CN115908574B - River dike encroaching, positioning and pushing method and system based on unmanned aerial vehicle monitoring - Google Patents

River dike encroaching, positioning and pushing method and system based on unmanned aerial vehicle monitoring Download PDF

Info

Publication number
CN115908574B
CN115908574B CN202310173901.XA CN202310173901A CN115908574B CN 115908574 B CN115908574 B CN 115908574B CN 202310173901 A CN202310173901 A CN 202310173901A CN 115908574 B CN115908574 B CN 115908574B
Authority
CN
China
Prior art keywords
river
encroaching
data
river levee
positioning
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
Application number
CN202310173901.XA
Other languages
Chinese (zh)
Other versions
CN115908574A (en
Inventor
杨翰翔
杨德润
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Lianhe Intelligent Technology Co ltd
Original Assignee
Shenzhen Lianhe Intelligent Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shenzhen Lianhe Intelligent Technology Co ltd filed Critical Shenzhen Lianhe Intelligent Technology Co ltd
Priority to CN202310173901.XA priority Critical patent/CN115908574B/en
Publication of CN115908574A publication Critical patent/CN115908574A/en
Application granted granted Critical
Publication of CN115908574B publication Critical patent/CN115908574B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Alarm Systems (AREA)

Abstract

The utility model provides a river levee encroaching location pushing method and system based on unmanned aerial vehicle monitoring, through extracting the river levee encroaching event data of the river levee district monitoring flow data of the appointed river levee district monitored by the target unmanned aerial vehicle in the appointed monitoring period, and combine the river levee encroaching event data analysis output of river levee district monitoring flow data to appoint the river levee encroaching event directed graph of river levee district in the appointed monitoring period, thereby carry out the analysis to the corresponding area characteristic data of river levee encroaching event directed graph and further obtain corresponding river levee encroaching location root cause information, then confirm with the target river levee encroaching location root cause that obtains the river levee district structural feature of appointed river levee district and push to corresponding river levee management server, from this, take into account the event linking characteristic data between the river levee encroaching events and carry out the determination of river levee encroaching location root cause, and further combine the river levee district structural feature to carry out the screening, can improve and refine the area encroaching location accuracy.

