A method, device, computer equipment and storage medium for inverting urban precipitation

Abstract

The present invention provides a method, device, computer equipment and storage medium for inverting urban precipitation, wherein the method for inverting urban precipitation includes: obtaining a precipitation event image of any target monitoring position among multiple target monitoring positions in a target city, wherein the precipitation event image contains a scale feature; inputting the precipitation event image into a pre-trained water accumulation depth monitoring model, so that the water accumulation depth monitoring model determines the water accumulation depth value of the target monitoring position based on the scale feature size; inputting the water accumulation depth value into a pre-trained precipitation monitoring model, so that the precipitation monitoring model inverts the current precipitation at the target monitoring position based on the water accumulation depth value. The present invention realizes real-time prediction and early warning of urban waterlogging and finely depicts the non-uniform spatial distribution of urban rainfall.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of urban flood warning and forecasting, and in particular to an urban precipitation inversion method, device, computer equipment and storage medium. Background Art

[0002] Carrying out scientific and timely waterlogging warning is an urgent need for urban flood control and drainage, and it is also an effective way to protect the lives and property of the people. At present, urban flood prevention mainly relies on manual monitoring, sensor monitoring, remote sensing image monitoring and other technologies, but the warning efficiency is low and it is difficult to support real-time prediction and warning. On the other hand, urban precipitation has a large spatial heterogeneity, but the number of urban rainfall observation stations is small and unevenly distributed, making it difficult to accurately depict the non-uniform spatial distribution of rainfall in the city. Therefore, how to achieve real-time prediction and warning for urban waterlogging and accurately depict the non-uniform spatial distribution of rainfall in the city has become a technical problem that needs to be solved urgently. Summary of the invention

[0003] The present application provides a method, device, computer equipment and storage medium for urban precipitation inversion, which realizes real-time prediction and early warning of urban waterlogging and fine depiction of the uneven spatial distribution of urban rainfall.

[0004] A first aspect of the present invention provides a method for inverting urban precipitation, comprising: obtaining a precipitation event image of any target monitoring location among multiple target monitoring locations in a target city, wherein the precipitation event image contains a scale feature; inputting the precipitation event image into a pre-trained water accumulation depth monitoring model, so that the water accumulation depth monitoring model determines the water accumulation depth value of the target monitoring location based on the size of the scale feature; inputting the water accumulation depth value into a pre-trained precipitation monitoring model, so that the precipitation monitoring model inverts the current precipitation at the target monitoring location based on the water accumulation depth value.

[0005] The urban precipitation inversion method provided by the embodiment of the present invention takes the city as the research area, identifies the image through the neural network deep learning method, obtains the water accumulation depth distribution in the entire city, and thus realizes the real-time forecast of the urban waterlogging situation; at the same time, the present invention further describes the correlation between the water accumulation depth and the precipitation based on the water accumulation depth identified by the image, and deduces the real-time precipitation of the city based on this, thereby providing a new monitoring method for the measurement of local precipitation in the city, and effectively solving the technical problems existing in the traditional classical physical process model. For example, most of the traditional classical physical process models construct linear models such as the least squares method between the two, which are easily affected by other "interferences" in the real world, resulting in the two not being in a strict linear relationship. The water accumulation situation of the entire city can be quickly and accurately grasped, and the urban waterlogging situation can be quickly located and grasped, which plays a vital technical support for the management of the city and the protection of the lives and property of the people, and overcomes the technical problems that the traditional urban rainfall monitoring method is difficult to predict and warn urban waterlogging in real time and finely depict the uneven spatial distribution of rainfall in the city.

[0006] Optionally, it also includes: determining a drainage infrastructure plan for the target monitoring location based on the current precipitation and water accumulation depth values ​​of multiple target monitoring locations obtained by inversion.

[0007] Optionally, the scale feature objects include roadsides and trash cans, and the method further includes: when the water depth value at any target monitoring position is less than or equal to the height of the road shoulder, the road shoulder is used as a scale feature object to determine the current water depth; when the water depth value at any target monitoring position is greater than the height of the road shoulder, the trash can is used as a scale feature object to determine the current water depth.

[0008] Optionally, it also includes: when the water depth value at any target monitoring location is less than a preset height, the current water depth is determined to be a low-threat water depth, and a low-threat warning is issued; when the water depth value at any target monitoring location is greater than a preset height, the current water depth is determined to be a high-threat water depth, and a high-threat warning is issued.

[0009] Optionally, it also includes: when the water depth value at any target monitoring location is less than or equal to the shoulder height, the current water depth is determined to be a light threat water depth; when the water depth value at any target monitoring location is greater than the shoulder height and less than a preset height, the current water depth is determined to be a low threat water depth, and a low threat warning is issued, wherein the preset height is less than the trash can height; when the water depth value at any target monitoring location is greater than a preset height, the current water depth is determined to be a high threat water depth, and a high threat warning is issued.

[0010] Optionally, the water accumulation depth monitoring model is trained through the following steps: obtaining a target event image dataset, wherein the target events include precipitation events and non-precipitation events, and the images in the image dataset contain scale features; annotating the images in the target event image dataset with scale features and water accumulation events; and training the water accumulation depth monitoring model using the annotated target event image dataset until the training results meet preset conditions.

[0011] Optionally, the precipitation monitoring model is trained by the following steps: obtaining historical water accumulation depth value information and corresponding precipitation information of multiple preset locations in the target city; and training the precipitation monitoring model using the water accumulation depth value information and precipitation information until the training results meet preset conditions.

