An urban waterlogging monitoring and risk analysis method and system based on artificial intelligence

Abstract

The present application relates to the field of artificial intelligence, in particular to a city waterlogging monitoring and risk analysis method and system based on artificial intelligence, comprising the following steps: acquiring river image and time, site, delay, determining offset reference according to river bank line, bridge pier and road edge, screening effective image, identifying water body, flood, bare shore and carrier response, dividing candidate overflow area, determining real overflow area according to boundary advancing speed and rising reference, generating risk level combined with carrier overlap, floating object and collapse object and uploading sampling instruction.In the present application, through perspective offset verification, time sequence speed verification and carrier flooding relationship linkage analysis, camera offset, transmission delay, shadow reflection and bare shore misjudgment are reduced, risk level is freed from single frame coverage area dependence, overflow expansion trend and on-site disaster object change are taken into account, early warning accuracy is improved, and uploading granularity and sampling interval are adjusted according to risk state, reducing invalid transmission pressure.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based method and system for urban flooding monitoring and risk analysis. Background Technology

[0002] The field of artificial intelligence technology involves computer processing methods that utilize deep learning models, multimodal computing models, and training samples to extract features, train parameters, and infer outputs from image data, video data, and environmental status data. Traditional urban flooding monitoring and risk analysis methods involve deploying cameras, water level sensors, and monitoring terminals in rivers, roads, or low-lying areas. The cameras collect images or video streams and transmit them to a server. The server performs format conversion, resolution standardization, and data storage on the returned images. Then, it uses a trained recognition model to identify normal water bodies, flooded areas, inundated roads, farmland, and street areas, generating risk level data based on the flood coverage, inundation depth, and affected objects.

[0003] Current urban flood monitoring typically relies on images or videos transmitted from cameras and water level sensors. The server then uses a standardized format to identify the coverage area of ​​water bodies and roads. However, when cameras are affected by vibration, rain, fog, or transmission delays, the image sequence cannot keep pace with the actual water level changes. The model is prone to misjudging shadows, exposed riverbanks, floating debris, or reflections of bridge piers as overflowing water. Risk levels are often generated based on the coverage area of ​​a single frame, lacking continuous verification of the relationship between boundary advancement and the flooding of load-bearing structures. The fixed upload frequency also increases communication pressure. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide an artificial intelligence-based method for urban flooding monitoring and risk analysis; To achieve the above objectives, the present invention adopts the following technical solution: an artificial intelligence-based method for urban flooding monitoring and risk analysis, comprising the following steps: The river image frames, image generation time, station identification and image transmission delay formed at the same flash flood station at continuous sampling time are obtained, and the view offset reference is determined based on the riverbank position, bridge pier bottom position and road edge position obtained by installation calibration. The effective river image frame that meets the effective viewpoint condition is determined by calculating the position difference between the fixed reference position in the river image frame and the viewpoint offset benchmark. Based on the effective river image frames, the normal water body response, flood response, exposed riverbank response, and carrying capacity response are determined, and candidate overflow water body areas are divided based on the boundary extension distance between the normal water body response and the flood response. Based on the candidate overflow water body region corresponding to continuous sampling time, the water body boundary advance speed is analyzed, and the water body boundary advance speed is compared with the rise speed benchmark generated by the average boundary displacement of adjacent sampling time to determine the real overflow water body region. The flash flood risk level is determined based on the overlap between the actual overflow water area and the calibrated position of the carrier, the advance speed of the water boundary, the response of floating objects and the response of the collapsed body, and the flash flood risk level is used to generate instructions for adjusting the image upload granularity and sampling interval.

[0005] As a further aspect of the present invention, the step of determining the viewing angle offset reference includes: Obtain river image frames corresponding to the same station identifier, write the image generation time into the frame sequence table in ascending order of sampling time, and bind the image transmission delay with the corresponding sampling time to generate a station time-series image record; The positions of the riverbank, the bottom of the bridge pier, and the edge of the road are read from the installation and calibration results. The pixel coordinate ranges of the riverbank, the bottom of the bridge pier, and the edge of the road in the reference image are extracted respectively. A fixed reference coordinate group is established based on the relative distance between each pixel coordinate range. The fixed reference coordinate group is associated with the time-series image records of the site, and the fixed reference coordinate group is determined as the viewpoint offset reference.

[0006] As a further aspect of the present invention, the generation process of the site time-series image record includes: Obtain the image generation time written by the edge camera terminal and the image reception time written by the upload server, calculate the time difference between the image generation time and the image reception time, and determine the time difference as the image transmission delay; When the image transmission delay is greater than the sampling interval between adjacent sampling times, the corresponding river image frame is marked as a delayed frame, and the original sampling time of the delayed frame is retained in the frame sequence table to generate a delay mark field; Write the delay flag field, image generation time, image transmission delay, and site identifier into the same time series record, and update the site time series image record.

