to obtain a non-precipitation sample specifically includes:
labeling each pixel in the non-precipitation area according to the position of each pixel in the non-precipitation area to obtain the pixel sequence number of each pixel in the non-precipitation area;
and randomly sampling the pixels of the non-precipitation area according to the number of the preset non-precipitation samples and the pixel serial number of each pixel to obtain the non-precipitation samples.
A system is selected to remote sensing precipitation inversion training sample self-adaptation of precipitation incident guide, includes:
the system comprises an acquisition module, a control module and a control module, wherein the acquisition module is used for acquiring a precipitation binary image which comprises a precipitation area and a non-precipitation area;
the precipitation connected domain determining module is used for dividing the precipitation connected domain into a plurality of precipitation connected domains by adopting a classification algorithm;
the quantity determining module is used for obtaining the sampling quantity of any precipitation connected domain according to the pixel number of the precipitation connected domain, the pixel number of the precipitation region and the preset precipitation sample quantity;
the rainfall sample determining module is used for sampling pixels of the rainfall connected domain according to the sampling number of the rainfall connected domain and the number of the pixels of the rainfall connected domain to obtain rainfall samples of the rainfall connected domain;
The non-precipitation sample determining module is used for randomly sampling the pixels of the non-precipitation area to obtain a non-precipitation sample;
and the inversion training sample determining module is used for determining precipitation samples of all precipitation connected domains and the non-precipitation samples as precipitation inversion training samples.
Optionally, the obtaining module includes:
the device comprises an acquisition unit, a calculation unit and a display unit, wherein the acquisition unit is used for acquiring a precipitation situation graph, the pixel value of the precipitation situation graph is the precipitation amount, and the pixel position of the precipitation situation graph is the geographic position corresponding to a pixel;
and the binary image determining unit is used for carrying out threshold segmentation on the precipitation situation image to obtain a precipitation binary image.
Optionally, the system for adaptively selecting the remote sensing precipitation inversion training sample guided by the precipitation event further includes: and the processing module is used for carrying out image mask processing on the precipitation binary image to obtain a processed precipitation area.
Optionally, the non-precipitation sample determination module includes:
a sequence number determining unit, configured to label each pixel in the non-precipitation region according to a position of each pixel in the non-precipitation region to obtain a pixel sequence number of each pixel in the non-precipitation region;
and the non-precipitation sample determining unit is used for randomly sampling the pixels of the non-precipitation area according to the number of the preset non-precipitation samples and the pixel serial number of each pixel to obtain the non-precipitation samples.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: the invention obtains the sampling number of each precipitation connected domain according to the pixel number of each precipitation connected domain, the pixel number of each precipitation connected domain and the preset precipitation sample number, so that each precipitation connected domain can be sampled, and the extracted sample number has no large repetition and redundancy due to covering each precipitation connected domain, thereby greatly solving the problem of insufficient representativeness of a random selection method.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
Fig. 1 is a flowchart of a method for adaptively selecting a training sample for remote sensing precipitation inversion guided by a precipitation event according to an embodiment of the present invention;
FIG. 2 is a detailed flow chart of an experiment using a remote sensing precipitation inversion training sample adaptive selection method guided by precipitation events;
FIG. 3 is a graph of comparison of different algorithm results in the present invention, FIG. 3(a) is a graph of the results of a random average sampling method, and FIG. 3(b) is a graph of the results of a sample adaptive selection method based on precipitation events according to the present invention;
FIG. 4 is a block diagram of a remote sensing precipitation inversion training sample adaptive selection system guided by precipitation events according to an embodiment of the present invention;
fig. 5 is a flowchart of a more specific method for adaptively selecting remote sensing precipitation inversion training samples guided by precipitation events according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
In order to make the aforementioned objects, features and advantages of the present invention more comprehensible, the present invention is described in detail with reference to the accompanying drawings and the detailed description thereof.