Description

River dike encroaching, positioning and pushing method and system based on unmanned aerial vehicle monitoring
Technical Field
The application relates to the technical field of river levee monitoring and unmanned aerial vehicles, in particular to a river levee encroaching, positioning and pushing method and system based on unmanned aerial vehicle monitoring.
Background
In the supervision and treatment process of the river levee, the attack of the abnormal river levee is a key treatment link. Therefore, how to find the encroachment problem of various river banks in time is the current direction to be studied urgently. In the related art, the analysis and classification of the river levee encroachment events are generally performed singly when the positioning of the river levee monitoring encroachment root causes are performed, and the event engagement between the river levee encroachment events and the structural characteristics of the river levee area are not considered, so that the accuracy of the river levee encroachment positioning is not high.
Disclosure of Invention
In order to at least overcome the defects in the prior art, the purpose of the application is to provide a river levee encroaching positioning pushing method and system based on unmanned aerial vehicle monitoring.
In a first aspect, the present application provides a method for positioning and pushing a river levee encroachment based on unmanned aerial vehicle monitoring, which is applied to a river levee encroachment positioning and pushing system based on unmanned aerial vehicle monitoring, the method includes:
extracting river bank encroaching event data of river bank monitoring flow data of a designated river bank area monitored by a target unmanned aerial vehicle in a designated monitoring period, and analyzing and outputting a river bank encroaching event directed graph of the designated river bank area in the designated monitoring period by combining the river bank encroaching event data of the river bank monitoring flow data, wherein the river bank encroaching event directed graph comprises a plurality of river bank encroaching events, different river bank encroaching events are configured with directed graph associated information through event directed relations, and the event directed relations represent event linking characteristics among different river bank encroaching events;
And inputting the graph rolling characteristic data corresponding to the river levee encroaching event directed graph into a river levee monitoring encroaching positioning network meeting the model convergence requirement, and generating river levee encroaching positioning root cause information corresponding to the river levee encroaching event directed graph.
And acquiring the river dike region structural characteristics of the designated river dike region, determining a target river dike encroaching locating root cause corresponding to the river dike region structural characteristics from the river dike encroaching locating root cause information, and pushing the target river dike encroaching locating root cause to a corresponding river dike management server.
Exemplary, the extracting the river levee intrusion event data of the river levee region monitoring flow data of the designated river levee region monitored by the target unmanned aerial vehicle in the designated monitoring period is realized by the following steps:
extracting river levee region monitoring stream data and an associated river levee monitoring stream data sequence, wherein the associated river levee monitoring stream data sequence comprises associated river levee monitoring stream data corresponding to one or more river levee encroaching event data;
performing video motion feature analysis on the river levee region monitoring stream data to generate a first video motion feature, and performing video motion feature analysis on the related river levee monitoring stream data to generate a second video motion feature of the related river levee monitoring stream data;
Performing encroaching event estimation on the river levee region monitoring flow data through a target estimation strategy by combining the first video motion characteristic and the second video motion characteristic, and generating estimation information of the river levee region monitoring flow data for each river levee encroaching event data under each target estimation strategy;
and generating river levee encroaching event data of the river levee region monitoring flow data by combining the estimated information of the river levee region monitoring flow data for each river levee encroaching event data under each target estimated strategy.
Illustratively, the first video motion feature and the second video motion feature are feature vectors obtained by performing video motion feature analysis on a video motion feature analysis model which is completed by prior training;
the method comprises the steps of carrying out video motion feature analysis on the river levee region monitoring stream data to generate a first video motion feature, carrying out video motion feature analysis on the related river levee monitoring stream data, and carrying out model convergence optimization on a video motion feature analysis model before generating a second video motion feature of the related river levee monitoring stream data, wherein the method is realized by the following steps:
acquiring a river levee region monitoring data sequence to be learned, wherein the river levee region monitoring data sequence to be learned comprises fuzzy river levee region monitoring data corresponding to reference river levee encroaching event data and example river levee region monitoring data corresponding to one or more river levee encroaching event data;
Performing video motion characteristic analysis on the fuzzy river levee region monitoring data through a video motion characteristic analysis model to be trained to generate characteristic vectors of the fuzzy river levee region monitoring data, and performing video motion characteristic analysis on the example river levee region monitoring data through the video motion characteristic analysis model to be trained to generate example river levee region monitoring data characteristic vectors;
combining the characteristic vector of the fuzzy river levee region monitoring data and the characteristic vector of the example river levee region monitoring data, deciding the fuzzy river levee region monitoring data according to the video motion characteristic analysis model to be trained and a target estimation strategy, and generating a river levee encroaching event data sequence of the fuzzy river levee region monitoring data, wherein the river levee encroaching event data sequence comprises target river levee encroaching event data corresponding to each target estimation strategy;
and combining the target river levee encroaching event data corresponding to each target estimation strategy and the reference river levee encroaching event data, updating traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, and generating a video motion feature analysis model for deployment.
Illustratively, the combining the feature vector of the fuzzy river levee region monitoring data with the feature vector of the example river levee region monitoring data makes a decision on the fuzzy river levee region monitoring data according to the video motion feature analysis model to be trained and a target estimation strategy, and generating a river levee encroaching event data sequence of the fuzzy river levee region monitoring data includes:
according to the video motion characteristic analysis model to be trained, determining association parameters between characteristic vectors of the fuzzy river levee region monitoring data and characteristic vectors of the example river levee region monitoring data in combination with a target estimation strategy, and generating estimated association parameters between the fuzzy river levee region monitoring data and the example river levee region monitoring data under each target estimation strategy;
determining target river levee encroachment event data of the fuzzy river levee region monitoring data under each target estimation strategy by estimating the associated parameters;
and generating a river bank encroaching event data sequence of the fuzzy river bank area monitoring data by combining target river bank encroaching event data of the fuzzy river bank area monitoring data under each target estimation strategy.
The method for generating the video motion feature analysis model for deployment by combining the target river levee encroaching event data and the reference river levee encroaching event data corresponding to each target estimation strategy, and updating the traversal model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, includes:
Determining a first LOSS value between target river levee encroachment event data corresponding to each target estimation strategy and the reference river levee encroachment event data;
fusing the first LOSS value to generate a fused LOSS value;
combining the fusion LOSS value, updating traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, and generating a video motion feature analysis model for deployment;
and combining the fusion LOSS value, updating traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, and before generating the video motion feature analysis model for deployment, the method further comprises the following steps:
performing monitoring priori frame expansion on the fuzzy river levee region monitoring data to generate expanded river levee region monitoring data;
estimating the encroaching event of the monitoring data of the extended river dike region according to a river dike encroaching decision branch of the video motion characteristic analysis model to be trained by combining the monitoring data of the extended river dike region, and generating estimated river dike encroaching event data of the monitoring data of the extended river dike region;
acquiring river levee encroaching event priori labeling data corresponding to the extended river levee region monitoring data, and determining a second LOSS value between the estimated river levee encroaching event data and the river levee encroaching event priori labeling data;
And combining the fusion LOSS value, updating traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, and generating a video motion feature analysis model for deployment, comprising:
and combining the fusion LOSS value and the second LOSS value, updating traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, and generating a video motion feature analysis model for deployment.
Illustratively, the extended river levee region monitoring data includes extended river levee region monitoring data obtained by extending the blurred river levee region monitoring data, the river levee encroaching decision branch includes a first river levee encroaching decision branch and a second river levee encroaching decision branch, and the estimated river levee encroaching event data includes first estimated river levee encroaching event data and second estimated river levee encroaching event data;
the method for estimating the intrusion event of the extended river levee region monitoring data according to the river levee intrusion decision branch of the video motion characteristic analysis model to be trained by combining the extended river levee region monitoring data, and generating estimated river levee intrusion event data of the extended river levee region monitoring data comprises the following steps:
Estimating an encroaching event through a first river levee encroaching decision branch of a video motion characteristic analysis model to be trained by combining the expanded river levee region monitoring data, and generating first estimated river levee encroaching event data of target expanded river levee region monitoring data, wherein the first estimated river levee encroaching event data is used for indicating that the target expanded river levee region monitoring data is the river levee encroaching event data of the fuzzy river levee region monitoring data, and the target expanded river levee region monitoring data is river levee region monitoring data related to the fuzzy river levee region monitoring data in the expanded river levee region monitoring data;
and carrying out encroaching event estimation through a second river levee encroaching decision branch of the video motion characteristic analysis model to be trained by combining the expanded river levee region monitoring data, and generating second estimated river levee encroaching event data of the expanded river levee region monitoring data, wherein the second estimated river levee encroaching event data is used for indicating that the expanded river levee region monitoring data corresponds to the expanded river levee encroaching event data.
Exemplary, in combination with the extended river levee region monitoring data, the estimating the encroaching event of the extended river levee region monitoring data according to a river levee encroaching decision branch of the video motion feature analysis model to be trained, and generating estimated river levee encroaching event data of the extended river levee region monitoring data includes:
Performing video motion characteristic analysis on the extended river levee region monitoring data according to the characteristic coding branches of the video motion characteristic analysis model to be trained, and generating an extended river levee region monitoring data characteristic vector;
estimating the encroaching event of the monitoring data of the extended river levee region according to a river levee encroaching decision branch of the video motion characteristic analysis model to be trained by combining the characteristic vector of the monitoring data of the extended river levee region, and generating estimated river levee encroaching event data of the monitoring data of the extended river levee region;
the video motion characteristic analysis is carried out on the fuzzy river levee region monitoring data through a video motion characteristic analysis model to be trained, and the feature vector of the fuzzy river levee region monitoring data is generated, which comprises the following steps:
and carrying out video motion characteristic analysis on the fuzzy river levee region monitoring data through the characteristic coding branch of the video motion characteristic analysis model to be trained, and generating characteristic vectors of the fuzzy river levee region monitoring data.
Illustratively, the estimating, by combining the first video motion feature and the second video motion feature, the encroaching event estimation on the river levee region monitoring stream data by using a target estimation policy, and generating estimation information of the river levee region monitoring stream data on each river levee encroaching event data under each target estimation policy includes:
Determining association parameters between the first video motion feature and the second video motion feature through a target estimation strategy, and generating target association parameters between the river bank monitoring stream data and the associated river bank monitoring stream data corresponding to each river bank encroaching event data under each target estimation strategy;
determining the target-related parameter as the estimation information;