[0012] The second aspect of the present invention provides an urban precipitation inversion device, comprising: a first acquisition module, used to obtain a precipitation event image of any target monitoring location among multiple target monitoring locations in a target city, wherein the precipitation event image contains a scale feature; a first input module, used to input the precipitation event image into a pre-trained water accumulation depth monitoring model, so that the water accumulation depth monitoring model determines the water accumulation depth value of the target monitoring location based on the size of the scale feature; a second input module, used to input the water accumulation depth value into a pre-trained precipitation monitoring model, so that the precipitation monitoring model inverts the current precipitation at the target monitoring location based on the water accumulation depth value.

[0013] The functions performed by each component in the urban precipitation inversion device provided by the present invention have been applied in any method embodiment of the first aspect above, and therefore will not be repeated here.

[0014] The third aspect of the present invention provides a computer device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; the memory is used to store computer programs; the processor is used to implement the steps of the urban precipitation inversion method of the first aspect when executing the program stored in the memory.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the urban precipitation inversion method provided in the first aspect of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 A schematic diagram of a flow chart of a method for inverting urban precipitation provided by an embodiment of the present invention;

[0018] Figure 2 A schematic diagram of the corresponding principle of the urban precipitation inversion method provided by one embodiment of the present invention;

[0019] Figure 3 A schematic diagram of water accumulation corresponding to the urban precipitation inversion method provided by an embodiment of the present invention;

[0020] Figure 4 A schematic diagram of a flow chart of a method for inverting urban precipitation provided by an embodiment of the present invention;

[0021] Figure 5 A schematic diagram of the structure of a method for inverting urban precipitation provided by an embodiment of the present invention;

[0022] Figure 6 A schematic diagram of the structure of an urban precipitation inversion device provided by an embodiment of the present invention;

[0023] Figure 7 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution of the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings of the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the described embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0025] Unless otherwise defined, the technical terms or scientific terms used in the present disclosure should be understood by people with ordinary skills in the field to which the present disclosure belongs. Similar words such as "one", "one" or "the" used in the present disclosure do not indicate a quantitative limitation, but rather indicate the presence of at least one. Similar words such as "include" or "comprise" mean that the element or object appearing before the word covers the element or object listed after the word and its equivalent, without excluding other elements or objects.

[0026] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0027] In view of the technical problems mentioned in the background technology, the embodiment of the present invention provides a method for inverting urban precipitation, such as Figure 1 As shown, the steps of the method include:

[0028] Step S110, obtaining a precipitation event image of any target monitoring location among a plurality of target monitoring locations in a target city, wherein the precipitation event image includes a scale feature object.

[0029] Exemplarily, the target monitoring location can be any location in the target city, such as an urban depression, a road section prone to waterlogging, etc. The acquisition of precipitation event images may include, but is not limited to, videos or images collected from multiple target monitoring locations and multiple time periods through intelligent equipment, such as road monitoring equipment, in actual rainfall scenes. It can also be obtained by photographing with a camera, such as images or videos obtained with a camera, a mobile phone, etc., or other methods known to those skilled in the art, which are not limited here. Specifically, precipitation event images can be obtained by extracting image frames from a video stream using artificial intelligence video analysis technology. Scale feature objects may include, but are not limited to, purposefully established scale equipment, existing urban facilities, such as traffic sign warning poles, signal light poles, isolation piers, etc., or other reference objects known to those skilled in the art that can serve as scale feature objects, which are not limited here.

[0030] Step S120: input the precipitation event image into a pre-trained water accumulation depth monitoring model, so that the water accumulation depth monitoring model determines the water accumulation depth value of the target monitoring location based on the size of the scale feature object.

[0031] Exemplarily, the water accumulation depth monitoring model is pre-trained based on the training method of a neural network. The size of the scale feature object may include but is not limited to the length, width, and height of the scale feature object. The water accumulation depth monitoring model determines the water accumulation depth value of the target monitoring location based on the scale feature object size. Specifically, the water accumulation depth monitoring model obtains the corresponding scale feature pixel value and water accumulation depth pixel value by identifying the input precipitation event image, and determines the water accumulation depth value of the target monitoring location according to the similarity theorem. Specifically, if the actual height of the scale feature object 1 is H, the actual water depth is h, the scale feature pixel height in the precipitation event image 2 (the shooting line of sight and the actual scale feature object may form a certain angle) is Hpx, and the water accumulation depth pixel height in the precipitation event image is hpx. According to Figure 2 As shown, the formula for the actual water depth h can be derived, as shown in formula (1):

[0032]

[0033] Where: h px To monitor the pixel size of the inverted water depth; H px is the pixel size of the monitoring inversion reference object; H is the actual reference object size; h is the actual water depth.

[0034] For example, taking the water depth monitoring model as a convolutional neural network (CNN) model, the actual water depth can be predicted after the CNN model is constructed, as shown in formula (2):

[0035]

[0036] Where: y pred is the predicted value of the convolutional neural network, argmax is the maximum possible function, y is the actual depth of the water, x is the input image, w is the weight parameter of the last layer, b is the corresponding bias parameter, and Y represents all possible prediction results.

[0037] Step S130, inputting the water depth value into a pre-trained precipitation monitoring model, so that the precipitation monitoring model inverts the current precipitation at the target monitoring location based on the water depth value.