The current training sample selection methods mainly include a random method, a numerical distribution control method and a sample selection method based on similarity (consistency) judgment. The representativeness and universality of the selected lower model in different sample sets are emphasized, and the characteristics are listed as follows:
the representativeness of the models under different sample selection strategies is from strong to weak: a sample selection method, a numerical distribution control method and a random selection method based on similarity judgment; the universality of models under different sample selection strategies is from strong to weak: a random selection method, a numerical distribution control method, a sample selection method based on similarity judgment; and the influence degree of the artificial interference caused by the prior setting on the sample set is from strong to weak: a numerical distribution control method, a sample selection method based on similarity judgment, and a random selection method.
The random selection method has the advantages of simple implementation, but lacks representativeness to the positive sample; the numerical value distribution control method can effectively improve the representativeness of the aligned sample, but the supervision in the numerical value distribution control method is greatly enhanced, and the result is easily influenced by the prior set parameters; the sample selection method based on similarity judgment can obviously improve the intra-set consistency of the sample set, but the process needs continuous iteration and is slow in calculation process, meanwhile, due to the improvement of the intra-set consistency, the probability of overfitting is obviously enhanced, and the universality of a training model is obviously reduced by overfitting, so that the model can not be used. The model with strong representativeness and weak universality is better simulated for the representative conditions in the sample set, but singular value results appear under the condition that the sample set does not appear or appears less, and the influence on the overall simulation result is larger.
According to the sample determination method provided by the embodiment, the sample selection denominations are distributed to the precipitation events according to the actual precipitation events, and a precipitation sample set with high representativeness and universality is finally generated for training and learning of a precipitation inversion model.
Step 101: and acquiring a precipitation binary diagram. The precipitation binary image comprises a precipitation area and a non-precipitation area.
Step 102: and carrying out connected domain division on the precipitation areas by adopting a classification algorithm to obtain a plurality of precipitation connected domains.
Step 103: and for any precipitation connected domain, obtaining the sampling number of the precipitation connected domain according to the pixel number of the precipitation connected domain, the pixel number of the precipitation region and the preset precipitation sample number.
Step 104: and sampling the pixels of the precipitation connected domain according to the sampling number of the precipitation connected domain and the number of the pixels of the precipitation connected domain to obtain a precipitation sample of the precipitation connected domain.
Step 105: and randomly sampling the pixels of the non-precipitation area to obtain a non-precipitation sample. For the non-precipitation area, the area is often large and the non-precipitation area is connected into slices, so the method is simple and only needs to be selected on average according to the serial number of the pixel position.
Step 106: and determining precipitation samples and non-precipitation samples of all precipitation connected domains as precipitation inversion training samples.
The input precipitation situation map needs to be thresholded to produce a distribution map of precipitation events (precipitation binary map) because precipitation predictions often have a large number of small estimates approaching 0, and whether such estimates need to be determined as precipitation events needs to be achieved by thresholding. In practical application, step 101 specifically includes:
and acquiring a precipitation situation graph in a grid format, wherein the pixel value of the precipitation situation graph is the precipitation amount, and the pixel position of the precipitation situation graph is the geographic position corresponding to the pixel.
And carrying out threshold segmentation on the precipitation situation graph to obtain a precipitation binary graph. The method adopts a relatively international universal value (0.1mm/hour) as a threshold value to judge the precipitation event: that is, the pixel greater than or equal to the threshold is marked as 1, and is judged as precipitation, and the pixel smaller than the threshold is marked as 0, and is judged as no precipitation.
In practical application, the method for dividing the connected domains of the precipitation regions by adopting the classification algorithm to obtain the plurality of precipitation connected domains is based on the connected domain principle, namely, adjacent and close pixels can be judged as subsets below the same connected domain, so that the condition that the pixels are independent does not occur in the extracted adjacent connected domains, and the adopted adjacent judgment is 4-adjacent judgment, namely, when a certain pixel is positioned right above another pixel, right below the another pixel, right left or right, the two pixels can be judged as adjacent. Thus, in this manner, a precipitation profile is divided into a plurality of connected components, each of which is identified as a geographically continuous precipitation event.