[0007] As a further aspect of the present invention, the step of determining a valid river channel image frame that satisfies the valid viewpoint condition includes: In each river image frame, the fixed reference detection positions corresponding to the riverbank calibration section, the bridge pier bottom calibration section, and the road edge calibration section are extracted, and the fixed reference detection positions are matched with the corresponding pixel coordinate intervals in the view offset reference to obtain the reference matching relationship. Based on the reference object matching relationship, calculate the position offset of each fixed reference object detection position relative to the corresponding pixel coordinate interval, and write each position offset into the offset record according to the reference object type to generate a view offset group; When the offset of each position in the view offset group does not exceed the corresponding view offset threshold, the corresponding river image frame is determined as a valid river image frame.

[0008] As a further aspect of the present invention, the process of generating the viewpoint allowable offset threshold includes: Acquire calibration repeat sampling images formed during installation and calibration, calculate the pixel position dispersion of the same riverbank calibration section, the same bridge pier bottom calibration section, and the same road edge calibration section in different calibration repeat sampling images, and determine the pixel position dispersion as the calibration stability parameter; Based on the calibration stabilization parameters and the resolution of the camera terminal, the allowable offset base value corresponding to the reference object type is determined, and the allowable offset base value is written into the threshold configuration table to establish the mapping relationship between the reference object type and the viewpoint allowable offset threshold. When calculating the viewpoint offset group, the viewpoint allowable offset threshold corresponding to the detection position of the fixed reference object is read from the threshold configuration table.

[0009] As a further aspect of the present invention, the step of dividing the candidate overflow water body area includes: The pixel regions located within the river calibration range in the effective river image frames are obtained. The normal water body response is determined based on the continuity of pixel brightness, the smoothness of water surface texture, and the intensity of edge reflection. The flood response is determined based on the wet diffusion area that is outside the calibration range of the riverbank and is continuously adjacent to the normal water body response. Obtain the bank slope area located outside the riverbank calibration range and with texture roughness reaching the riverbank texture threshold in the effective river channel image frame, determine the response of the exposed riverbank, and obtain the stable contour area that coincides with the calibration position of the bridge pier, road surface or guardrail, to determine the response of the load-bearing body. Calculate the continuous boundary position between the normal water body response and the flood response, and count the boundary expansion distance of the flood response in the direction away from the normal water body response. When the boundary expansion distance reaches the overflow judgment distance threshold, the area where the corresponding flood response is located is divided into candidate overflow water body areas.

[0010] As a further aspect of the present invention, the step of determining the actual overflow water area includes: Multiple candidate overflow water bodies corresponding to the same station identifier within adjacent sampling times are obtained. A temporal correspondence relationship of overflow areas is established based on the area of ​​overlap and the direction of boundary connectivity. Isolated candidate areas that do not form a correspondence relationship in adjacent sampling times are eliminated to generate a continuous candidate area sequence. Extract the outer boundary of each candidate region in the continuous candidate region sequence along the direction away from the normal water body response, calculate the displacement distance of the outer boundary between adjacent sampling times, and determine the water body boundary advancement speed by combining the time interval between the corresponding image generation times. The water body boundary advance speed is compared with the rise speed benchmark. When the water body boundary advance speed reaches the rise speed benchmark and the corresponding candidate region remains outwardly connected during continuous sampling, the corresponding candidate region is determined as the actual overflow water body region.

[0011] As a further aspect of the present invention, the time correction process for the water body boundary propulsion velocity includes: Obtain the image generation time and image transmission delay corresponding to adjacent sampling times, use the image generation time as the time base corresponding to the boundary displacement, and write the image transmission delay into the delay verification field of the corresponding boundary displacement record to generate a boundary displacement time record. When the difference in image transmission delay between adjacent boundary displacement records exceeds the delay fluctuation threshold, the sampling order of the corresponding candidate overflow water body area is rearranged according to the image generation time, and boundary displacement records with reversed time order are removed to obtain a corrected boundary displacement sequence. The water body boundary advance velocity is recalculated based on the corrected boundary displacement sequence.

[0012] As a further aspect of the present invention, the step of determining the flash flood risk level and generating image upload granularity and sampling interval adjustment instructions includes: Obtain the overlap area, overlap duration, and overlap location type between the actual overflow water area and the calibrated location of the carrier, and determine the flooding risk value of the carrier based on the overlap area, overlap duration, and overlap location type; The water body boundary advance velocity, floating object response area, and collapse body response area are obtained. The advance risk value, floating object blockage risk value, and collapse impact risk value are determined respectively. The flooding risk value of the bearing body, the advance risk value, floating object blockage risk value, and collapse impact risk value are jointly determined as the flash flood risk assessment parameters. The flash flood risk assessment parameters are compared with the preset risk level range to determine the flash flood risk level. Based on the flash flood risk level, the image upload granularity and sampling interval are read from the upload sampling strategy table to generate image upload granularity and sampling interval adjustment instructions.