[0038] For example, based on the traditional classical physical process model, it can be inferred that there is a close relationship between precipitation and water depth. Figure 3 And formula (3)-(5) as shown:

[0039]

[0040] S=f(h)=P+(RFE)=P+Δο≈P (4)

[0041] P≈f -1 (h) = g(h) (5)

[0042] Where: P is precipitation; R is the surrounding runoff; F is the water infiltration loss; E is the evaporation during heavy rain; S is the water accumulation in the depression during heavy rain; h is the depth of water accumulation in the depression; Δo is the residual of runoff, infiltration and evaporation; f is the functional relationship between regional water accumulation depth and water accumulation; g is the inverse functional relationship between water accumulation depth and water accumulation. The specific inverse functional relationship between water accumulation depth and water accumulation can be obtained by statistically analyzing the historical water accumulation depth data and the corresponding historical water accumulation data.

[0043] Figure 3The figure shows the form of waterlogging in urban precipitation, and formula (3) is the corresponding water balance equation for waterlogging, in which the difference between evaporation, seepage and boundary runoff can be treated as a small error. It can be seen from the equation that the waterlogging depth and the waterlogging volume have a certain functional relationship. When the remaining losses have a small effect on the waterlogging, the waterlogging depth and the precipitation have a close correlation. From the perspective of water balance, it is considered that the feasibility of the theory of deducing precipitation based on the waterlogging depth provides theoretical support for the subsequent inversion of urban precipitation based on the waterlogging depth. The present invention obtains a precipitation monitoring model by pre-training based on a neural network training method, and constructs an inversion model between the waterlogging depth and precipitation, thereby realizing the deduction between the waterlogging depth and the precipitation.

[0044] The urban precipitation inversion method provided by the embodiment of the present invention takes the city as the research area, identifies the image through the neural network deep learning method, obtains the water accumulation depth distribution in the entire city, and thus realizes the real-time forecast of the urban waterlogging situation; at the same time, the present invention further describes the correlation between the water accumulation depth and the precipitation based on the water accumulation depth identified by the image, and deduces the real-time precipitation of the city based on this, thereby providing a new monitoring method for the measurement of local precipitation in the city, and effectively solving the technical problems existing in the traditional classical physical process model. For example, most of the traditional classical physical process models construct linear models such as the least squares method between the two, which are easily affected by other "interferences" in the real world, resulting in the two not being in a strict linear relationship. The water accumulation situation of the entire city can be quickly and accurately grasped, and the urban waterlogging situation can be quickly located and grasped, which plays a vital technical support for the management of the city and the protection of the lives and property of the people, and overcomes the technical problems that the traditional urban rainfall monitoring method is difficult to predict and warn urban waterlogging in real time and finely depict the uneven spatial distribution of rainfall in the city.

[0045] As an optional embodiment of the present invention, it also includes:

[0046] Step S210, determining a drainage infrastructure plan for the target monitoring location based on the current precipitation and water accumulation depth values ​​at multiple target monitoring locations obtained by inversion.

[0047] Exemplarily, the water accumulation depth value at the target monitoring location is correlated with the drainage system at the target monitoring location. Under the same precipitation conditions, the better the drainage system, the smaller the water accumulation depth, and the worse the drainage system, the greater the water accumulation depth. That is, under the same precipitation, the greater the water accumulation depth, the worse the drainage system at the current location, and the drainage infrastructure plan at the current location needs to be improved or further improved. Based on this correspondence, the precipitation and water accumulation depth values ​​at the current multiple target monitoring locations obtained by inversion can provide a reference for the drainage infrastructure plan at any location in the target city. Specifically, a "precipitation-drainage system-water accumulation depth" mathematical model that can achieve good operation is constructed, and this is used as a reference to determine the drainage infrastructure plan for the target monitoring location.

[0048] This embodiment provides a scientific reference for the drainage infrastructure plan of the target monitoring location by comparing the precipitation and water accumulation depth values ​​of multiple target monitoring locations in the target city.

[0049] Exemplarily, the drainage system design parameters of multiple target monitoring locations are obtained, wherein the drainage system design parameters include the drainage per unit time; the water accumulation level of the target monitoring location is predicted according to the precipitation of the current multiple target monitoring locations obtained by inversion and the drainage system drainage per unit time, wherein the water accumulation level can be divided into light water accumulation and heavy water accumulation, or divided into slight water accumulation, light water accumulation, moderate water accumulation and heavy water accumulation. Water accumulation personnel in this field can divide the water accumulation level according to actual conditions, which is not limited here. The determination of the water accumulation level can be determined by the difference between the precipitation and the drainage. When the difference between the two is less than the preset threshold, the current target monitoring location is determined to be light water accumulation; when the difference between the two is greater than the preset threshold, the current target monitoring location is determined to be heavy water accumulation. Based on different water accumulation levels, different levels of warnings are issued. Those skilled in the art can set the size and number of thresholds according to actual conditions to define the water accumulation level, which is not limited here. It should be noted that when the difference between the two is equal to the preset threshold, those skilled in the art can classify the level according to actual needs, which is not limited here.

[0050] This embodiment determines the waterlogging level of the current target monitoring location in the city by comparing the urban precipitation with the drainage volume of the drainage system, quickly locates and grasps the urban waterlogging situation, and issues different levels of warnings based on different waterlogging levels, which provides vital technical support for urban management and protecting the lives and property of the people.

[0051] As an optional embodiment of the present invention, the scale feature includes a road shoulder and a trash can, and the method further includes:

[0052] Step S310, when the water depth value at any target monitoring position is less than or equal to the road shoulder height, the road shoulder is used as a scale feature to determine the current water depth.