Image dilation and erosion are one of the basic steps in the graphics operation. In practical application, before the step of dividing the precipitation connected domain into a plurality of precipitation connected domains by using the classification algorithm, the method further comprises the following steps:
and carrying out image mask processing on the precipitation binary image to obtain a processed precipitation area, and finally expanding or reducing the range of the connected area. The threshold value of the graph expansion or erosion is selected as 4 connected domains, and the operation is only carried out once, the erosion is carried out before the expansion, and the purpose of the step is to reduce the influence of single precipitation pixels and reduce the influence of random noise, so that the whole algorithm is focused on large-scale precipitation events.
In practical application, the sampling number of the precipitation connected domain is obtained according to the number of pixels of the precipitation connected domain, the number of pixels of the precipitation region and the number of preset precipitation samples, and the method specifically comprises the following steps:
according to Mi＝(Ni/Ntotal)*MtotalFormula (1)
Calculating the sampling number of the ith precipitation connected domain, wherein MiNumber of samples, M, for the ith precipitation connected fieldtotalThe number of the precipitation samples is preset; n is a radical ofiIs the number of pixels of the ith precipitation connected domain, NtotalIs the number of pixels of the precipitation area.
In practical application, the pixels of the precipitation connected domain are sampled according to the sampling number of the precipitation connected domain and the number of the pixels of the precipitation connected domain to obtain precipitation samples of the precipitation connected domain, and the sampling method specifically comprises the following steps:
according to Si＝{Pk|k＝[(t-0.5)/Mi*Ni]，t∈N+，1≤t≤MiEquation (2)
Obtaining a precipitation sample of the ith precipitation connected domain, wherein SiPrecipitation sample for the ith precipitation connected field, PkThe kth precipitation pixel in the precipitation sample of the ith precipitation connected domain, t is the sample number of the precipitation connected domain, and MiNumber of samples, N, for the ith precipitation connected fieldiThe number of pixels of the ith precipitation connected domain.
In practical application, the randomly sampling the pixels in the non-precipitation area to obtain a non-precipitation sample specifically includes:
And labeling each pixel in the non-precipitation area according to the position of each pixel in the non-precipitation area to obtain the pixel sequence number of each pixel in the non-precipitation area.
And randomly sampling the pixels of the non-precipitation area according to the number of the preset non-precipitation samples and the pixel serial number of each pixel to obtain the non-precipitation samples.
As shown in fig. 2, the present embodiment further provides a method for testing the rainfall of 2018 at 6 th and 1 st longitude latitudes (105.0 ° to 130.0 ° E, 18.2 ° to 38.0 ° N) by using the above method, and finally obtaining 150 rainfall samples and 450 non-rainfall samples which are consistent with the preset nominal value, which includes the specific steps of:
step S210: and performing threshold segmentation on the precipitation map to obtain a binary image of the precipitation area and the non-precipitation area.
Step S220: and carrying out an average sampling selection method according to the geographical position on the non-precipitation area.
Step S230: and carrying out image erosion and image expansion operation on the precipitation area.
Step S240: and carrying out precipitation event connected domain division on the precipitation region.
Step S250: and (4) carrying out precipitation denomination distribution on the well-divided precipitation areas according to the areas of the precipitation areas. Specifically, precipitation denomination distribution is carried out according to a formula (1) and precipitation samples are selected according to a formula (2).
Step S260: and integrating the precipitation event samples and the non-precipitation event samples into a final precipitation sample set.
For example, as shown in fig. 3, fig. 3(a) is a result graph obtained by a common average random sampling method, and fig. 3(b) is a result graph obtained by applying the adaptive sample selection method based on precipitation events provided by the present invention, it can be known from fig. 3 that the method provided by this embodiment can replace the original common average random sample sampling method to generate a precipitation training sample set having precipitation event characteristics, meanwhile, high representativeness and universality, and the number of precipitation and non-precipitation samples can be adjusted according to a preset nominal value. And inputting the data set into a machine learning framework, so that precipitation inversion can be carried out.
As shown in fig. 4, the present embodiment further provides a system for adaptively selecting a remote sensing precipitation inversion training sample guided by a precipitation event corresponding to the above method, where the system includes:
an obtaining module a1, configured to obtain a precipitation binary map, where the precipitation binary map includes a precipitation area and a non-precipitation area.