[0013] An artificial intelligence-based urban flooding monitoring and risk analysis system includes: The perspective calibration module acquires river image frames, image generation time, station identifier and image transmission delay formed at the same flash flood station at continuous sampling time, and determines the perspective offset benchmark based on the riverbank position, bridge pier bottom position and road edge position obtained by installation calibration. The water body identification module calculates the position difference between the fixed reference position in the river image frame and the view offset benchmark to determine the valid river image frame that meets the view validity condition. The overflow delineation module determines the normal water body response, flood response, exposed riverbank response, and carrying capacity response based on the effective river channel image frames, and delineates candidate overflow water body areas based on the boundary expansion distance between the normal water body response and the flood response. The advancement determination module analyzes the advancement speed of the water body boundary based on the candidate overflow water body region corresponding to continuous sampling time, and compares the advancement speed of the water body boundary with the rising speed benchmark generated by the average boundary displacement of adjacent sampling time to determine the real overflow water body region. The risk rating module determines the flash flood risk level based on the overlap between the actual overflow water area and the calibrated position of the carrier, the advance speed of the water boundary, the response of floating objects and the response of the collapsed body, and generates instructions for adjusting the image upload granularity and sampling interval based on the flash flood risk level.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, continuous sampling of river images is correlated with generation time, station identification, and transmission delay. Combined with the riverbank, bridge pier bottom, and road edge to form a viewpoint offset benchmark, the position difference of fixed reference objects is effectively screened to ensure that the images entering the analysis have a stable viewpoint and reliable time sequence. Boundary discrimination relationships are established between water body response, flood response, exposed riverbank response, and load-bearing body response. The average displacement of the overflow boundary at adjacent times is used to form the rise rate benchmark, eliminating interference from pseudo-water bodies such as shadows, reflections, and debris. The risk level is jointly constrained by the overlap of the overflow area and the load-bearing body, the advance speed, floating objects, and collapse bodies. The upload granularity and sampling interval are adjusted synchronously to improve the accuracy of early warning and reduce invalid transmission. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the overall process of an AI-based urban flood monitoring and risk analysis method. Figure 2 Establish a flowchart for site time-series recording and viewpoint offset benchmark; Figure 3 Flowchart for perspective verification and effective river channel image frame selection; Figure 4 Flowchart for water body response identification and candidate overflow water body region delineation; Figure 5 Flowchart for continuous sampling to advance analysis and confirm actual spillover; Figure 6 Flowchart for risk rating and sampling adjustment; Figure 7 This is a block diagram of an AI-based urban flood monitoring and risk analysis system. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments. Example

[0019] Please see Figures 1 to 7 This embodiment provides an AI-based method and system for urban flooding monitoring and risk analysis. In the actual operation of urban flash flood monitoring stations continuously recording images of rivers, bridge piers, riverbanks, and adjacent roads, factors such as water surface reflection, rain and fog obstruction, sampling delays, and framing angle shifts can cause overflowing water to be confused with wet riverbanks, road flooding, or shaded areas. This embodiment first verifies the framing location, then identifies the water body response, and subsequently uses the boundary progression relationship of continuous sampling times to confirm the true overflow area. Furthermore, it incorporates the status of submerged structures, floating objects, and collapsed structures into risk classification, enabling the image upload granularity and sampling interval to form a closed-loop adjustment according to the flash flood risk level. This includes an AI-based method and system for urban flooding monitoring and risk analysis.

[0020] S1: Acquire river image frames, image generation time, station identifier, and image transmission delay generated at consecutive sampling times from the same flash flood station. Determine the viewing angle offset benchmark based on the riverbank position, bridge pier bottom position, and road edge position obtained from installation and calibration. Here, a river image frame refers to image data collected at the same flash flood station that reflects the river surface, riverbank, bridge pier bottom, and road edge; image generation time refers to the sampling time written when the river image frame is formed; station identifier is a unique record field used to distinguish different flash flood stations; image transmission delay refers to the time difference between image generation time and image reception time; and the viewing angle offset benchmark refers to the benchmark data composed of the fixed reference object pixel coordinate range and relative distance relationships formed during the installation and calibration phase, used to subsequently determine whether the framing position remains within the allowable range. This step binds the image, time, and delay fields under the same station identifier, forming a common input for subsequent viewing angle verification, temporal sequencing, and boundary advancement analysis.