[0053] For example, the road shoulder has design standard specifications in urban construction. Generally speaking, the height of the road shoulder is 12 cm and 15 cm, and the maximum does not exceed 20 cm. If it is set according to the standard, the height of the road shoulder is known. Under urban monitoring, the road shoulder is a ruler feature that is very convenient for image recognition. When the water depth value is less than or equal to the road shoulder height, the road shoulder can be used as a ruler feature to determine the current water depth.

[0054] Step S320, when the water depth value at any target monitoring position is greater than the road shoulder height, the trash can is used as a scale feature to determine the current water depth.

[0055] Exemplarily, when the water depth value is greater than the shoulder height, the trash can is used as a scale feature to determine the water depth at the current target monitoring location. According to urban construction requirements, trash cans are set at intervals of 25 to 50 meters in commercial areas, 50 to 80 meters on traffic arteries, and 80 to 100 meters on general roads. The placement density of trash cans is closely related to the flow of vehicles and people, so monitoring with trash cans as scale features meets the requirements for identifying the depth of water. The height of trash cans in cities is generally 70 to 100 cm. The types of trash cans in the same city are generally known, and the relative heights of different trash cans can be basically determined. It should be noted that trash cans are usually placed on the shoulders of the road.

[0056] The urban precipitation inversion method provided by the embodiment of the present invention selects the commonly existing road shoulders and trash cans in the city as scale feature objects from the perspective of the system, and can construct an early warning system of different threat levels according to the degree of threat of the water depth to the city. The two reference objects constitute a reference object system, which can quickly and qualitatively judge the threat degree of urban waterlogging, and can also quickly determine the specific depth of waterlogging. The two complement each other to form a reference object system for monitoring the depth of waterlogging. It can quickly and accurately grasp the waterlogging situation of the entire city under heavy rain, and quickly locate and grasp the waterlogging situation in the city, which plays a vital technical support for the management of the city and the losses when waterlogging occurs. At the same time, in the selection of markers, the traditional method of installing the ruler is abandoned, and the existing facilities in the city are selected as the ruler feature objects to determine the depth of waterlogging. On the one hand, the two selected ruler feature objects are very standard, and the height is basically the same in the same city. Based on the consistency of the reference object height, the determined waterlogging depth is relatively accurate; on the other hand, it saves costs, has a wide range of applications, and is highly scalable.

[0057] As an optional embodiment of the present invention, it also includes:

[0058] Step S410: When the water depth value at any target monitoring position is less than a preset height, the current water depth is determined to be a low threat water depth, and a low threat warning is issued.

[0059] Step S420: When the water depth value at any target monitoring position is greater than a preset height, the current water depth is determined to be a high-threat water depth, and a high-threat warning is issued.

[0060] Exemplarily, the preset height can be determined according to the degree of threat to personal and property from the depth of urban waterlogging, such as the preset height can be determined by the degree of flooding of vehicle tires by the depth of waterlogging. The specific basis or method for determining the preset height can be determined by those skilled in the art according to actual conditions, and is not limited here. The degree of threat to personal and property from the depth of urban waterlogging can be obtained as a known quantity, but is not limited to, from existing research materials, recorded data, or other methods known to those skilled in the art, and is not limited here. Low threat warnings or high threat warnings can be sent to decision makers or mobile clients, etc. through the service platform. The specific sending method and target sending location can be determined by those skilled in the art according to actual application needs, and are not limited here. The specific warning information sent may include but is not limited to the flooded road section or location, the depth of waterlogging, and prompt information, such as driving cautiously on the flooded road section, carefully selecting the flooded road section, please detour the flooded road section, "a certain road section" is now closed due to deep waterlogging, please detour, etc. Those skilled in the art can set the content of the warning information sent according to actual needs, and are not limited here. It should be noted that when the water depth value of any target monitoring location is equal to the preset height, those skilled in the art can classify it according to actual needs, and no limitation is made here. The urban precipitation inversion method provided by the embodiment of the present invention determines the threat level of the current target monitoring location in the city by comparing the water depth with the preset height, and issues different levels of warnings based on different threat levels, which plays a vital technical support for the management of the city and the protection of the lives and property safety of the people.

[0061] As an optional embodiment of the present invention, it also includes:

[0062] Step S510: When the water depth value at any target monitoring position is less than or equal to the road shoulder height, it is determined that the current water depth is a lightly threatening water depth.

[0063] For example, when the shoulder is used as a scale feature, it means that the current waterlogging is not serious, that is, the waterlogging depth is slightly threatening. By comparing it with whether the shoulder is submerged, it can be quickly determined whether the current road waterlogging is a big threat, and the current waterlogging depth can be quickly obtained; when the waterlogging depth value at any target monitoring position is less than or equal to the shoulder height, the current waterlogging depth is determined to be a slightly threatening depth, and a slight threat warning can be issued or no warning can be issued, which can be determined according to the actual situation and is not limited here.

[0064] Step S520: When the water depth value at any target monitoring position is greater than the shoulder height and less than a preset height, the current water depth is determined to be a low threat water depth, and a low threat warning is issued, wherein the preset height is less than the trash can height.

[0065] Exemplarily, the preset height can be determined based on the degree of threat posed by the depth of urban waterlogging to personal and property, such as the degree of submergence of vehicle tires by the depth of waterlogging. The specific basis or method for determining the preset height can be determined by those skilled in the art according to actual conditions, and is not limited here. The degree of threat posed by the depth of urban waterlogging to personal and property as a known quantity can be obtained, but not limited to, from existing research materials, recorded data, or other methods known to those skilled in the art, and is not limited here. When the waterlogging depth value is greater than the shoulder height, the waterlogging depth cannot be determined by the shoulder. At this time, the trash can is used as a scale feature to judge the waterlogging depth at the current target monitoring location, which means that the waterlogging depth is relatively deep and the disaster to the city is already serious. When the waterlogging depth value at any target monitoring location is greater than the shoulder height and less than the preset height, the current waterlogging depth is determined to be a low-threat depth, and a low-threat warning is issued.