the generating, in combination with the river bank area monitoring flow data, estimation information of each river bank encroaching event data under each target estimation policy, the river bank encroaching event data of the river bank area monitoring flow data includes:
acquiring influence weight information corresponding to each target estimation strategy;
according to the influence weight information, merging the estimation information corresponding to each target estimation strategy to generate target estimation information corresponding to the river levee encroaching event data;
and determining the river levee encroaching event data of the river levee region monitoring flow data by combining target estimated information corresponding to the river levee encroaching event data.
In a second aspect, an embodiment of the present application further provides a river levee encroaching and positioning pushing system based on unmanned aerial vehicle monitoring, where the river levee encroaching and positioning pushing system based on unmanned aerial vehicle monitoring includes a processor and a machine-readable storage medium, where the machine-readable storage medium stores a computer program, where the computer program is loaded and executed in conjunction with the processor to implement the river levee encroaching and positioning pushing method based on unmanned aerial vehicle monitoring of the first aspect.
According to the technical scheme, the method comprises the steps of extracting the river bank encroaching event data of the river bank area monitoring flow data of the designated river bank area in the designated monitoring period, analyzing and outputting the river bank encroaching event directed graph of the designated river bank area in the designated monitoring period by combining the river bank encroaching event data of the river bank area monitoring flow data, analyzing the rolling characteristic data corresponding to the river bank encroaching event directed graph to further obtain corresponding river bank encroaching locating root cause information, determining the target river bank encroaching locating root cause corresponding to the river bank area structural characteristics of the designated river bank area from the river bank encroaching locating root cause information, pushing the target river bank encroaching locating root cause corresponding to the river bank area structural characteristics of the designated river bank area to the corresponding river bank management server, determining the river bank encroaching locating root cause by taking the event linking characteristics among the river bank encroaching events into consideration, and further combining the river bank area structural characteristics to refine and screen, so that the encroaching locating accuracy can be improved.
Drawings
For a clearer description of the technical solutions of the embodiments of the present application, reference will be made to the accompanying drawings, which are needed to be activated, for the sake of simplicity, and it should be understood that the following drawings only illustrate some embodiments of the present application and should therefore not be considered as limiting the scope, and that other related drawings can be obtained by those skilled in the art without the inventive effort.
Fig. 1 is a schematic flow chart of a method for positioning and pushing a river levee encroachment based on unmanned aerial vehicle monitoring according to an embodiment of the present application;
fig. 2 is a schematic block diagram of a river levee encroaching positioning pushing system based on unmanned aerial vehicle monitoring for implementing the river levee encroaching positioning pushing method based on unmanned aerial vehicle monitoring according to an embodiment of the present application.
Detailed Description
The following description is provided in connection with the accompanying drawings, and the specific operation method in the method embodiment may also be applied to the device embodiment or the system embodiment.
As shown in fig. 1, the embodiment of the application discloses a river levee encroaching, positioning and pushing method based on unmanned aerial vehicle monitoring, which comprises the following specific steps: and 110, extracting river bank encroaching event data of river bank area monitoring flow data of a designated river bank area monitored by the target unmanned aerial vehicle in a designated monitoring period, and analyzing and outputting a river bank encroaching event directed graph of the designated river bank area in the designated monitoring period by combining the river bank encroaching event data of the river bank area monitoring flow data.
For example, the dyke encroachment event directed graph may include a plurality of dyke encroachment events, different dyke encroachment events configured with directed graph association information by event directed relationships that characterize event engagement features between different dyke encroachment events. For example, the river levee intrusion event data of the river levee region monitoring stream data may include river levee intrusion events of each monitoring segment in the river levee region monitoring stream data, where event linking features of each monitoring segment with each other may be resolved (e.g., a garbage stacking event starts from a river levee segment A1 and links to a river levee break event of a river levee segment A2, and then the river levee segments A1-A2 may constitute event linking features).
And 120, inputting the graph convolution characteristic data corresponding to the directed graph of the river bank encroaching event into a river bank monitoring encroaching positioning network meeting the model convergence requirement, and generating river bank encroaching positioning root cause information corresponding to the directed graph of the river bank encroaching event.
Illustratively, the graph rolling feature data may refer to river levee encroachment event data and event engagement feature data between river levee encroachment events, forming a directed graph. The river levee monitoring encroaching positioning network can be trained in advance, so that the capability of determining the river levee encroaching positioning root cause can be obtained. The river levee encroaching locating root cause information may include a plurality of river levee encroaching locating root causes.
And 130, obtaining the river levee region structural characteristics of the designated river levee region, determining a target river levee encroaching locating root cause corresponding to the river levee region structural characteristics from the river levee encroaching locating root cause information, and pushing the target river levee encroaching locating root cause to a corresponding river levee management server.
For example, a sequence of river bank encroaching locating root factors associated with different river bank region structural features may be stored in advance, so that for the different river bank region structural features, only river bank encroaching locating root factors within the range of the sequence of river bank encroaching locating root factors may occur, and therefore on this basis, a target river bank encroaching locating root factor corresponding to the river bank region structural features may be determined from the river bank encroaching locating root factor information, and the target river bank encroaching locating root factor may be pushed to a corresponding river bank management server.
Therefore, in the embodiment, the river levee invasion positioning root information corresponding to the river levee region structural characteristics of the designated river levee region is further obtained by extracting the river levee invasion event data of the designated river levee region monitoring flow data in the designated monitoring period and analyzing and outputting the river levee invasion event directed graph of the designated river levee region in the designated monitoring period in combination with the river levee invasion event data of the river levee region monitoring flow data, so that the river levee invasion positioning root information corresponding to the river levee invasion event directed graph is further obtained by analyzing, and then the target river levee invasion positioning root corresponding to the river levee region structural characteristics of the designated river levee region is determined from the river levee invasion positioning root information and pushed to the corresponding river levee management server.
In one embodiment, in step 120, a training step for monitoring the encroachment positioning network of the river levee is described below.
Step 101, template graph convolution characteristic data of a river bank monitoring and encroaching positioning network to be optimized are obtained; the template graph convolution characteristic data carries encroaching positioning learning data, and the encroaching positioning learning data is used for representing a calibrated river bank encroaching positioning root cause of the template graph convolution characteristic data;
102, configuring a first intermediate positioning network by combining network construction configuration data of the river bank monitoring encroachment positioning network to be optimized;
step 103, combining the template graph convolution feature data with a second intermediate positioning network to obtain root cause cluster vector representations of each estimated river levee encroaching positioning root cause corresponding to the second intermediate positioning network; the river levee monitoring and encroaching positioning network to be optimized and the second intermediate positioning network are deep learning networks with consistent network construction attribute;
104, determining positioning output branch weight information of the first intermediate positioning network by combining root factor cluster vector representations of the positioning root factors of each estimated river levee invasion;
and 105, carrying out loop network configuration information optimization on the first intermediate positioning network by combining the template graph convolution characteristic data and the encroaching positioning learning data, and updating the branch weight information of the first intermediate positioning network except the positioning output branch weight information.
In one embodiment, the template graph rolling feature data is multiple, the loop network configuration information optimization is performed on the first intermediate positioning network by combining the template graph rolling feature data and the encroaching positioning learning data, and the updating of the branch weight information of the first intermediate positioning network except for the positioning output branch weight information includes: respectively inputting the template graph convolution characteristic data into the river bank monitoring and encroaching positioning network to be optimized and the first intermediate positioning network to obtain first root cause knowledge fields which are analyzed and extracted from the template graph convolution characteristic data by the river bank monitoring and encroaching positioning network to be optimized and second root cause knowledge fields which are respectively extracted from the template graph convolution characteristic data by the first intermediate positioning network; combining statistics of the template graph rolling characteristic data, river bank encroaching locating root factor statistics of the estimated river bank encroaching locating root factor, each first root factor knowledge field, each second root factor knowledge field, the encroaching locating learning data and the locating output branch weight information to construct a learning convergence index function; and carrying out cyclic network configuration information optimization on the first intermediate positioning network by taking the learning convergence index function as a network learning effect portrait and combining a first-order optimization algorithm, and updating the branch weight information of the first intermediate positioning network except the positioning output branch weight information until the function value of the learning convergence index function meets the preset condition.
In one embodiment, the learning convergence index function includes a first learning convergence calculation sub-function, and the constructing the learning convergence index function includes: and constructing the first learning convergence calculation sub-function by combining the statistic of the template graph rolling characteristic data, the statistic of the river dike encroaching positioning root cause of the estimated river dike encroaching positioning root cause, each second root cause knowledge field, the encroaching positioning learning data, the positioning output branch weight information and a preset cross entropy loss function.
In one embodiment, the learning convergence index function further includes a second learning convergence calculation sub-function, the learning convergence index function being a sum of the first learning convergence calculation sub-function and the second learning convergence calculation sub-function, and after constructing the first learning convergence calculation sub-function, further includes: combining the knowledge field of each first root cause and the encroaching positioning learning data, determining a first graph convolution characteristic span between template graph convolution characteristic data of the same estimated river dike encroaching positioning root cause in each estimated river dike encroaching positioning root cause, and determining the corresponding estimated learning times of each estimated river dike encroaching positioning root cause; determining a rolling characteristic average span of a first rolling characteristic span corresponding to each estimated river dike encroaching locating root cause by combining the corresponding estimated learning times of each estimated river dike encroaching locating root cause and each first rolling characteristic span; combining the knowledge field of each second root cause and the encroaching positioning learning data to determine a second graph convolution feature span between template graph convolution feature data of the same estimated river dike encroaching positioning root cause in each estimated river dike encroaching positioning root cause; determining a rolling characteristic average span of a second rolling characteristic span corresponding to each estimated river dike encroaching positioning root cause by combining the corresponding estimated learning times of each estimated river dike encroaching positioning root cause and each second rolling characteristic span; and constructing a second learning convergence calculation sub-function by combining the convolution characteristic average span of the first convolution characteristic span, the convolution characteristic average span of the second convolution characteristic span and a set L1 learning convergence index function.