[0021] S101: Obtain river image frames corresponding to the same station identifier, write the image generation time into the frame sequence table in ascending order of sampling time, and bind the image transmission delay to the corresponding sampling time to generate a station time-series image record. The station time-series image record refers to a time-series record with the station identifier as the attribution field, the image generation time as the sorting basis, and the river image frame and image transmission delay as content fields. The frame sequence table receives consecutive river image frames under the same station identifier, first determines the sampling order by the image generation time, and then writes the image transmission delay into the corresponding record, so that subsequent steps can call images according to the actual sampling order, rather than just according to the receiving order.

[0022] S102: Read the riverbank position, pier base position, and road edge position from the installation calibration results. Extract the pixel coordinate intervals of the riverbank position, pier base position, and road edge position in the reference image, and establish a fixed reference coordinate group based on the relative distance between each pixel coordinate interval. The fixed reference coordinate group refers to the reference data composed of the pixel coordinate intervals of the riverbank calibration segment, pier base calibration segment, and road edge calibration segment, as well as the relative positional relationships between these pixel coordinate intervals. The pixel coordinate intervals are not used as individual points, but as range fields covering the corresponding fixed reference object contours to absorb minor imaging differences during installation calibration.

[0023] S103: Establish a connection between the fixed reference coordinate group and the time-series image records of the station, and determine the fixed reference coordinate group as the viewpoint offset reference. After the connection is completed, each river image frame in the time-series image records of the same station can call the same viewpoint offset reference. It is not necessary to reselect the reference object when judging the viewpoint offset in the future, so as to maintain the consistency of the verification caliber of the same station in the continuous sampling process.

[0024] S104: Obtain the image generation time written by the edge camera terminal and the image reception time written by the upload server, calculate the time difference between the image generation time and the image reception time, and determine the time difference as the image transmission delay. The calculation process here is based on the order of the time fields, and the output result is written to the image transmission delay field. The image transmission delay field does not directly change the image content, but is only used for subsequent judgment of sampling order, delay frame status, and boundary displacement time record.

[0025] S105: When the image transmission delay exceeds the sampling interval between adjacent sampling times, the corresponding river image frame is marked as a delayed frame, and the original sampling time of the delayed frame is retained in the frame sequence table, generating a delay marker field. The delay marker field is a status field indicating that the river image frame has a transmission lag but still retains the original image generation time. Delayed frames are not moved to incorrect sampling positions due to their later reception order; subsequent steps still use the image generation time as the time reference when calculating the water body boundary advancement velocity.

[0026] S106: Write the delay marker field, image generation time, image transmission delay, and site identifier into the same time-series record, and update the site time-series image record. When there are missing image generation times, unmatched site identifiers, or river image frames written repeatedly at the same sampling time, prioritize retaining records that can match both the site identifier and image generation time. Duplicate records are selected as valid candidate records for subsequent steps based on image sharpness and delay marker status. Records that cannot be time-bound are not included in the boundary advancement analysis.

[0027] In this embodiment, by writing the image generation time, image transmission delay, and fixed reference coordinate set into the station time-series image record, the river image frame has a clear time assignment and viewpoint reference before entering the water body identification, thereby providing a traceable data starting point for subsequent continuous boundary judgment.

[0028] S2: Calculate the position difference between the fixed reference points in the river image frame and the viewpoint offset reference to determine the valid river image frames that meet the viewpoint validity condition. The fixed reference point position refers to the location range of the detected riverbank calibration section, bridge pier bottom calibration section, and road edge calibration section in the current river image frame. The viewpoint validity condition refers to the judgment rule that the offset state of each fixed reference point's detected position relative to the viewpoint offset reference is within the allowable range for the corresponding reference point type. A valid river image frame is a river image frame that has passed viewpoint verification and can be used for water body response recognition.

[0029] S201: Extract the fixed reference object detection positions corresponding to the riverbank calibration section, bridge pier bottom calibration section, and road edge calibration section in each river image frame, and match the fixed reference object detection positions with the corresponding pixel coordinate intervals in the viewpoint offset reference to obtain the reference object matching relationship. The reference object matching relationship refers to the corresponding state between the current detection position and the reference pixel coordinate interval, and includes at least the reference object type, detection position range, reference position range, and valid matching state. If a fixed reference object cannot form a stable detection position due to occlusion, rain / fog, strong reflection, or image truncation, the reference object matching relationship is written as missing, and the current river image frame does not directly enter the real overflow water area for judgment.

[0030] S202: Calculate the positional offset of each fixed reference object detection position relative to its corresponding pixel coordinate interval based on the reference object matching relationship, and write each positional offset into an offset record according to the reference object type, generating a viewpoint offset group. The positional offset refers to the deviation of the current detection position from the reference pixel coordinate interval in the image plane. The offset record includes the reference object type, offset direction, offset state, and the corresponding threshold call source. The viewpoint offset group aggregates the offset states of different fixed reference objects in the same river image frame to avoid judging the effectiveness of framing based on a single reference object.