And the precipitation connected domain determining module A2 is used for dividing the precipitation connected domain into a plurality of precipitation connected domains by adopting a classification algorithm.
And the quantity determining module A3 is used for obtaining the sampling quantity of any precipitation connected domain according to the number of pixels of the precipitation connected domain, the number of pixels of the precipitation region and the preset precipitation sample quantity.
And the precipitation sample determining module A4 is used for sampling the pixels of the precipitation connected domain according to the sampling number of the precipitation connected domain and the number of the pixels of the precipitation connected domain to obtain the precipitation sample of the precipitation connected domain.
And the non-precipitation sample determination module A5 is used for randomly sampling the pixels of the non-precipitation area to obtain a non-precipitation sample.
And the inversion training sample determining module A6 is used for determining precipitation samples and non-precipitation samples of all precipitation connected domains as precipitation inversion training samples.
As an optional implementation manner, the obtaining module includes:
the acquiring unit is used for acquiring a precipitation situation graph, the pixel value of the precipitation situation graph is the precipitation amount, and the pixel position of the precipitation situation graph is the geographic position corresponding to the pixel.
And the binary image determining unit is used for carrying out threshold segmentation on the precipitation situation image to obtain a precipitation binary image.
As an optional implementation, the adaptive selection system for remote sensing precipitation inversion training samples guided by precipitation events further includes: and the processing module is used for carrying out image mask processing on the precipitation binary image to obtain a processed precipitation area.
As an optional embodiment, the non-precipitation sample determination module comprises:
and the sequence number determining unit is used for labeling each pixel in the non-precipitation area according to the position of each pixel in the non-precipitation area to obtain the pixel sequence number of each pixel in the non-precipitation area.
And the non-precipitation sample determining unit is used for randomly sampling the pixels of the non-precipitation area according to the number of the preset non-precipitation samples and the pixel serial number of each pixel to obtain the non-precipitation samples.
Fig. 5 is a flowchart of a more specific method for adaptively selecting a remote sensing precipitation inversion training sample guided by a precipitation event according to this embodiment, for an actual precipitation map, a binary map is obtained by threshold segmentation, whether a pixel has precipitation is determined according to a pixel value in the binary map, a precipitation area and a non-precipitation area are obtained, the non-precipitation area is subjected to simple average sampling, the precipitation area is subjected to image erosion and expansion, then connected areas which are independent from each other but are internally connected are obtained by division of the precipitation event, then a precipitation sample is obtained by selection according to quantile points, and the precipitation sample and the non-precipitation sample are determined as a sample position set.
The invention has the following advantages:
1. The method divides the whole rainfall into various rainfall areas according to the concept of the connected domain in the graphics, then selects samples in each rainfall area according to the assigned denominations and quantiles, can effectively balance the disadvantages of different methods, and can greatly solve the problem of insufficient representativeness of a random selection method because the samples are distributed to the rainfall events according to the denominations and have no large amount of repetition and redundancy because the samples are used as the basis according to the rainfall events during selection and are distributed to the rainfall events according to the denominations
2. The sample set selected by the method is distributed to all precipitation events according to the precipitation denominations, so that the sample selection is not over centralized and has enough variability, and the precipitation events with different conditions and different characteristics can be effectively represented, so that the universality of the trained model is improved, overfitting is avoided, and the method is superior to a discrimination method based on similarity.
3. The method has the advantages that no more key parameters need to be input, only precipitation denominations need to be input in advance, the parameters only basically influence the volume of the data, and the distribution and selection strategies of the data are not directly influenced, so that the relative numerical distribution control method does not depend on the setting of artificial priori parameters seriously, and the method is a sample selection method with high robustness.
4. Precipitation events can be directionally searched, and sample names are divided and selected according to the precipitation events, so that the representativeness of a sample set can be effectively improved; the prior parameters have little influence on the sample selection result and have enough stability; meanwhile, the sample set has enough variability according to the selection of the precipitation cluster, so that overfitting can be effectively prevented, and the universality of the sample set is improved.
5. The method can apply an image corrosion technology, and effectively reduce the influence of noise points on the final sample selection.