In one embodiment, the template graph rolling feature data is plural, and the obtaining, by combining the template graph rolling feature data with a second intermediate positioning network, root cause cluster vector representations of each estimated river levee encroaching positioning root cause corresponding to the second intermediate positioning network includes: inputting the template graph convolution feature data into the second intermediate positioning network, and acquiring third root cause knowledge fields which are analyzed and extracted from the template graph convolution feature data by the second intermediate positioning network; and combining each third root cause knowledge field with the encroaching positioning learning data, determining a third rolling characteristic average span of a third root cause knowledge field corresponding to the same estimated river dike encroaching positioning root cause in each estimated river dike encroaching positioning root cause, and using the third rolling characteristic average span of the third root cause knowledge field corresponding to the same estimated river dike encroaching positioning root cause as a root cause clustering vector representation of the estimated river dike encroaching positioning root cause corresponding to the second intermediate positioning network.
In one embodiment, the determining the location output branch weight information of the first intermediate location network in combination with root cause cluster vector representation of each estimated river levee encroachment location root cause includes: transposition is carried out on root cause clustering vector representations of the root causes of the invasion of each estimated river levee; the root cause clustering vector representation of the transposed estimated river levee encroaching positioning root cause is converged into a weight matrix; and taking the weight matrix as the positioning output branch weight information of the first intermediate positioning network.
In one embodiment, in step 110, the process of extracting the river levee intrusion event data of the river levee region monitoring flow data of the specified river levee region within the specified monitoring period monitored by the target unmanned aerial vehicle may be implemented by the following exemplary embodiments.
S100: and extracting river levee region monitoring stream data and an associated river levee monitoring stream data sequence.
The sequence of associated river levee monitored stream data includes associated river levee monitored stream data corresponding to one or more river levee encroachment event data. The correlated river bank monitoring flow data is a sequence of river bank area monitoring data as referenced. After the relevant river levee monitoring stream data are obtained, classifying the relevant river levee monitoring stream data through marks of the relevant river levee monitoring stream data, and generating relevant river levee monitoring stream data corresponding to each river levee encroaching event data, wherein the data marks are related to the river levee encroaching event data, and the relevant river levee monitoring stream data can be classified through the data marks.
S200: and carrying out video motion characteristic analysis on the river levee monitoring stream data to generate a first video motion characteristic, and carrying out video motion characteristic analysis on the related river levee monitoring stream data to generate a second video motion characteristic of the related river levee monitoring stream data.
The method includes the steps that video motion feature analysis is conducted on river levee monitoring stream data through a video motion feature analysis model available for deployment, a first video motion feature is generated, video motion feature analysis is conducted on relevant river levee monitoring stream data through a video motion feature analysis model available for deployment, and a second video motion feature relevant to river levee monitoring stream data is generated, wherein the video motion feature analysis model available for deployment can be a neural network model. The video motion feature analysis model needs to be trained, for example, see the following steps S10-S40:
s10: and acquiring a monitoring data sequence of the river levee region to be learned.
The sequence of river bank area monitoring data to be learned comprises fuzzy river bank area monitoring data corresponding to reference river bank encroachment event data and example river bank area monitoring data corresponding to one or more river bank encroachment event data. The carrying of the reference river levee encroachment event data is to mark the river levee encroachment event data in advance, and the example river levee region monitoring data is used as a training sample.
S20: and carrying out video motion characteristic analysis on the fuzzy river levee region monitoring data through a video motion characteristic analysis model to be trained to generate characteristic vectors of the fuzzy river levee region monitoring data, and carrying out video motion characteristic analysis on the example river levee region monitoring data through the video motion characteristic analysis model to be trained to generate characteristic vectors of the example river levee region monitoring data.
The feature vector of the fuzzy river levee region monitoring data can be obtained by performing video motion feature analysis on the fuzzy river levee region monitoring data through a residual network, and the feature vector of the example river levee region monitoring data can also be obtained by performing video motion feature analysis on the example river levee region monitoring data through the network.
S30: and combining the feature vector of the fuzzy river levee region monitoring data with the feature vector of the example river levee region monitoring data, and deciding the fuzzy river levee region monitoring data by combining a target estimation strategy through a video motion feature analysis model to be trained to generate a river levee intrusion event data sequence of the fuzzy river levee region monitoring data.
The sequence of river levee encroachment event data includes target river levee encroachment event data corresponding to each target estimation strategy. The method comprises the steps of making a decision on fuzzy river levee region monitoring data through feature vectors of the fuzzy river levee region monitoring data and example river levee region monitoring data feature vectors and combining a target estimation strategy through a video motion feature analysis model to be trained, and generating a river levee intrusion event data sequence of the fuzzy river levee region monitoring data, for example: and determining association parameters (such as similarity) between characteristic vectors of the fuzzy river levee region monitoring data and characteristic vectors of the example river levee region monitoring data by combining a video motion characteristic analysis model to be trained with target estimation strategies, generating estimated association parameters between the fuzzy river levee region monitoring data and the example river levee region monitoring data under each target estimation strategy, determining target river levee encroaching event data of the fuzzy river levee region monitoring data under each target estimation strategy by combining the estimated association parameters, and generating a river levee encroaching event data sequence of the fuzzy river levee region monitoring data by the target river levee encroaching event data of the fuzzy river levee region monitoring data under each target estimation strategy. In the embodiment of the application, the target estimation strategies can be various, and the target estimation strategies are methods for estimating the encroachment event of the monitored stream data of the river levee region.
And determining association parameters between the characteristic vectors of the fuzzy river levee region monitoring data and the characteristic vectors of the example river levee region monitoring data through a target estimation strategy, and generating estimated association parameters between the fuzzy river levee region monitoring data and the example river levee region monitoring data under each target estimation strategy.
The estimated association parameter with the largest association parameter can be selected from the estimated association parameters, and the target river dike encroachment event data of the fuzzy river dike area monitoring data can be determined through the characteristic vector of the exemplary river dike area monitoring data corresponding to the estimated association parameter with the largest association parameter. The target river levee encroaching event data of the fuzzy river levee region monitoring data is river levee encroaching event data corresponding to the estimated association parameter with the largest association parameter.
S40: and combining target river levee encroaching event data corresponding to each target estimation strategy with reference river levee encroaching event data, updating traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, and generating a video motion feature analysis model for deployment.
The process of generating the video motion feature analysis model for deployment by updating the traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met through target river levee encroachment event data and reference river levee encroachment event data corresponding to each target estimation strategy specifically comprises the following steps: and calculating a first LOSS value between target river levee encroaching event data and reference river levee encroaching event data corresponding to each target estimation strategy, fusing the first LOSS value to generate a fused LOSS value, and updating traversing model weight information of a video motion feature analysis model to be trained until the model convergence requirement is met through the fused LOSS value to generate a video motion feature analysis model for deployment.
The training steps above specifically may include: performing monitoring priori frame expansion on the fuzzy river levee region monitoring data to generate expanded river levee region monitoring data, combining the expanded river levee region monitoring data, estimating an encroaching event of the expanded river levee region monitoring data through a river levee encroaching decision branch of a video motion characteristic analysis model to be trained, generating estimated river levee encroaching event data of the expanded river levee region monitoring data, acquiring river levee encroaching event priori labeling data corresponding to the expanded river levee region monitoring data, and determining a second LOSS value between the estimated river levee encroaching event data and the river levee encroaching event priori labeling data.
Therefore, the process of updating the traversing model weight information of the video motion feature analysis model to be trained by fusing the LOSS value until the model convergence requirement is met may include: and combining the fusion LOSS value and the second LOSS value to update traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, and generating the video motion feature analysis model for deployment.
As an implementation manner, the process of updating the traversing model weight information of the video motion feature analysis model to be trained by combining the fusion LOSS value and the second LOSS value until the model convergence requirement is met may include: and adding the fusion LOSS value and the second LOSS value to obtain a total LOSS value, and updating traversing model weight information of the video motion feature analysis model to be trained by combining the total LOSS value until the model convergence requirement is met, so as to generate a video motion feature analysis model for deployment.
If the extended river bank area monitoring data comprises extended river bank area monitoring data obtained by extending the fuzzy river bank area monitoring data, the river bank encroaching decision branch comprises a first river bank encroaching decision branch and a second river bank encroaching decision branch, and the estimated river bank encroaching event data comprises first estimated river bank encroaching event data and second estimated river bank encroaching event data. Then, in combination with the extended river bank area monitoring data, the process of estimating the intrusion event data of the extended river bank area monitoring data by the river bank intrusion decision-making branch of the video motion feature analysis model to be trained may include: and performing encroaching event estimation through a first river bank encroaching decision branch of the video motion feature analysis model to be trained in combination with the expanded river bank monitoring data to generate first estimated river bank encroaching event data of target expanded river bank monitoring data, and performing encroaching event estimation through a second river bank encroaching decision branch of the video motion feature analysis model to be trained in combination with the expanded river bank monitoring data to generate second estimated river bank encroaching event data of the expanded river bank monitoring data, wherein the first estimated river bank encroaching event data is used for indicating that the target expanded river bank monitoring data is the river bank encroaching event data of the fuzzy river bank monitoring data, the target expanded river bank monitoring data is the river bank monitoring data related to the fuzzy river bank monitoring data, in other words, the target expanded river bank monitoring data can represent the river bank monitoring data with the largest related parameters among the expanded river bank monitoring data, and if the expanded river bank monitoring data and the fuzzy river bank monitoring data are considered to be the corresponding to the fuzzy river bank monitoring data. In this embodiment, the first dyke encroaching decision branch is combined to perform encroaching event estimation on the extended dyke area monitoring data, where one dyke encroaching event data of the first dyke encroaching decision branch is dyke encroaching event data of the fuzzy dyke area monitoring data, so that the dyke encroaching event data of the target extended dyke area monitoring data, namely, the first estimated dyke encroaching event data, can be resolved in a plurality of dyke encroaching event data obtained after the decision operation of the first dyke encroaching decision branch. The first dyke encroaching decision branch may be inconsistent with the second dyke encroaching decision branch.
The second estimated river levee encroachment event data is used to indicate that the extended river levee region monitoring data is for the extended river levee encroachment event data.
S300: and according to the first video motion characteristic and the second video motion characteristic, performing encroachment event estimation on the river levee region monitoring flow data through a target estimation strategy, and generating estimation information of the river levee region monitoring flow data on each river levee encroachment event data under each target estimation strategy.
The method for estimating the encroaching event of the river levee region monitoring flow data by combining the first video motion feature and the second video motion feature through a target estimation strategy, and generating the estimated information of the river levee region monitoring flow data for each river levee encroaching event data under each target estimation strategy specifically comprises the following steps:
and determining the association parameters between the first video motion feature and the second video motion feature through a target estimation strategy, generating target association parameters between the river bank monitoring flow data and the associated river bank monitoring flow data corresponding to the river bank encroaching event data under each target estimation strategy, and taking the target association parameters as estimation information. The target estimation strategy includes a first target estimation strategy and a second target estimation strategy, wherein the first target estimation strategy is used for determining association parameters between a first video motion characteristic and a second video motion characteristic, generating target association parameters between river bank area monitoring stream data and associated river bank monitoring stream data corresponding to each river bank intrusion event data under the first target estimation strategy, and the second target estimation strategy is used for determining association parameters between the first video motion characteristic and the second video motion characteristic, and generating target association parameters between river bank area monitoring stream data and associated river bank monitoring stream data corresponding to each river bank intrusion event data under the second target estimation strategy.
S400: and determining the river levee encroaching event data of the river levee region monitoring flow data according to the estimated information of the river levee encroaching event data of the river levee region monitoring flow data under each target estimated strategy.
The process of determining the river bank encroaching event data of the river bank area monitoring flow data by combining the estimated information of the river bank area monitoring flow data for each river bank encroaching event data under each target estimated strategy specifically may include the following steps:
s401: and merging the estimated information corresponding to each target estimated strategy for each river levee encroachment event data to generate target estimated information corresponding to the river levee encroachment event data.
The estimation information may be a confidence, and the estimation information corresponding to each target estimation policy may be weighted so that the estimation information corresponding to each target estimation policy is fused, for example, the influence weight information corresponding to each target estimation policy is obtained, and the estimation information corresponding to each target estimation policy is fused by the influence weight information, so as to generate the target estimation information corresponding to the river levee intrusion event data. The influence weight information is a weight parameter obtained by repeatedly updating a video motion characteristic analysis model for deployment. In addition, the estimation information corresponding to each target estimation policy may be added to fuse the estimation information corresponding to each target estimation policy.
S402: and determining the river levee encroaching event data of the river levee region monitoring flow data according to the target estimated information corresponding to the river levee encroaching event data.
The estimation information may be a confidence, each piece of the river levee encroachment event data corresponds to the target estimation information, and the maximum target estimation information may be determined from the target estimation information, and the corresponding river levee encroachment event data is river levee encroachment event data of the river levee region monitoring flow data. For example, the river levee region monitoring stream data and the associated river levee monitoring stream data sequence may be obtained, the associated river levee monitoring stream data sequence includes one or more pieces of associated river levee monitoring stream data corresponding to the river levee intrusion event data, video motion feature analysis is performed on the river levee region monitoring stream data to generate a first video motion feature, video motion feature analysis is performed on the associated river levee monitoring stream data to generate a second video motion feature of the associated river levee monitoring stream data, the first video motion feature and the second video motion feature are combined, the intrusion event is performed on the river levee region monitoring stream data through the target estimation strategy to obtain estimated information of the river levee intrusion event data for each river levee under each target estimation strategy, and the river levee intrusion event data of the river levee region monitoring stream data is determined in combination with the estimated information of the river levee intrusion event data for each river levee under each target estimation strategy. Because the river levee encroaching event data of the river levee region monitoring flow data can be generated by combining the generalized estimation information of each river levee encroaching event data under each target estimation strategy, the estimation information corresponding to different target estimation strategies can be measured simultaneously.
Based on the above, further description of embodiments is provided below, for example, see the following steps S1000-S8000:
s1000: and acquiring a monitoring data sequence of the river levee region to be learned.
The sequence of river bank area monitoring data to be learned comprises fuzzy river bank area monitoring data samples corresponding to reference river bank encroachment event data and one or more example river bank area monitoring data corresponding to river bank encroachment event data.
S2000: and carrying out video motion characteristic analysis on the fuzzy river levee region monitoring data through a video motion characteristic analysis model to be trained to generate characteristic vectors of the fuzzy river levee region monitoring data, and carrying out video motion characteristic analysis on the example river levee region monitoring data through the video motion characteristic analysis model to be trained to generate characteristic vectors of the example river levee region monitoring data.
The method comprises the steps of carrying out video motion feature analysis on fuzzy river levee region monitoring data through a video motion feature analysis model to be trained to generate feature vectors of the fuzzy river levee region monitoring data, carrying out video motion feature analysis on example river levee region monitoring data through the video motion feature analysis model to be trained to generate example river levee region monitoring data feature vectors, carrying out depth video motion feature analysis on the fuzzy river levee region monitoring data through a residual error network through the video motion feature analysis model to be trained to generate feature vectors of the fuzzy river levee region monitoring data, and carrying out depth video motion feature analysis on the example river levee region monitoring data through the residual error network through the video motion feature analysis model to be trained to generate example river levee region monitoring data feature vectors.
S3000: and combining the feature vector of the fuzzy river levee region monitoring data with the feature vector of the example river levee region monitoring data, and deciding the fuzzy river levee region monitoring data by combining a target estimation strategy through a video motion feature analysis model to be trained to generate a river levee intrusion event data sequence of the fuzzy river levee region monitoring data.
The river levee encroaching event data sequence comprises target river levee encroaching event data corresponding to each target estimation strategy, the characteristic vector of the fuzzy river levee region monitoring data and the characteristic vector of the example river levee region monitoring data are combined, the fuzzy river levee region monitoring data are decided by combining the target estimation strategy through a video motion characteristic analysis model to be trained, and the process of generating the river levee encroaching event data sequence of the fuzzy river levee region monitoring data comprises the following steps:
and determining association parameters between characteristic vectors of the fuzzy river levee region monitoring data and characteristic vectors of the example river levee region monitoring data by combining target estimation strategies through a video motion characteristic analysis model to be trained, generating estimated association parameters between the fuzzy river levee region monitoring data and the example river levee region monitoring data under each target estimation strategy, determining target river levee encroaching event data of the fuzzy river levee region monitoring data under each target estimation strategy by combining the estimated association parameters, and generating a river levee encroaching event data sequence of the fuzzy river levee region monitoring data by combining the target river levee encroaching event data of the fuzzy river levee region monitoring data under each target estimation strategy.
If the example river levee region monitoring data corresponding to the river levee intrusion event data comprises a plurality of groups, video motion feature analysis can be performed on all the example river levee region monitoring data corresponding to the river levee intrusion event data with a plurality of groups of the example river levee region monitoring data in combination with a video motion feature analysis model to be trained, a prepared example river levee region monitoring data feature vector corresponding to each example river levee region monitoring data under the river levee intrusion event data is generated, and then the prepared example river levee region monitoring data feature vectors of each example river levee region monitoring data under the river levee intrusion event data are fused to obtain an example river levee region monitoring data feature vector of the example river levee region monitoring data under the river levee intrusion event data.
For example, the exemplary river bank area monitoring data in the river bank intrusion event data X includes xa, xb and xc, video motion feature analysis is performed on the exemplary river bank area monitoring data in the river bank intrusion event data X through a video motion feature analysis model to be trained, a prepared exemplary river bank area monitoring data feature vector F (xa) corresponding to xa is generated, a prepared exemplary river bank area monitoring data feature vector F (xb) corresponding to xb and a prepared exemplary river bank area monitoring data feature vector F (xc) corresponding to xc are generated, and then an average number of all prepared exemplary river bank area monitoring data feature vectors under the river bank intrusion event data is determined, so that feature vectors F (X), F (X) = (F (xa) +f (xb) +f (xc))/3 are obtained.
Thus, estimated association parameters between the fuzzy river bank region monitoring data and the example river bank region monitoring data under each target estimation policy are obtained, the estimated association parameters can be expressed by percentage parameters, and the generated estimated association parameters of the fuzzy river bank region monitoring data under each target estimation policy for each river bank encroachment event data, for example, the estimated information of the determined fuzzy river bank region monitoring data is (a; B, C), A is the estimated association parameter of the fuzzy river bank area monitoring data in the mode for each river bank intrusion event data, B is the estimated association parameter of the fuzzy river bank area monitoring data in the mode II for each river bank intrusion event data, C is the estimated association parameter of the fuzzy river bank area monitoring data in the mode III for each river bank intrusion event data, one, two and three are different vector distance calculation modes, the river bank intrusion event data comprises the river bank intrusion event data A, the river bank intrusion event data B and the river bank intrusion event data C, the estimated association parameter of the fuzzy river bank area monitoring data in the mode A= (40%; 60%; 80%) is the estimated association parameter of the fuzzy river bank area monitoring data for the river bank intrusion event data A, and 60% is the estimated association parameter of the fuzzy river bank area monitoring data for the river bank intrusion event data B, and 80% is the estimated association parameter of the fuzzy river bank intrusion event data C. For each target estimation strategy, determining the estimated association parameter with the largest number from the estimated association parameters of the fuzzy river bank area monitoring data under the target estimation strategy for each river bank intrusion event data, namely the estimated association parameter with the highest association parameter, and then determining the river bank intrusion event data corresponding to the largest estimated association parameter as the target river bank intrusion event data of the fuzzy river bank area monitoring data under the target estimation strategy. In the above example, the target estimation policy is one mode, a= (40%, 60%, 80%), 80% is the maximum estimated correlation parameter, and 80% is the estimated correlation parameter of the river bank intrusion event data C, and the river bank intrusion event data C is the target river bank intrusion event data of the fuzzy river bank area monitoring data.
S4000: and combining target river levee encroaching event data corresponding to each target estimation strategy with reference river levee encroaching event data, updating traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, and generating a video motion feature analysis model for deployment.
Combining the target river levee encroaching event data and the reference river levee encroaching event data corresponding to each target estimation strategy, updating the traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, and generating the video motion feature analysis model for deployment can comprise: determining a first LOSS value between target river levee encroaching event data and reference river levee encroaching event data corresponding to each target estimation strategy, fusing the first LOSS value to obtain a fused LOSS value, and updating traversing model weight information of a video motion feature analysis model to be trained by combining the fused LOSS value until the model convergence requirement is met, so as to generate a video motion feature analysis model for deployment.
The above method may further comprise the steps of:
performing monitoring priori frame expansion on the fuzzy river levee region monitoring data to generate expanded river levee region monitoring data, combining the expanded river levee region monitoring data, estimating an encroaching event of the expanded river levee region monitoring data through a river levee encroaching decision branch of a video motion characteristic analysis model to be trained, generating estimated river levee encroaching event data of the expanded river levee region monitoring data, acquiring river levee encroaching event priori labeling data corresponding to the expanded river levee region monitoring data, and determining a second LOSS value between the estimated river levee encroaching event data and the river levee encroaching event priori labeling data.