[0031] S203: When the offset of each position in the viewpoint offset group does not exceed the corresponding viewpoint allowable offset threshold, the corresponding river image frame is determined as a valid river image frame. The viewpoint allowable offset threshold refers to the allowable offset boundary formed during the installation and calibration phase and written into the threshold configuration table according to the reference object type. If any reference object offset exceeds the corresponding allowable boundary, or if the reference object matching relationship is missing, the river image frame is marked as an invalid viewpoint frame and is only retained in the station's time-series image record for tracing purposes, not as a basis for advancing the overflow boundary.

[0032] S204: Acquire repeated sampling images generated during installation and calibration. Calculate the pixel position dispersion of the same riverbank calibration segment, the same bridge pier bottom calibration segment, and the same road edge calibration segment in different repeated sampling images, and determine the pixel position dispersion as the calibration stability parameter. The calibration stability parameter refers to the stable imaging position of the same fixed reference object during the installation and calibration phase, used to indicate whether the reference object itself is suitable as a basis for viewpoint calibration. The calibration stability parameter is derived from the installation and calibration results, not from subsequent risk assessment results, to avoid subsequent water changes negatively affecting the viewpoint calibration standard.

[0033] S205: Based on the calibration stability parameters and camera terminal resolution, determine the allowable offset baseline value corresponding to the reference object type, and write the allowable offset baseline value into the threshold configuration table to establish a mapping relationship between the reference object type and the viewpoint allowable offset threshold. The threshold configuration table stores the allowable offset boundaries of the riverbank calibration section, the bridge pier bottom calibration section, and the road edge calibration section according to the reference object type, and records its source as installation calibration stability and image resolution configuration. The mapping relationship remains consistent within the same site, and when the installation calibration results are updated, the threshold configuration table is updated with the new calibration record.

[0034] S206: When calculating the viewpoint offset group, read the viewpoint allowable offset threshold corresponding to the fixed reference object detection position from the threshold configuration table. If the threshold configuration table lacks a record for the corresponding reference object type, the current reference object will not participate in the valid frame confirmation, and the relevant river channel image frame will enter the pending verification state until the installation calibration record is supplemented and then the viewpoint verification is performed again.

[0035] In this embodiment, by matching the position of the fixed reference object with the viewpoint offset benchmark item by item, the water position error caused by the framing deflection is eliminated before entering the overflow identification, so that the subsequent water boundary advancement judgment is established under the same viewpoint.

[0036] S3: Based on valid river channel image frames, determine the normal water body response, flood response, exposed bank response, and load-bearing body response, and delineate candidate overflow water body regions based on the boundary extension distance between the normal water body response and the flood response. The normal water body response refers to the image region state located within the river channel calibration range and meeting the characteristics of continuous water brightness, smooth texture, and edge reflection. The flood response refers to the moist diffusion region state that extends beyond the riverbank calibration range and is continuously adjacent to the normal water body response. The exposed bank response refers to the region located outside the riverbank calibration range and exhibiting a rough bank texture. The load-bearing body response refers to the region state that overlaps with the calibration locations of bridge piers, road surfaces, or guardrails and presents a stable outline. Candidate overflow water body regions refer to regions that meet the criteria for extension direction and extension distance but have not yet been verified through continuous sampling.

[0037] S301: Obtain pixel regions within the river calibration range in valid river image frames. Determine normal water body responses based on pixel brightness continuity, water surface texture smoothness, and edge reflection intensity. Determine flood response based on wet diffusion regions that extend beyond the riverbank calibration range and are continuously adjacent to normal water body responses. Brightness continuity, water surface texture smoothness, and edge reflection intensity are first converted into water body response state fields within the same image, and then used to identify normal water body responses according to a preset judgment order, avoiding directly using brightness changes under different image conditions as a basis for cross-image comparison. For flood response, only wet diffusion regions that are spatially adjacent to normal water body responses and located outside the riverbank are included in the candidate spillover judgment; isolated reflections, shadows, and local wet patches are not considered as evidence of continuous spillover.

[0038] S302: Obtain the bank slope area located outside the riverbank calibration range in the valid river image frame and whose texture roughness reaches the riverbank texture threshold, determine the bare riverbank response, and obtain the stable contour area that coincides with the calibration position of bridge piers, road surfaces, or guardrails to determine the load-bearing body response. The riverbank texture threshold is derived from the texture state configuration of the bare bank slope area in the installation calibration stage and historical valid images, used to distinguish between bank slopes not covered by water and areas of wet diffusion. The load-bearing body response is not used to directly prove that overflow has occurred, but is used to subsequently determine whether the actual overflow water area has affected the calibration position of load-bearing bodies such as bridge piers, road surfaces, or guardrails.