[0066] Step S530: When the water depth value at any target monitoring position is greater than a preset height, the current water depth is determined to be a high-threat water depth, and a high-threat warning is issued.

[0067] Step S530 of this embodiment is similar to the description of the corresponding part above and will not be repeated here.

[0068] The urban precipitation inversion method provided by the embodiment of the present invention determines the threat level of the current target monitoring location in the city by comparing the water accumulation depth with the scale feature and the preset height, and issues different levels of warnings based on different threat levels, which plays a vital technical support for the management of the city and the protection of people's lives and property safety.

[0069] As an optional implementation of the present invention, the water depth monitoring model is trained by the following steps:

[0070] Step S610, obtaining a target event image dataset, wherein the target event includes precipitation events and non-precipitation events, and the images in the image dataset contain scale features.

[0071] Exemplarily, the target event image may be, but is not limited to, from a video of an actual rainfall scene collected from multiple scenes and multiple time periods by intelligent equipment, such as road monitoring equipment, or may be obtained by photographing a camera, such as an image or video obtained by a camera, a mobile phone, etc., or other methods known to those skilled in the art, which are not limited here. Specifically, the target event image data set can be obtained by extracting image frames from a video stream using artificial intelligence video analysis technology. The target event includes precipitation events and non-precipitation events, and the target event image data set includes a precipitation event image data set and a non-precipitation event image data set, wherein the non-precipitation event image data set is used as a reference image data set, and needs to include the same scale feature as the precipitation event image data set to avoid the influence of various factors (sunlight, camera angle, noise) in different images on the consistency of image content. The scale feature may include, but is not limited to, corresponding scale equipment established for monitoring, existing urban facilities, such as traffic sign warning poles, signal light poles, isolation piers, etc., or other reference objects known to those skilled in the art that can be used as scale features, such as pedestrians, vehicles, etc.

[0072] Step S620: marking the images in the target event image data set with scale features and water accumulation events.

[0073] For example, the images in the target event image dataset are annotated with scale features, such as road shoulders, trash cans, etc.; waterlogging events may include but are not limited to waterlogging shapes and waterlogging depths. Specifically, LabelImg can be used to annotate images to obtain the corresponding annotation file, i.e., the annotated target event image dataset.

[0074] Step S630: training the water depth monitoring model using the labeled target event image dataset until the training result meets the preset conditions.

[0075] For example, the preset conditions may include but are not limited to satisfying the preset recognition precision and the preset recognition accuracy. Those skilled in the art may select them according to actual needs and are not limited here. For example, the scale feature objects are shoulders and trash cans, the target event image data set is obtained through road monitoring equipment, the water depth monitoring model is CNN, and the preset condition is the preset recognition accuracy. The specific process of identifying the water depth in the monitoring video is as follows: Figure 4 shown.

[0076] CNN is a hot topic in the current image recognition field. It is a mature technology with many advantages. When the input network is a multi-dimensional image, it can avoid weight explosion and achieve better performance. At the same time, the video image can be directly used as the input of the network, avoiding the complex feature extraction and data reconstruction process in traditional recognition methods. CNN mainly includes 5 layers of structure: input layer, convolution layer, pooling layer, fully connected layer and output layer. The convolution layer, pooling layer and fully connected layer can have a multi-layer structure. The training process of CNN is mainly to learn the network parameters such as the convolution kernel parameters and the inter-layer connection weights in the convolution layer. At the same time, the back propagation algorithm is added to correct the simulation accuracy of the model. The prediction process is mainly based on the input image and the trained network parameters for prediction. The hierarchical architecture of CNN is shown below:

[0077] a. Input layer:

[0078] The first layer of the convolutional neural network is the input layer. The data set of the input layer is the image data recorded in the past precipitation events in the target city recorded by the monitoring. The precipitation events and non-precipitation images recorded in the past monitoring are used as data sets respectively, among which the non-precipitation images are used as comparison images. Specifically, the precipitation images are used as training sets at a ratio of 80%, and 20% of the data are used as verification sets. The CNN model is trained and verified. The specific distribution ratio of the training set and the verification set can be set by technicians in this field according to actual conditions, and there is no restriction here. Since different monitoring has different angles, a video image recorded by a monitoring can be used as a data set when training the CNN model, and the monitoring must at least include the shoulder image, and for areas with deep water accumulation, it must also include images of urban garbage cans, etc. Considering the universality of shoulders and garbage cans in urban construction, this type of data is relatively common.

[0079] b. Convolutional layer:

[0080] The convolution layer uses a learnable convolution kernel to perform a convolution operation on the image of the input layer. The convolution kernel is generally a 3×3 convolution kernel, which transforms the surveillance video image into a convolution image layer with different features. Then, through an activation function, the output feature map can be obtained. Each output feature map can combine the values ​​of multiple convolution feature maps. The specific formulas are shown in formulas (6) and (7).