And then, the video motion characteristic analysis model utilizes a residual network to carry out depth video motion characteristic analysis on the extended river levee region monitoring data, and an extended river levee region monitoring data characteristic vector corresponding to the extended river levee region monitoring data is generated. And combining the characteristic vector of the monitoring data of the extended river levee region, estimating the encroaching event of the monitoring data of the extended river levee region through a river levee encroaching decision branch of a video motion characteristic analysis model to be trained, and generating estimated river levee encroaching event data of the monitoring data of the extended river levee region.
And after the feature vector of the monitoring data of the extended river levee region is input into a video motion feature analysis model to be trained, the monitoring data of the extended river levee region is estimated through a river levee encroaching decision branch. The river levee encroaching decision branch may include a first river levee encroaching decision branch and a second river levee encroaching decision branch, the estimated river levee encroaching event data includes a first estimated river levee encroaching event data and a second estimated river levee encroaching event data, so that, in combination with the extended river levee region monitoring data, the encroaching event estimation is performed on the extended river levee region monitoring data through the river levee encroaching decision branch of the video motion feature analysis model to be trained, and the process of generating the estimated river levee encroaching event data of the extended river levee region monitoring data may include:
S4100: and combining the expanded monitoring data of the expanded river dike region, estimating an encroaching event through a first river dike encroaching decision branch of a video motion characteristic analysis model to be trained, and generating first estimated river dike encroaching event data of the target expanded river dike region monitoring data.
The first estimated river bank encroaching event data is used for indicating river bank encroaching event data of target extended river bank area monitoring data relative to the fuzzy river bank area monitoring data, wherein the target extended river bank area monitoring data is river bank area monitoring data related to the fuzzy river bank area monitoring data in the extended river bank area monitoring data.
S4200: and (3) combining the expanded monitoring data of the expanded river dike region, estimating an encroaching event through a second river dike encroaching decision branch of the video motion characteristic analysis model to be trained, and generating second estimated river dike encroaching event data of the monitoring data of the expanded river dike region.
The second estimated river levee encroachment event data is used for indicating that the expanded river levee region monitoring data is updated by self-supervised learning for the expanded river levee encroachment event data.
The second LOSS value comprises a first branch LOSS value and a second branch LOSS value, the river bank encroaching event priori annotation data comprises a first river bank encroaching event priori annotation data and a second river bank encroaching event priori annotation data, the first branch LOSS value between the first river bank encroaching event priori annotation data and the first estimated river bank encroaching event data is determined, the second branch LOSS value between the second river bank encroaching event priori annotation data and the second estimated river bank encroaching event data is determined, according to the fusion LOSS value and the second LOSS value, traversing model weight information of the video motion feature analysis model to be trained is updated until the model convergence requirement is met, a video motion feature analysis model which can be used for deployment is generated, in other words, the fusion LOSS value, the first branch LOSS value and the second branch LOSS value can be fused, the fused LOSS value is generated, the video motion feature model to be trained is updated until the model convergence requirement is met by combining the fused LOSS value, and the motion feature analysis model to be used for deployment is generated.
S5000: and acquiring the river levee region monitoring stream data and the related river levee monitoring stream data sequence.
The sequence of associated river levee monitored stream data includes associated river levee monitored stream data corresponding to one or more river levee encroachment event data.
S6000: and carrying out video motion characteristic analysis on the river levee monitoring stream data through a video motion characteristic analysis model which can be used for deployment to generate a first video motion characteristic, and carrying out video motion characteristic analysis on the relevant river levee monitoring stream data through a video motion characteristic analysis model which can be used for deployment to generate a second video motion characteristic of the relevant river levee monitoring stream data.
S7000: and according to the first video motion characteristic and the second video motion characteristic, performing encroachment event estimation on the river levee region monitoring flow data through a target estimation strategy, and generating estimation information of the river levee region monitoring flow data on each river levee encroachment event data under each target estimation strategy.
And carrying out encroaching event estimation on the river levee region monitoring flow data through a target estimation strategy by combining the first video motion characteristic and the second video motion characteristic, wherein the specific process for generating the estimation information of the river levee region monitoring flow data on the encroaching event data of each river levee under each target estimation strategy comprises the following steps:
And determining association parameters between the first video motion feature and the second video motion feature through a target estimation strategy by combining a video motion feature analysis model available for deployment, generating target association parameters between river bank monitoring flow data and associated river bank monitoring flow data corresponding to each river bank intrusion event data under each target estimation strategy, and determining the target association parameters as estimation information.
S8000: and determining the river bank encroaching event data of the river bank area monitoring flow data according to the estimated information of the river bank area monitoring flow data for each river bank encroaching event data under the respective target estimated strategies.
Combining the estimated information of the river bank area monitoring flow data for each river bank encroachment event data under each target estimated strategy, and determining the river bank encroachment event data of the river bank area monitoring flow data comprises the following steps: and merging the estimated information corresponding to each target estimated strategy for each river bank encroachment event data, generating target estimated information corresponding to the river bank encroachment event data, and determining the river bank encroachment event data of the river bank area monitoring flow data by combining the target estimated information corresponding to the river bank encroachment event data.
For each river levee encroachment event data, fusing the estimation information corresponding to each target estimation strategy, and generating the target estimation information corresponding to the river levee encroachment event data specifically comprises the following steps: and acquiring influence weight information corresponding to each target estimation strategy, and fusing the estimation information corresponding to each target estimation strategy by combining the influence weight information to generate target estimation information corresponding to river levee encroaching event data, wherein the influence weight information is a weight parameter obtained by repeatedly updating a video motion characteristic analysis model for deployment.
The method comprises the steps of generating first video motion characteristics by video motion characteristic analysis of the river bank area monitoring stream data, generating second video motion characteristics of the river bank area monitoring stream data by video motion characteristic analysis of the river bank area monitoring stream data, performing invasion event estimation of the river bank area monitoring stream data by a target estimation strategy in combination with the first video motion characteristics and the second video motion characteristics, generating estimated information of the river bank area monitoring stream data for each river bank invasion event data under each target estimation strategy, determining the river bank invasion event data of the river bank area monitoring stream data by combining with the estimated information of each river bank invasion event data under each target estimation strategy, and determining the information of the river bank invasion event data for each river bank invasion event data under the corresponding to each estimated river bank invasion event data.
Fig. 2 schematically illustrates a riverbank intrusion positioning and pushing system 100 based on drone monitoring that may be used to implement the various embodiments described herein.
For one embodiment, fig. 2 shows a drone-monitoring-based riverbank intrusion positioning pushing system 100 having a plurality of processors 102, a control module (chipset) 104 coupled to at least one of the processor(s) 102, a memory 106 coupled to the control module 104, a non-volatile memory (NVM)/storage 108 coupled to the control module 104, a plurality of input/output devices 110 coupled to the control module 104, and a network interface 112 coupled to the control module 106.
Processor 102 may include a plurality of single-core or multi-core processors, and processor 102 may include any combination of general-purpose or special-purpose processors (e.g., graphics processors, application processors, baseband processors, etc.). In some embodiments, the riverbank intrusion positioning pushing system 100 based on unmanned aerial vehicle monitoring can be used as a server device such as a gateway in the embodiments of the present application.
In some embodiments, the unmanned-monitoring-based riverbank intrusion positioning pushing system 100 may include a plurality of computer-readable media (e.g., memory 106 or NVM/storage 108) having instructions 114 and a plurality of processors 102 in combination with the plurality of computer-readable media configured to execute the instructions 114 to implement the modules to perform the actions described in this disclosure.
For one embodiment, the control module 104 may include any suitable interface controller to provide any suitable interface to one or more of the processor(s) 102 and/or any suitable device or component in communication with the control module 104.
The control module 104 may include a memory controller module to provide an interface to the memory 106. The memory controller modules may be hardware modules, software modules, and/or firmware modules.
The memory 106 may be used to load and store data and/or instructions 114 for, for example, a riverbank intrusion location-based system 100 based on unmanned aerial vehicle monitoring. For one embodiment, memory 106 may comprise any suitable volatile memory, such as, for example, a suitable DRAM. In some embodiments, memory 106 may comprise double data rate type four synchronous dynamic random access memory (DDR 4 SDRAM).
For one embodiment, the control module 104 may include a plurality of input/output controllers to provide interfaces to the NVM/storage 108 and the input/output device(s) 110.
For example, NVM/storage 108 may be used to store data and/or instructions 114. NVM/storage 108 may include any suitable nonvolatile memory (e.g., flash memory) and/or may include any suitable nonvolatile storage(s) (e.g., a plurality of Hard Disk Drives (HDDs), a plurality of Compact Disc (CD) drives, and/or a plurality of Digital Versatile Disc (DVD) drives).
NVM/storage 108 may include storage resources that are physically part of the device on which the drone-based monitoring of river levee intrusion positioning and pushing system 100 is installed, or which may be accessible by the device, but may not be necessary as part of the device. For example, NVM/storage 108 may be accessed via input/output device(s) 110 in connection with a network.
The input/output device(s) 110 may provide an interface for the unmanned-based riverbank intrusion positioning and pushing system 100 to communicate with any other suitable device, and the input/output device 110 may include a communication component, pinyin component, sensor component, and the like. The network interface 112 may provide an interface for the unmanned-vehicle-monitoring-based riverbank intrusion positioning and pushing system 100 to communicate in accordance with a plurality of networks, and the unmanned-monitoring-based riverbank intrusion positioning and pushing system 100 may communicate wirelessly with a plurality of components of a wireless network in accordance with any of a plurality of wireless network standards and/or protocols, such as accessing a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G, 5G, etc., or a combination thereof.
For one embodiment, at least one of the processor(s) 102 may be packaged together with logic of a plurality of controllers (e.g., memory controller modules) of the control module 104. For one embodiment, at least one of the processor(s) 102 may be packaged together with logic of a plurality of controllers of the control module 104 to form a System In Package (SiP). For one embodiment, at least one of the processor(s) 102 may be integrated on the same die with logic of multiple controllers of the control module 104. For one embodiment, at least one of the processor(s) 102 may be integrated on the same die with logic of multiple controllers of the control module 104 to form a system on chip (SoC).
In various embodiments, the riverbank intrusion positioning pushing system 100 based on unmanned aerial vehicle monitoring may be, but is not limited to, the following: a desktop computing device or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet, a netbook, etc.), and the like. In various embodiments, the unmanned-based riverbank intrusion positioning and pushing system 100 may have more or fewer components and/or different architectures. For example, in some embodiments, the unmanned-vehicle-monitoring-based riverbank intrusion positioning and pushing system 100 includes multiple cameras, a keyboard, a Liquid Crystal Display (LCD) screen (including a touch screen display), a non-volatile memory port, multiple antennas, a graphics chip, an Application Specific Integrated Circuit (ASIC), and a speaker.
The foregoing has outlined rather broadly the more detailed description of the present application, wherein specific examples have been provided to illustrate the principles and embodiments of the present application, the description of the examples being provided solely to assist in the understanding of the method of the present application and the core concepts thereof; meanwhile, as those skilled in the art will have modifications in the specific embodiments and application scope in accordance with the ideas of the present application, the present description should not be construed as limiting the present application in view of the above.