6. The method can select the samples according to quantiles in each precipitation event, so that the finally selected precipitation sample set has universality, is insensitive to the final sample generation result, and can get rid of dependence on prior parameters.
In the present specification, the embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the description of the method part.
The principle and the embodiment of the present invention are explained by applying specific examples, and the above description of the embodiments is only used to help understanding the method and the core idea of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the foregoing, the description is not to be taken in a limiting sense.
Claims (8)
1. A rainfall event guided remote sensing rainfall inversion training sample self-adaptive selection method is characterized by comprising the following steps:
acquiring a precipitation binary image, wherein the precipitation binary image comprises a precipitation area and a non-precipitation area;
carrying out connected domain division on the precipitation areas by adopting a classification algorithm to obtain a plurality of precipitation connected domains;
for any precipitation connected domain, obtaining the sampling number of the precipitation connected domain according to the pixel number of the precipitation connected domain, the pixel number of the precipitation region and the number of preset precipitation samples;
sampling pixels of the precipitation connected domain according to the sampling number of the precipitation connected domain and the number of the pixels of the precipitation connected domain to obtain a precipitation sample of the precipitation connected domain;
randomly sampling pixels of the non-precipitation area to obtain a non-precipitation sample;
Determining precipitation samples and non-precipitation samples of all precipitation connected domains as precipitation inversion training samples;
the sampling number of the precipitation connected domain is obtained according to the number of pixels of the precipitation connected domain, the number of pixels of the precipitation region and the number of preset precipitation samples, and the method specifically comprises the following steps:
according to the formula Mi＝(Ni/Ntotal)*MtotalCalculating the sampling number of the ith precipitation connected domain, wherein MiNumber of samples, M, for the ith precipitation connected domaintotalThe number of the precipitation samples is preset; n is a radical ofiIs the number of pixels of the ith precipitation connected domain, NtotalThe number of pixels of the dewatering area is set;
the pixel of the precipitation connected domain is sampled according to the sampling quantity of the precipitation connected domain and the pixel number of the precipitation connected domain to obtain a precipitation sample of the precipitation connected domain, and the sampling method specifically comprises the following steps:
according to the formula Si＝{Pk|k＝[(t-0.5)/Mi*Ni]，t∈N+，1≤t≤MiObtaining a precipitation sample of the ith precipitation connected domain, wherein SiPrecipitation sample for the ith precipitation connected field, PkThe kth precipitation pixel in the precipitation sample of the ith precipitation connected domain, t is the sample number of the precipitation connected domain, and MiNumber of samples, N, for the ith precipitation connected fieldiThe number of pixels of the ith precipitation connected domain.
2. The method of claim 1, wherein the obtaining of the precipitation event-guided remote sensing precipitation inversion training sample comprises:
Acquiring a precipitation condition graph, wherein the pixel value of the precipitation condition graph is the precipitation amount, and the pixel position of the precipitation condition graph is the geographic position corresponding to the pixel;
and carrying out threshold segmentation on the precipitation condition graph to obtain a precipitation binary graph.
3. The adaptive selection method for the remote sensing rainfall inversion training sample guided by the rainfall event according to claim 1, wherein before the connected domain division of the rainfall region by the classification algorithm to obtain a plurality of rainfall connected domains, the method further comprises:
and carrying out image mask processing on the precipitation binary image to obtain a processed precipitation area.
4. The adaptive selection method for the remote sensing precipitation inversion training sample guided by the precipitation event according to claim 1, wherein the randomly sampling the pixels in the non-precipitation area to obtain the non-precipitation sample specifically comprises:
labeling each pixel in the non-precipitation area according to the position of each pixel in the non-precipitation area to obtain the pixel sequence number of each pixel in the non-precipitation area;
and randomly sampling the pixels of the non-precipitation area according to the number of the preset non-precipitation samples and the pixel serial number of each pixel to obtain the non-precipitation samples.