[0039] S303: Calculate the continuous boundary position between the normal water body response and the flood response, and calculate the boundary expansion distance of the flood response along the direction away from the normal water body response. When the boundary expansion distance reaches the overflow determination distance threshold, the area where the corresponding flood response is located is classified as a candidate overflow water body area. Here, the continuous boundary position refers to the boundary zone where the normal water body response and the flood response are adjacent to each other. The boundary expansion distance refers to the range formed by the expansion from this boundary zone outwards from the riverbank. The overflow determination distance threshold is derived from the riverbank calibration range and the station installation calibration rules. If the flood response only occurs briefly in a local location and does not expand outwards along the direction away from the normal water body response, the area is retained as a suspected flood record and does not enter the candidate overflow water body area.

[0040] In this embodiment, by limiting the responses of normal water bodies, floodwaters, exposed riverbanks, and carrier bodies to different calibration ranges and texture states, the wet bank slopes, road edge shadows, and real outward-expanding water bodies are distinguished from the real outward-expanding water bodies in the candidate stage, thereby providing candidate areas with clear boundaries for subsequent continuous verification.

[0041] S4: Analyze the water boundary advancement velocity based on the candidate overflow water body regions corresponding to continuous sampling times, and compare the water boundary advancement velocity with the rising velocity benchmark generated from the average boundary displacement at adjacent sampling times to determine the true overflow water body region. Here, the water boundary advancement velocity refers to the speed at which the boundary of the same candidate overflow water body region advances along the direction away from the normal water body response at adjacent sampling times; the rising velocity benchmark is a benchmark field generated from the average boundary displacement at adjacent sampling times and used to determine whether the water body expansion has a continuous rising characteristic; and the true overflow water body region refers to the overflow water body region confirmed after verification through temporal correspondence, boundary advancement, and expansion connectivity.

[0042] S401: Obtain multiple candidate overflow water areas corresponding to the same site identifier within adjacent sampling times. Establish a temporal correspondence between overflow areas based on the overlapping area and boundary connectivity direction, and eliminate isolated candidate areas that do not form a correspondence between adjacent sampling times, generating a continuous candidate area sequence. The temporal correspondence between overflow areas refers to the corresponding state of candidate overflow water areas in adjacent sampling times that can be identified as belonging to the same overflow process in terms of spatial coverage and outward expansion direction. Isolated candidate areas fail to maintain a spatial continuity between adjacent sampling times and cannot prove continuous water body expansion; therefore, they do not enter the actual overflow confirmation process.

[0043] S402: Extract the outer boundary of each candidate region in the continuous candidate region sequence along the direction away from the normal water body response. Calculate the displacement distance of the outer boundary between adjacent sampling times, and determine the water body boundary advancement speed by combining this with the time interval between corresponding image generation times. The outer boundary refers to the boundary band of the candidate overflow water body region away from the normal water body response. The displacement distance is recorded in the boundary displacement record and simultaneously bound to the image generation time of the preceding and following sampling times. The water body boundary advancement speed is formed by the boundary displacement record and the image generation time interval, used to indicate whether the water body expansion has a continuous advancement state.

[0044] S403: Compare the water boundary advance velocity with the rise velocity benchmark. When the water boundary advance velocity reaches the rise velocity benchmark and the corresponding candidate region remains outwardly connected during continuous sampling, the corresponding candidate region is determined as the actual overflow water region. The rise velocity benchmark is formed from the initial boundary state that has been installed and calibrated during the initial phase. When the sequence of consecutive candidate regions is insufficient, the obtained adjacent boundary displacement records are used for temporary benchmark updates. After a complete and usable sequence is formed during continuous sampling, it switches to a stable benchmark generated from the average boundary displacement of adjacent sampling times. If the water boundary advance velocity does not reach the rise velocity benchmark, or the candidate region does not remain outwardly connected, the candidate region is retained as an unconfirmed overflow record.

[0045] S404: Obtain the image generation time and image transmission delay corresponding to adjacent sampling times, use the image generation time as the time base for the boundary displacement, and write the image transmission delay into the delay verification field of the corresponding boundary displacement record to generate a boundary displacement time record. The boundary displacement time record is a time-series record composed of the outer boundary of the candidate region, the boundary displacement state, the image generation time, and the delay verification field, used to avoid the order of image reception affecting the actual sampling order.

[0046] S405: When the difference in image transmission delay between adjacent boundary displacement records exceeds the delay fluctuation threshold, the sampling order of the corresponding candidate overflow water areas is rearranged according to the image generation time, and boundary displacement records with reversed time order are removed to obtain a corrected boundary displacement sequence. The delay fluctuation threshold is derived from the site sampling interval configuration and transmission stability configuration. Reversed time order refers to a record state where the receiving order is inconsistent with the image generation time order, which could lead to misjudgment of the boundary advancement direction. The removed boundary displacement records retain a delay verification field for subsequent traceability but are not used to confirm the water body boundary advancement speed.