[0081]

[0082]

[0083] In formula (6)-(7): It is called the net activation of the jth channel of the lth convolutional layer, which is obtained by filtering the feature map output by the previous layer. After convolution summation and bias, is the output of the jth channel of the convolutional layer l. f is called the activation function, and functions such as sigmoid and tanh can usually be used. j Indicates the calculation A subset of the input feature maps, is the convolution kernel matrix, It is the bias of the feature map after convolution. For an output feature map Each input feature map The corresponding convolution kernel Might be different, * is the convolution symbol.

[0084] c. Pooling layer:

[0085] Since the images of surveillance videos are generally large and have three channels, the network parameters generated when the initial data is used for CNN recognition are very large. In order to reduce the huge amount of input data and the amount of parameters generated, it is necessary to further compress the feature map in the convolution layer through the pooling layer, retain the most significant features in the convolution layer, and discard other features, thereby reducing the amount of network data. The pooling layer processes each input feature map. The specific formulas are shown in formulas (8) and (9):

[0086]

[0087]

[0088] In formula (8)-(9): The net activation of the jth channel of the lth layer is called the pooling layer, which is obtained by filtering the feature map output by the previous layer After pooling weighting and biasing, β is the weight coefficient of the pooling layer. It is the bias term of the downsampling layer. The symbol down represents the downsampling function, which is obtained by The image is divided into multiple non-overlapping image blocks through the sliding window method, and then the pixels in each image block are summed, averaged or maximized, so that the output image is reduced in two dimensions, thereby reducing the amount of calculation.

[0089] d. Fully connected layer:

[0090] In the fully connected network, the feature maps of all the two-dimensional images processed by the pooling layer are concatenated into one-dimensional features as the input layer of the fully connected network. The output of the fully connected layer can be obtained by weighted summation of the input and the response of the activation function, as shown in formulas (10) and (11):

[0091]

[0092] k l =ω lx l-1 +b l (11)

[0093] In formula (5)-(6): where k l It is called the net activation of the fully connected layer l, which is composed of the output feature map x of the previous layer l-1 After weighting and biasing, ω l is the weight coefficient of the fully connected network, b l is the bias term of the fully connected layer l.

[0094] e. Output layer:

[0095] Finally, the calculation results of the fully connected layer are accumulated to obtain the image recognition results, and the relative heights of the water depth and the road shoulder (or trash can) and other reference objects under monitoring are determined and verified with the given values. The CNN model is trained according to the back propagation algorithm, as shown in formula (12), and finally a model with higher accuracy is obtained.

[0096]

[0097] Where: J is the loss function; y n is the category to which the nth sample belongs; x n is the feature vector of the nth sample; w is the weight parameter of the last layer; b is the corresponding bias parameter. Specifically, Figure 4 As shown, when the output result differs greatly from the given training value, the convolution layer is returned to re-adjust the parameters to build the model until the output result differs from the given training value by a small error and meets the preset accuracy (such as within 5%). At this time, the loop can be ended to output the pixel value of the scale feature object and the pixel value of the water depth for depth recognition, so that the water depth monitoring model can infer the actual water depth of the city based on the similarity principle.

[0098] As an optional implementation of the present invention, the precipitation monitoring model is trained by the following steps:

[0099] Step S710, obtaining historical water depth value information and corresponding precipitation information of multiple preset locations in the target city.

[0100] Exemplarily, the water depth value information and the corresponding precipitation information can be obtained, but not limited to, from meteorological monitoring stations, documents recording the hydrological information of the target city, or other methods known to those skilled in the art, which are not limited here.

[0101] Step S720: training a precipitation monitoring model using the water depth information and precipitation information until the training result meets a preset condition.

[0102] For example, the precipitation monitoring model can be constructed by, but is not limited to, a BP back propagation (BP) neural network. Taking the BP neural network as an example, the specific construction network is as follows: Figure 4 , 5 As shown, it includes input layer, hidden layer and output layer. Figure 5 Taking hidden layer 1 and hidden layer 2 as an example, the input layer mainly includes the water depth, and the output layer is the precipitation at the target monitoring location.

[0103] Figure 6 An embodiment of the present invention provides a device for inverting urban precipitation. The device for inverting urban precipitation in this embodiment includes:

[0104] The first acquisition module 810 is used to acquire a precipitation event image of any target monitoring position among multiple target monitoring positions in the target city, wherein the precipitation event image contains a scale feature. For details, please refer to the description of the corresponding part in the above embodiment, which will not be repeated here.

[0105] The first input module 820 is used to input the precipitation event image into a pre-trained water depth monitoring model, so that the water depth monitoring model determines the water depth value of the target monitoring position based on the scale feature size. For details, please refer to the description of the corresponding part in the above embodiment, which will not be repeated here.

[0106] The second input module 830 is used to input the water depth value into a pre-trained precipitation monitoring model, so that the precipitation monitoring model can invert the current precipitation at the target monitoring location based on the water depth value. For details, please refer to the description of the corresponding part in the above embodiment, which will not be repeated here.

[0107] As an optional implementation device of the present invention, it also includes:

[0108] The first determination module is used to determine the drainage infrastructure plan of the target monitoring location according to the precipitation and water accumulation depth values ​​of the current multiple target monitoring locations obtained by inversion. For details, please refer to the description of the corresponding part of the above embodiment, which will not be repeated here.

[0109] As an optional implementation device of the present invention, the scale feature includes a road shoulder and a trash can, and further includes:

[0110] The first determination module is used to determine the current water depth when the water depth value of any target monitoring position is less than or equal to the road shoulder height, using the road shoulder as a scale feature. For details, please refer to the description of the corresponding part of the above embodiment, which will not be repeated here.