5. The utility model provides a system is selected in remote sensing precipitation inversion training sample self-adaptation of precipitation event guide which characterized in that includes:
the system comprises an acquisition module, a processing module and a display module, wherein the acquisition module is used for acquiring a precipitation binary image which comprises a precipitation area and a non-precipitation area;
the precipitation connected domain determining module is used for dividing the precipitation area into a plurality of precipitation connected domains by adopting a classification algorithm;
the quantity determining module is used for obtaining the sampling quantity of any precipitation connected domain according to the pixel number of the precipitation connected domain, the pixel number of the precipitation area and the preset precipitation sample quantity;
the rainfall sample determining module is used for sampling the pixels of the rainfall connected domain according to the sampling number of the rainfall connected domain and the number of the pixels of the rainfall connected domain to obtain the rainfall sample of the rainfall connected domain;
the non-precipitation sample determining module is used for randomly sampling the pixels of the non-precipitation area to obtain a non-precipitation sample;
the inversion training sample determining module is used for determining precipitation samples and non-precipitation samples of all precipitation connected domains as precipitation inversion training samples;
the sampling number of the precipitation connected domain is obtained according to the number of pixels of the precipitation connected domain, the number of pixels of the precipitation region and the number of preset precipitation samples, and the method specifically comprises the following steps:
According to the formula Mi＝(Ni/Ntotal)*MtotalCalculating the sampling number of the ith precipitation connected domain, wherein MiNumber of samples, M, for the ith precipitation connected domaintotalPresetting the number of precipitation samples; n is a radical of hydrogeniIs the number of pixels of the ith precipitation connected domain, NtotalThe number of pixels of the dewatering area is set;
the pixel of the precipitation connected domain is sampled according to the sampling quantity of the precipitation connected domain and the pixel number of the precipitation connected domain to obtain a precipitation sample of the precipitation connected domain, and the sampling method specifically comprises the following steps:
according to the formula Si＝{Pk|k＝[(t-0.5)/Mi*Ni]，t∈N+，1≤t≤MiObtaining a precipitation sample of the ith precipitation connected domain, wherein SiPrecipitation sample for the ith precipitation connected field, PkThe kth precipitation pixel in the precipitation sample of the ith precipitation connected domain, t is the sample number of the precipitation connected domain, and MiNumber of samples, N, for the ith precipitation connected fieldiThe number of pixels of the ith precipitation connected domain.
6. The system of claim 5, wherein the acquisition module comprises:
the device comprises an acquisition unit, a calculation unit and a display unit, wherein the acquisition unit is used for acquiring a precipitation situation graph, the pixel value of the precipitation situation graph is the precipitation amount, and the pixel position of the precipitation situation graph is the geographic position corresponding to a pixel;
And the binary image determining unit is used for carrying out threshold segmentation on the precipitation condition image to obtain a precipitation binary image.
7. The adaptive selection system for remote sensing precipitation inversion training sample guided by precipitation event of claim 5, further comprising: and the processing module is used for carrying out image mask processing on the precipitation binary image to obtain a processed precipitation area.