[0047] S406: Recalculate the water body boundary advance velocity based on the corrected boundary displacement sequence. During recalculation, only records with consistent time sequence, identifiable outer boundaries, and maintained outward connectivity of candidate regions are used. If the corrected continuous candidate region sequence is insufficient, the current sampling round will not output a new actual overflow water body region; instead, the previous valid confirmation state will be used as the input state for risk assessment, and stable operation rules will be restored after new valid sampling is obtained.

[0048] In this embodiment, the water body boundary advancement process is constrained by the region overlap, boundary connectivity direction and image generation time, so that short-term reflection, transmission delay and isolated wet areas will not be directly identified as real overflow, thus providing continuous sampling basis and time correction basis for real overflow water areas.

[0049] S5: Determine the flash flood risk level based on the overlap between the actual overflow water area and the calibrated location of the carrying capacity, the advance speed of the water boundary, the response of floating objects, and the response of collapsed structures. Then, generate image upload granularity and sampling interval adjustment instructions based on the flash flood risk level. Floating object response refers to the state of a non-fixed contour area that moves with the water in the actual overflow water area or its adjacent water bodies. Collapse response refers to the state of a bank slope, revetment, or adjacent structure in the image that exhibits a broken contour, abrupt location, or accumulation and diffusion. The flash flood risk level is a risk state jointly mapped from the carrying capacity flooding risk value, advance risk value, floating object blockage risk value, and collapse impact risk value. The image upload granularity and sampling interval adjustment instructions refer to the image upload clarity and sampling rhythm control results generated after calling the upload sampling strategy table according to the risk state.

[0050] S501: Obtain the overlap area, overlap duration, and overlap location type between the actual overflow water area and the marked location of the bearing structure. Determine the flood risk value of the bearing structure based on the overlap area, overlap duration, and overlap location type. The overlap area represents the extent to which the actual overflow water area covers the marked location of the bearing structure; the overlap duration represents the duration of this coverage during continuous sampling; and the overlap location type indicates whether the covered location belongs to the bottom of the bridge pier, the road surface, or the guardrail marked location. The flood risk value of the bearing structure is mapped textually according to a preset risk level range. The closer the marked location of the bearing structure is to the passageway or supporting area, and the more continuous the coverage, the higher the risk level that requires intervention.

[0051] S502: Obtain the water body boundary advance velocity, floating debris response area, and collapse body response area to determine the advance risk value, floating debris blockage risk value, and collapse impact risk value, respectively. These values, along with the load-bearing body flooding risk value, advance risk value, floating debris blockage risk value, and collapse impact risk value, are collectively defined as flash flood risk assessment parameters. The advance risk value reflects the state of the actual overflow water area continuing to advance towards the load-bearing body's designated location. The floating debris blockage risk value reflects the state of floating debris response blocking water passage and the area around bridge piers. The collapse impact risk value reflects the state of the collapse body response impacting water flow and the load-bearing body. All of the above risk values ​​are written into the flash flood risk assessment parameters according to the same risk level caliber; different image physical quantities are not directly substituted for each other.

[0052] S503: Compare the flash flood risk assessment parameters with the preset risk level range to determine the flash flood risk level. Based on the flash flood risk level, read the image upload granularity and sampling interval from the upload sampling strategy table and generate image upload granularity and sampling interval adjustment instructions. The preset risk level range is formed by the site risk management rules, the carrier calibration location type, and the configuration of historical valid spillover states. The upload sampling strategy table records the image upload granularity and sampling interval corresponding to different flash flood risk levels. When the flash flood risk level is in the basic monitoring state, the image upload granularity and sampling interval maintain the conventional strategy; when the flash flood risk level enters the spillover confirmation state, a finer upload granularity and a denser sampling strategy are read; when the flash flood risk level enters the carrier flooding or combined impact state, adjustment instructions for encrypted sampling and refined upload are output. If there are missing key inputs in the risk assessment parameters, the previous valid flash flood risk level is retained, and the missing fields are written to the risk assessment abnormal state. The flash flood risk level is re-determined after the next valid sampling is completed.

[0053] In this embodiment, by incorporating the overlap between the actual overflow water area and the calibrated location of the carrier, the water boundary advancement status, the floating object response, and the collapse body response into the flash flood risk assessment parameters, the flash flood risk level is made independent of a single water surface expansion phenomenon. This allows the image upload granularity and sampling interval to be adjusted according to the overflow impact object and the impact status.