[0111] The second determination module is used to determine the current water depth when the water depth value of any target monitoring position is greater than the road shoulder height, using the trash can as a scale feature. For details, please refer to the description of the corresponding part of the above embodiment, which will not be repeated here.

[0112] As an optional implementation device of the present invention, it also includes:

[0113] The third determination module is used to determine that the current water depth is a low threat water depth and issue a low threat warning when the water depth value of any target monitoring position is less than a preset height. For details, please refer to the description of the corresponding part of the above embodiment, which will not be repeated here.

[0114] The fourth determination module is used to determine that the current water depth is a high threat water depth and issue a high threat warning when the water depth value of any target monitoring position is greater than a preset height. For details, please refer to the description of the corresponding part in the above embodiment, which will not be repeated here.

[0115] As an optional implementation device of the present invention, it also includes:

[0116] The fifth determination module is used to determine that the current water depth is a lightly threatening water depth when the water depth value at any target monitoring position is less than or equal to the road shoulder height. For details, please refer to the description of the corresponding part in the above embodiment, which will not be repeated here.

[0117] The sixth determination module is used to determine that the current water depth is a low threat water depth and issue a low threat warning when the water depth value at any target monitoring position is greater than the shoulder height and less than a preset height, wherein the preset height is less than the trash can height. For details, please refer to the description of the corresponding part in the above embodiment, which will not be repeated here.

[0118] The seventh determination module is used to determine that the current water depth is a high threat water depth and issue a high threat warning when the water depth value of any target monitoring position is greater than a preset height. For details, please refer to the description of the corresponding part in the above embodiment, which will not be repeated here.

[0119] As an optional implementation device of the present invention, the water depth monitoring model is trained by the following device:

[0120] The second acquisition module is used to acquire a target event image data set, wherein the target event includes a precipitation event and a non-precipitation event, and the image in the image data set includes a scale feature. For details, please refer to the description of the corresponding part in the above embodiment, which will not be repeated here.

[0121] The first labeling module is used to label the images in the target event image data set with scale features and water accumulation events. For details, please refer to the description of the corresponding parts in the above embodiment, which will not be repeated here.

[0122] The first training module is used to train the water depth monitoring model using the labeled target event image data set until the training result meets the preset conditions. For details, please refer to the description of the corresponding part of the above embodiment, which will not be repeated here.

[0123] As an optional implementation device of the present invention, the precipitation monitoring model is trained by the following device:

[0124] The third acquisition module is used to obtain historical water depth value information and corresponding precipitation information of multiple preset locations in the target city. For details, please refer to the description of the corresponding part in the above embodiment, which will not be repeated here.

[0125] The second training module is used to train the precipitation monitoring model using the water depth value information and the precipitation information until the training result meets the preset conditions. For details, please refer to the description of the corresponding part in the above embodiment, which will not be repeated here.

[0126] An embodiment of the present invention provides a computer device, such as Figure 7 As shown, the device includes one or more processors 3010 and a memory 3020, and the memory 3020 includes a persistent memory, a volatile memory, and a hard disk. Figure 7 A processor 3010 is taken as an example. The device may also include: an input device 3030 and an output device 3040.

[0127] The processor 3010, the memory 3020, the input device 3030 and the output device 3040 may be connected via a bus or other means. Figure 7 The example of connecting through bus is taken in the following.

[0128] The processor 3010 may include, but is not limited to, a central processing unit (CPU) and a graphics processing unit (GPU). The processor 3010 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, or a combination of the above-mentioned chips. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The memory 3020 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the urban precipitation inversion device, etc. In addition, the memory 3020 may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory 3020 may optionally include a memory remotely arranged relative to the processor 3010, and these remote memories may be connected to the urban precipitation inversion device via a network. The input device 3030 may receive a calculation request (or other digital or character information) input by a user, and generate a key signal input related to the urban precipitation inversion device. The output device 3040 may include a display device such as a display screen to output the calculation results.

[0129] The embodiment of the present invention provides a computer-readable storage medium, which stores computer instructions, and the computer storage medium stores computer executable instructions, which can execute the urban precipitation inversion method in any of the above method embodiments. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory (Flash Memory), a hard disk (HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above types of memory.

[0130] The logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be embodied in any computer-readable storage medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, apparatus, or device and execute instructions), or in conjunction with such instruction execution system, apparatus, or device. For purposes of this specification, "computer-readable storage medium" may be any device that can contain, store, communicate, propagate, or transmit a program for use by an instruction execution system, apparatus, or device, or in conjunction with such instruction execution system, apparatus, or device. More specific examples of computer-readable storage media (a non-exhaustive list) include the following: an electrical connection with one or more wirings (electronic devices), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable storage medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0131] It should be understood that the various parts of the present disclosure can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0132] In the description of this specification, the description with reference to the terms "this embodiment", "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless there is any contradiction. In the description of the present disclosure, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0133] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from these are still within the scope of protection of the invention.