Claims (10)

1. The method is characterized by being applied to a river levee encroaching positioning pushing system based on unmanned aerial vehicle monitoring, and comprises the following steps:
extracting river bank encroaching event data of river bank monitoring flow data of a designated river bank area monitored by a target unmanned aerial vehicle in a designated monitoring period, and analyzing and outputting a river bank encroaching event directed graph of the designated river bank area in the designated monitoring period by combining the river bank encroaching event data of the river bank monitoring flow data, wherein the river bank encroaching event directed graph comprises a plurality of river bank encroaching events, different river bank encroaching events are configured with directed graph associated information through event directed relations, and the event directed relations represent event linking characteristics among different river bank encroaching events;
inputting the graph rolling characteristic data corresponding to the directed graph of the river levee encroaching event into a river levee monitoring encroaching positioning network meeting the model convergence requirement, and generating river levee encroaching positioning root cause information corresponding to the directed graph of the river levee encroaching event;
and acquiring the river dike region structural characteristics of the designated river dike region, determining a target river dike encroaching locating root cause corresponding to the river dike region structural characteristics from the river dike encroaching locating root cause information, and pushing the target river dike encroaching locating root cause to a corresponding river dike management server.
2. The method for positioning and pushing a river levee encroachment based on unmanned aerial vehicle monitoring according to claim 1, wherein the river levee monitoring encroachment positioning network is generated by the following steps:
acquiring template graph convolution characteristic data of a river bank monitoring encroaching positioning network to be optimized; the template graph convolution characteristic data carries encroaching positioning learning data, and the encroaching positioning learning data is used for representing a calibrated river bank encroaching positioning root cause of the template graph convolution characteristic data;
the network construction configuration data of the encroaching positioning network is combined with the river bank monitoring to be optimized, and a first intermediate positioning network is configured;
combining the template graph rolling characteristic data with a second intermediate positioning network to obtain root cause clustering vector representations of each estimated river levee encroaching positioning root cause corresponding to the second intermediate positioning network; the river levee monitoring and encroaching positioning network to be optimized and the second intermediate positioning network are deep learning networks with consistent network construction attribute;
combining root cause cluster vector representations of the estimated river levee encroaching positioning root causes to determine positioning output branch weight information of the first intermediate positioning network;
And carrying out cyclic network configuration information optimization on the first intermediate positioning network by combining the template graph convolution characteristic data and the encroaching positioning learning data, and updating the branch weight information of the first intermediate positioning network except the positioning output branch weight information.
3. The method for positioning and pushing the river levee encroachment based on unmanned aerial vehicle monitoring according to claim 2, wherein the template graph convolution feature data is a plurality of, the loop network configuration information optimization is performed on the first intermediate positioning network by combining the template graph convolution feature data and the encroachment positioning learning data, and updating the branch weight information of the first intermediate positioning network except the positioning output branch weight information comprises:
respectively inputting the template graph convolution characteristic data into the river bank monitoring and encroaching positioning network to be optimized and the first intermediate positioning network to obtain first root cause knowledge fields which are analyzed and extracted from the template graph convolution characteristic data by the river bank monitoring and encroaching positioning network to be optimized and second root cause knowledge fields which are respectively extracted from the template graph convolution characteristic data by the first intermediate positioning network;
Combining statistics of the template graph rolling characteristic data, river bank encroaching locating root factor statistics of the estimated river bank encroaching locating root factor, each first root factor knowledge field, each second root factor knowledge field, the encroaching locating learning data and the locating output branch weight information to construct a learning convergence index function;
and carrying out cyclic network configuration information optimization on the first intermediate positioning network by taking the learning convergence index function as a network learning effect portrait and combining a first-order optimization algorithm, and updating the branch weight information of the first intermediate positioning network except the positioning output branch weight information until the function value of the learning convergence index function meets the preset condition.
4. A method for locating and pushing a river levee encroachment based on unmanned aerial vehicle monitoring as claimed in claim 3, wherein the learning convergence index function comprises a first learning convergence calculation sub-function, and the constructing the learning convergence index function comprises:
and constructing the first learning convergence calculation sub-function by combining the statistic of the template graph rolling characteristic data, the statistic of the river dike encroaching positioning root cause of the estimated river dike encroaching positioning root cause, each second root cause knowledge field, the encroaching positioning learning data, the positioning output branch weight information and a preset cross entropy loss function.
5. The method for locating and pushing a river levee based on unmanned aerial vehicle monitoring according to claim 4, wherein the learning convergence index function further comprises a second learning convergence calculation sub-function, the learning convergence index function being a sum of the first learning convergence calculation sub-function and the second learning convergence calculation sub-function, and further comprising, after constructing the first learning convergence calculation sub-function:
combining the knowledge field of each first root cause and the encroaching positioning learning data, determining a first graph convolution characteristic span between template graph convolution characteristic data of the same estimated river dike encroaching positioning root cause in each estimated river dike encroaching positioning root cause, and determining the corresponding estimated learning times of each estimated river dike encroaching positioning root cause;
determining a rolling characteristic average span of a first rolling characteristic span corresponding to each estimated river dike encroaching locating root cause by combining the corresponding estimated learning times of each estimated river dike encroaching locating root cause and each first rolling characteristic span;
combining the knowledge field of each second root cause and the encroaching positioning learning data to determine a second graph convolution feature span between template graph convolution feature data of the same estimated river dike encroaching positioning root cause in each estimated river dike encroaching positioning root cause;
Determining a rolling characteristic average span of a second rolling characteristic span corresponding to each estimated river dike encroaching positioning root cause by combining the corresponding estimated learning times of each estimated river dike encroaching positioning root cause and each second rolling characteristic span;
and constructing a second learning convergence calculation sub-function by combining the convolution characteristic average span of the first convolution characteristic span, the convolution characteristic average span of the second convolution characteristic span and a set L1 learning convergence index function.
6. The method for positioning and pushing the estuary based on unmanned aerial vehicle monitoring according to claim 2, wherein the template map convolution feature data is a plurality of, the combining the template map convolution feature data with the second intermediate positioning network, obtaining root cause cluster vector representations of each estimated estuary intrusion positioning root cause corresponding to the second intermediate positioning network, includes:
inputting the template graph convolution feature data into the second intermediate positioning network, and acquiring third root cause knowledge fields which are analyzed and extracted from the template graph convolution feature data by the second intermediate positioning network;
and combining each third root cause knowledge field with the encroaching positioning learning data, determining a third rolling characteristic average span of a third root cause knowledge field corresponding to the same estimated river dike encroaching positioning root cause in each estimated river dike encroaching positioning root cause, and using the third rolling characteristic average span of the third root cause knowledge field corresponding to the same estimated river dike encroaching positioning root cause as a root cause clustering vector representation of the estimated river dike encroaching positioning root cause corresponding to the second intermediate positioning network.
7. The method for locating and pushing a river levee encroachment based on unmanned aerial vehicle monitoring according to claim 6, wherein the determining the location output branch weight information of the first intermediate location network by combining root cause cluster vector representations of the estimated river levee encroachment locating root causes comprises:
transposition is carried out on root cause clustering vector representations of the root causes of the invasion of each estimated river levee;
the root cause clustering vector representation of the transposed estimated river levee encroaching positioning root cause is converged into a weight matrix;
and taking the weight matrix as the positioning output branch weight information of the first intermediate positioning network.
8. The method for positioning and pushing the river levee encroachment based on the unmanned aerial vehicle monitoring according to any one of claims 1 to 7, wherein the extracting of the river levee encroachment event data of the river levee region monitoring flow data of the specified river levee region monitored by the target unmanned aerial vehicle in the specified monitoring period is realized by the following steps:
extracting an associated river levee monitoring stream data sequence corresponding to the river levee area monitoring stream data, wherein the associated river levee monitoring stream data sequence comprises one or more associated river levee monitoring stream data corresponding to river levee encroaching event data;
Performing video motion feature analysis on the river levee region monitoring stream data to generate a first video motion feature, and performing video motion feature analysis on the related river levee monitoring stream data to generate a second video motion feature of the related river levee monitoring stream data;
performing encroaching event estimation on the river levee region monitoring flow data through a target estimation strategy by combining the first video motion characteristic and the second video motion characteristic, and generating estimation information of the river levee region monitoring flow data for each river levee encroaching event data under each target estimation strategy;
and generating river levee encroaching event data of the river levee region monitoring flow data by combining the estimated information of the river levee region monitoring flow data for each river levee encroaching event data under each target estimated strategy.
9. The method for positioning and pushing the river levee encroachment based on unmanned aerial vehicle monitoring according to claim 8, wherein the first video motion feature and the second video motion feature are feature vectors obtained by performing video motion feature analysis on a video motion feature analysis model completed by prior training;
the method comprises the steps of carrying out video motion feature analysis on the river levee region monitoring stream data to generate a first video motion feature, carrying out video motion feature analysis on the related river levee monitoring stream data, and carrying out model convergence optimization on a video motion feature analysis model before generating a second video motion feature of the related river levee monitoring stream data, wherein the method is realized by the following steps:
Acquiring a river levee region monitoring data sequence to be learned, wherein the river levee region monitoring data sequence to be learned comprises fuzzy river levee region monitoring data corresponding to reference river levee encroaching event data and example river levee region monitoring data corresponding to one or more river levee encroaching event data;
performing video motion characteristic analysis on the fuzzy river levee region monitoring data through a video motion characteristic analysis model to be trained to generate characteristic vectors of the fuzzy river levee region monitoring data, and performing video motion characteristic analysis on the example river levee region monitoring data through the video motion characteristic analysis model to be trained to generate example river levee region monitoring data characteristic vectors;
according to the video motion characteristic analysis model to be trained, determining association parameters between characteristic vectors of the fuzzy river levee region monitoring data and characteristic vectors of the example river levee region monitoring data in combination with a target estimation strategy, and generating estimated association parameters between the fuzzy river levee region monitoring data and the example river levee region monitoring data under each target estimation strategy;
determining target river levee encroachment event data of the fuzzy river levee region monitoring data under each target estimation strategy by estimating the associated parameters;
Combining target river levee encroachment event data of the fuzzy river levee region monitoring data under each target estimation strategy to generate a river levee encroachment event data sequence of the fuzzy river levee region monitoring data, wherein the river levee encroachment event data sequence comprises target river levee encroachment event data corresponding to each target estimation strategy;
determining a first LOSS value between target river levee encroachment event data corresponding to each target estimation strategy and the reference river levee encroachment event data;
fusing the first LOSS value to generate a fused LOSS value;
combining the fusion LOSS value, updating traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, and generating a video motion feature analysis model for deployment;
and combining the fusion LOSS value, updating traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, and before generating the video motion feature analysis model for deployment, the method further comprises the following steps:
performing monitoring priori frame expansion on the fuzzy river levee region monitoring data to generate expanded river levee region monitoring data;
Estimating the encroaching event of the monitoring data of the extended river dike region according to a river dike encroaching decision branch of the video motion characteristic analysis model to be trained by combining the monitoring data of the extended river dike region, and generating estimated river dike encroaching event data of the monitoring data of the extended river dike region;
acquiring river levee encroaching event priori labeling data corresponding to the extended river levee region monitoring data, and determining a second LOSS value between the estimated river levee encroaching event data and the river levee encroaching event priori labeling data;
and combining the fusion LOSS value, updating traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, and generating a video motion feature analysis model for deployment, comprising:
and combining the fusion LOSS value and the second LOSS value, updating traversing model weight information of the video motion feature analysis model to be trained until the model convergence requirement is met, and generating a video motion feature analysis model for deployment.
10. A riverbank intrusion positioning and pushing system based on unmanned aerial vehicle monitoring, characterized in that the riverbank intrusion positioning and pushing system based on unmanned aerial vehicle monitoring comprises a processor and a machine-readable storage medium, in which machine-executable instructions are stored, which are loaded and executed by the processor to implement the riverbank intrusion positioning and pushing method based on unmanned aerial vehicle monitoring of any one of claims 1-9.
CN202310173901.XA 2023-02-28 2023-02-28 River dike encroaching, positioning and pushing method and system based on unmanned aerial vehicle monitoring Active CN115908574B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310173901.XA CN115908574B (en) 2023-02-28 2023-02-28 River dike encroaching, positioning and pushing method and system based on unmanned aerial vehicle monitoring