8. The adaptive selection system for remote sensing precipitation inversion training samples guided by precipitation events of claim 5, wherein the non-precipitation sample determination module comprises:
a sequence number determining unit, configured to label each pixel in the non-precipitation area according to a position of each pixel in the non-precipitation area to obtain a pixel sequence number of each pixel in the non-precipitation area;
and the non-precipitation sample determining unit is used for randomly sampling the pixels of the non-precipitation area according to the number of the preset non-precipitation samples and the pixel serial number of each pixel to obtain the non-precipitation samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110655233.5A CN113269805B (en) | 2021-06-11 | 2021-06-11 | Rainfall event guided remote sensing rainfall inversion training sample self-adaptive selection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110655233.5A CN113269805B (en) | 2021-06-11 | 2021-06-11 | Rainfall event guided remote sensing rainfall inversion training sample self-adaptive selection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113269805A CN113269805A (en) | 2021-08-17 |
| CN113269805B true CN113269805B (en) | 2022-06-28 |
Family
ID=77234884
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110655233.5A Active CN113269805B (en) | 2021-06-11 | 2021-06-11 | Rainfall event guided remote sensing rainfall inversion training sample self-adaptive selection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113269805B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115034317A (en) * | 2022-06-17 | 2022-09-09 | 中国平安人寿保险股份有限公司 | Training method and device of policy identification model and policy identification method and device |
| CN116468737B (en) * | 2023-04-10 | 2024-07-26 | 华能澜沧江水电股份有限公司 | Rainfall information acquisition method and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112053371A (en) * | 2020-09-29 | 2020-12-08 | 成都星时代宇航科技有限公司 | Water body extraction method and device in remote sensing image |
| WO2021094753A1 (en) * | 2019-11-11 | 2021-05-20 | University Of Newcastle Upon Tyne | Configuration of global satellite navigation system infrastructure |
-
2021
- 2021-06-11 CN CN202110655233.5A patent/CN113269805B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021094753A1 (en) * | 2019-11-11 | 2021-05-20 | University Of Newcastle Upon Tyne | Configuration of global satellite navigation system infrastructure |
| CN112053371A (en) * | 2020-09-29 | 2020-12-08 | 成都星时代宇航科技有限公司 | Water body extraction method and device in remote sensing image |
Non-Patent Citations (3)
| Title |
|---|
| A Method for Detecting Rainfall From X-Band Marine Radar Images;Yan Zheng et al.;《IEEE Access》;20200122;第08卷;全文 * |
| Himawari-8/AHI红外光谱资料降水信号识别与反演初步应用研究;王根等;《红外与毫米波学报》;20200415(第02期);全文 * |
| KNN不同距离度量对FY-4A/ AGRI红外亮温反演降水的影响研究;王根等;《红外》;20200425(第04期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113269805A (en) | 2021-08-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113496104B (en) | Precipitation prediction correction method and system based on deep learning | |
| CN113269805B (en) | Rainfall event guided remote sensing rainfall inversion training sample self-adaptive selection method | |
| CN111915058B (en) | Flood prediction method and device based on long-time memory network and transfer learning | |
| CN108629452A (en) | A kind of Weather Risk decision-making technique based on multi-mode multi-parameter DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM | |
| CN109636171A (en) | A kind of comprehensive diagnos and risk evaluating method that regional vegetation restores | |
| CN110706213B (en) | Bridge cluster structure damage judgment method based on strain response cumulative distribution function difference | |
| CN107230214B (en) | SAR image water area automatic detection method based on recursive OTSU algorithm | |
| CN111105068B (en) | Numerical mode correction method based on sequential regression learning | |
| CN115423163A (en) | Method and device for predicting short-term flood events of drainage basin and terminal equipment | |
| CN111767517A (en) | BiGRU multi-step prediction method and system applied to flood prediction and storage medium | |
| Chen et al. | Tempee: Temporal-spatial parallel transformer for radar echo extrapolation beyond auto-regression | |
| CN112907970A (en) | Variable lane steering control method based on vehicle queuing length change rate | |
| CN117494871A (en) | Ship track prediction method considering ship interaction influence | |
| CN116861306A (en) | Abnormal electricity utilization detection method based on electricity utilization trend quantification network | |
| CN117975131A (en) | Urban road traffic accident black point prediction method and system based on deep learning and density clustering | |
| CN117892162A (en) | Watershed runoff forecasting method based on nonnegative matrix factorization and machine learning | |
| CN113591608A (en) | High-resolution remote sensing image impervious surface extraction method based on deep learning | |
| CN117253344A (en) | Seawater acidification early warning and forecasting method, system and electronic equipment | |
| CN116884269A (en) | Regional ship traffic flow prediction method and system | |
| Qiu et al. | A short-term precipitation prediction model based on spatiotemporal convolution network and ensemble empirical mode decomposition | |
| Shaw et al. | A methodology for preserving channel flow networks and connectivity patterns in large‐scale distributed hydrological models | |
| Yu et al. | A Deep Learning-Based Multi-model Ensemble Method for Hydrological Forecasting | |
| Wang et al. | Perceptual loss function for generating high-resolution climate data | |
| CN111752976A (en) | Meteorological data processing method and device, computer equipment and readable storage medium | |
| CN111914751A (en) | Image crowd density identification and detection method and system |
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 |