Claims

1. A method for inverting urban precipitation, characterized in that: include: Acquire a precipitation event image of any target monitoring location among a plurality of target monitoring locations in a target city, wherein the precipitation event image includes a scale feature object; The precipitation event image is input into a pre-trained water accumulation depth monitoring model, so that the water accumulation depth monitoring model determines the water accumulation depth value of the target monitoring position based on the size of the scale feature object; wherein the water accumulation depth monitoring model obtains the corresponding scale feature pixel value and water accumulation depth pixel value by identifying the input precipitation event image, and determines the water accumulation depth value of the target monitoring position according to the similarity theorem; assuming that the actual height of the scale feature object 1 is H, the actual water depth is h, the scale feature pixel height in the precipitation event image 2 is Hpx, and the water accumulation depth pixel height in the precipitation event image is hpx, the formula for the actual water accumulation depth value h is as follows: Where: h px To monitor the pixel size of the inverted water depth; H px is the pixel size of the monitoring inversion reference object; H is the actual reference object size; h is the actual water depth; The water depth value is input into a pre-trained precipitation monitoring model, so that the precipitation monitoring model inverts the current precipitation at the target monitoring location based on the water depth value; wherein the relationship between the current precipitation and the water depth value is as follows: S=f(h)=P+(RFE)=P+Δο≈P P≈f -1 (h)=g(h) Where: P is precipitation; R is the surrounding runoff; F is the water infiltration loss; E is the evaporation during heavy rain; S is the water accumulation in the depression during heavy rain; h is the depth of water accumulation in the depression; Δo is the residual of the defined runoff, infiltration and evaporation; f is the functional relationship between the regional water accumulation depth and water accumulation; g is the inverse functional relationship between the water accumulation depth and water accumulation; The precipitation monitoring model is trained by the following steps: Obtain historical water depth value information and corresponding precipitation information at multiple preset locations in the target city; The precipitation monitoring model is trained using the water accumulation depth value information and precipitation information until the training result meets the preset conditions.

2. The urban precipitation inversion method according to claim 1, characterized in that: The method further comprises: The drainage infrastructure plan for the target monitoring location is determined based on the current precipitation and water accumulation depth values ​​at multiple target monitoring locations obtained by inversion.

3. The urban precipitation inversion method according to claim 1, characterized in that: The scale feature includes a road shoulder and a trash can, and the method further includes: When the water depth value at any target monitoring location is less than or equal to the shoulder height, the shoulder is used as a scale feature to determine the current water depth; When the water depth value at any target monitoring location is greater than the road shoulder height, the trash can is used as a scale feature to determine the current water depth.

4. The urban precipitation inversion method according to claim 1 or 3, characterized in that: The method further comprises: When the water depth value at any target monitoring location is less than the preset height, the current water depth is determined to be a low threat depth and a low threat warning is issued; When the water depth value at any target monitoring location is greater than the preset height, the current water depth is determined to be a high-threat water depth and a high-threat warning is issued.

5. The urban precipitation inversion method according to claim 3 is characterized in that: Also includes: When the water depth value at any target monitoring location is less than or equal to the road shoulder height, the current water depth is determined to be a light threat depth; When the water depth value at any target monitoring location is greater than the shoulder height and less than a preset height, the current water depth is determined to be a low threat depth and a low threat warning is issued, wherein the preset height is less than the trash can height; When the water depth value at any target monitoring location is greater than the preset height, the current water depth is determined to be a high-threat water depth and a high-threat warning is issued.

6. The urban precipitation inversion method according to claim 1, characterized in that: The water depth monitoring model is trained by the following steps: Acquire a target event image data set, wherein the target event includes a precipitation event and a non-precipitation event, and the images in the image data set include scale features; Marking the images in the target event image dataset with scale features and water accumulation events; The water depth monitoring model is trained using the labeled target event image dataset until the training results meet the preset conditions.

7. An urban precipitation inversion device, characterized in that: include: A first acquisition module is used to acquire a precipitation event image of any target monitoring location among a plurality of target monitoring locations in a target city, wherein the precipitation event image contains a scale feature object; The first input module is used to input the precipitation event image into a pre-trained water accumulation depth monitoring model, so that the water accumulation depth monitoring model determines the water accumulation depth value of the target monitoring position based on the size of the scale feature object; wherein the water accumulation depth monitoring model obtains the corresponding scale feature pixel value and water accumulation depth pixel value by identifying the input precipitation event image, and determines the water accumulation depth value of the target monitoring position according to the similarity theorem; assuming that the actual height of the scale feature object 1 is H, the actual water depth is h, the scale feature pixel height in the precipitation event image 2 is Hpx, and the water accumulation depth pixel height in the precipitation event image is hpx, the formula for the actual water accumulation depth value h is as follows: Where: h px To monitor the pixel size of the inverted water depth; H px is the pixel size of the monitoring inversion reference object; H is the actual reference object size; h is the actual water depth; The second input module is used to input the water depth value into a pre-trained precipitation monitoring model, so that the precipitation monitoring model inverts the current precipitation at the target monitoring location based on the water depth value; wherein the relationship between the current precipitation and the water depth value is as follows: S=f(h)=P+(RFE)=P+Δο≈P P≈f -1 (h)=g(h) Where: P is precipitation; R is the surrounding runoff; F is the water infiltration loss; E is the evaporation during heavy rain; S is the water accumulation in the depression during heavy rain; h is the depth of water accumulation in the depression; Δo is the residual of the defined runoff, infiltration and evaporation; f is the functional relationship between the regional water accumulation depth and water accumulation; g is the inverse functional relationship between the water accumulation depth and water accumulation; The precipitation monitoring model is trained by the following devices: The third acquisition module is used to obtain historical water depth value information and corresponding precipitation information at multiple preset locations in the target city; The second training module is used to train the precipitation monitoring model using the water accumulation depth value information and the precipitation information until the training result meets the preset conditions.

8. A computer device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor is used to implement the steps of the urban precipitation inversion method described in any one of claims 1 to 6 when executing the program stored in the memory.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the urban precipitation inversion method according to any one of claims 1 to 6 are implemented.

Images