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310173901.XA CN115908574B (en) 2023-02-28 2023-02-28 River dike encroaching, positioning and pushing method and system based on unmanned aerial vehicle monitoring

Publications (2)

Publication Number Publication Date
CN115908574A CN115908574A (en) 2023-04-04
CN115908574B true CN115908574B (en) 2023-05-09

Family

ID=85733771

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310173901.XA Active CN115908574B (en) 2023-02-28 2023-02-28 River dike encroaching, positioning and pushing method and system based on unmanned aerial vehicle monitoring

Country Status (1)

Country Link
CN (1) CN115908574B (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020042024A1 (en) * 2018-08-29 2020-03-05 区链通网络有限公司 Node abnormality detection method and device based on graph algorithm and storage device
CN114065307A (en) * 2021-11-18 2022-02-18 福州大学 Hardware Trojan horse detection method and system based on bipartite graph convolutional neural network
CN115184568A (en) * 2022-06-14 2022-10-14 河北省水利工程局集团有限公司 Real-time monitoring method for river water environmental pollutants
CN115394348A (en) * 2022-07-15 2022-11-25 中南大学 IncRNA subcellular localization prediction method, equipment and medium based on graph convolution network
CN115564838A (en) * 2022-12-06 2023-01-03 深圳联和智慧科技有限公司 Unmanned aerial vehicle-based river bank detection encroachment positioning method and system
CN115661916A (en) * 2021-07-07 2023-01-31 阿里巴巴新加坡控股有限公司 Information processing method and device

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020042024A1 (en) * 2018-08-29 2020-03-05 区链通网络有限公司 Node abnormality detection method and device based on graph algorithm and storage device
CN115661916A (en) * 2021-07-07 2023-01-31 阿里巴巴新加坡控股有限公司 Information processing method and device
CN114065307A (en) * 2021-11-18 2022-02-18 福州大学 Hardware Trojan horse detection method and system based on bipartite graph convolutional neural network
CN115184568A (en) * 2022-06-14 2022-10-14 河北省水利工程局集团有限公司 Real-time monitoring method for river water environmental pollutants
CN115394348A (en) * 2022-07-15 2022-11-25 中南大学 IncRNA subcellular localization prediction method, equipment and medium based on graph convolution network
CN115564838A (en) * 2022-12-06 2023-01-03 深圳联和智慧科技有限公司 Unmanned aerial vehicle-based river bank detection encroachment positioning method and system

Also Published As

Publication number Publication date
CN115908574A (en) 2023-04-04

Similar Documents

Publication Publication Date Title
CN114529825B (en) Target detection model, method and application for fire fighting access occupied target detection
KR102141324B1 (en) Fast computation of convolutional neural networks
WO2023138300A1 (en) Target detection method, and moving-target tracking method using same
CN110309747B (en) Support quick degree of depth pedestrian detection model of multiscale
CN112232426B (en) Training method, device and equipment of target detection model and readable storage medium
CN110956255A (en) Difficult sample mining method and device, electronic equipment and computer readable storage medium
CN113852432A (en) RCS-GRU model-based spectrum prediction sensing method
CN114861172B (en) Data processing method and system based on government affair service system
KR20210093875A (en) Video analysis methods and associated model training methods, devices, and devices
CN111428854A (en) Structure searching method and structure searching device
CN114169389A (en) Class-expanded target detection model training method and storage device
CN115937703A (en) Enhanced feature extraction method for remote sensing image target detection
CN110146855B (en) Radar intermittent interference suppression threshold calculation method and device
CN114639102A (en) Cell segmentation method and device based on key point and size regression
US20210326475A1 (en) Systems and method for evaluating identity disclosure risks in synthetic personal data
CN115908574B (en) River dike encroaching, positioning and pushing method and system based on unmanned aerial vehicle monitoring
CN116542673B (en) Fraud identification method and system applied to machine learning
CN116051118B (en) Analysis method and device of behavior time sequence model
CN117313141A (en) Abnormality detection method, abnormality detection device, abnormality detection equipment and readable storage medium
CN116797973A (en) Data mining method and system applied to sanitation intelligent management platform
CN115909347A (en) Instrument reading identification method, device, equipment and medium
CN112464989A (en) Closed loop detection method based on target detection network
CN116051880A (en) Result prediction method based on uncertainty evaluation under label noise
CN114549849A (en) Image recognition method and device, computer equipment and storage medium
CN116736174B (en) Method, apparatus, computer device and storage medium for predicting remaining life of battery

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