[0054] This embodiment also provides an artificial intelligence-based urban flooding monitoring and risk analysis system, which is used to implement the aforementioned artificial intelligence-based urban flooding monitoring and risk analysis method. The system includes a perspective verification module, a water body identification module, an overflow demarcation module, a progress determination module, and a risk rating module. Each module transmits data based on the river image frame, time field, calibration location, and risk output results of the same flash flood station, without introducing any processing objects unrelated to the method.

[0055] The perspective verification module acquires river image frames, image generation times, station identifiers, and image transmission delays generated at consecutive sampling times from the same flash flood station. It then determines the perspective offset reference based on the riverbank position, bridge pier bottom position, and road edge position obtained from installation calibration. The module receives river image frames, image generation times, station identifiers, image reception times, and installation calibration results. It writes image records under the same station identifier into the station's time-series image record and outputs the fixed reference coordinate set formed by the riverbank calibration segment, bridge pier bottom calibration segment, and road edge calibration segment as the perspective offset reference to the water body identification module. The module also generates a delay marker field when the image transmission delay exceeds the sampling interval and retains the original image generation time, ensuring that subsequent modules call river image frames according to the actual sampling order.

[0056] The water body identification module calculates the position difference between the fixed reference object positions in the river image frames and the viewpoint offset benchmark to determine valid river image frames that meet the viewpoint validity conditions. The water body identification module receives the station time-series image records and viewpoint offset benchmark output by the viewpoint verification module. It extracts the fixed reference object detection positions in each river image frame, establishes reference object matching relationships, generates viewpoint offset sets, and reads the viewpoint allowable offset threshold for the corresponding reference object type from the threshold configuration table. The water body identification module only outputs valid river image frames where all reference object offset states are within the allowable range. For river image frames with missing reference objects, viewpoint offset, or missing threshold configurations, it outputs an invalid viewpoint or a state pending verification, and does not output it to the overflow delimitation module as an overflow judgment input.

[0057] The overflow demarcation module, based on valid river image frames, determines the normal water body response, flood response, exposed bank response, and carrying capacity response, and delineates candidate overflow water body areas based on the boundary expansion distance between the normal water body response and the flood response. The overflow demarcation module receives valid river image frames output by the water body identification module and uses the river channel calibration range, bank line calibration range, and carrying capacity calibration position to generate normal water body response, flood response, exposed bank response, and carrying capacity response according to the water brightness continuity, water surface texture smoothness, edge reflection, bank slope texture, and stable contour state, respectively. The overflow demarcation module identifies the expansion state of the flood response along the direction away from the normal water body response, outputting areas that reach the overflow determination distance threshold and are continuously adjacent to the normal water body response as candidate overflow water body areas, and also transmits the exposed bank response and carrying capacity response to the advancement determination module and risk classification module.

[0058] The advancement determination module analyzes the water body boundary advancement velocity based on the candidate overflow water body regions corresponding to continuous sampling times, and compares the water body boundary advancement velocity with the rising velocity benchmark generated by the average boundary displacement of adjacent sampling times to determine the actual overflow water body region. The advancement determination module receives candidate overflow water body regions, normal water body response boundaries, and station time-series image records output by the overflow demarcation module. It establishes a time-series correspondence between overflow regions based on the overlapping area and boundary connectivity direction, generating a continuous candidate region sequence. The advancement determination module writes the outer boundary displacement status, image generation time, and image transmission delay into the boundary displacement time record. When delay fluctuations affect the sampling order, the sampling order is rearranged according to the image generation time, and records with reversed time order are removed. The water body boundary advancement velocity is then determined using the corrected boundary displacement sequence. When the water body boundary advancement velocity reaches the rising velocity benchmark and the candidate region maintains outward connectivity, the advancement determination module outputs the actual overflow water body region to the risk classification module. When there are insufficient consecutive frames or the boundary is unidentifiable, it outputs a pending confirmation status and a call flag for the previous valid confirmation status.

[0059] The risk assessment module determines the flash flood risk level based on the overlap between the actual overflow water area and the calibrated location of the carrying capacity, the water boundary advance speed, the response of floating objects, and the response of collapsed structures. It also generates instructions for adjusting the image upload granularity and sampling interval based on the flash flood risk level. The module receives the actual overflow water area, water boundary advance speed, and correction status output by the advance judgment module, and the carrying capacity response output by the overflow demarcation module. It then identifies the floating object response and collapsed structure response in valid river image frames. The module writes the overlap area, overlap duration, and overlap location type between the actual overflow water area and the calibrated location of the carrying capacity into the carrying capacity flooding risk value. It writes the water boundary advance speed, floating object response area, and collapsed structure response area into the advance risk value, floating object blockage risk value, and collapsed impact risk value, respectively. These risk values ​​are then used together as flash flood risk assessment parameters, compared with a preset risk level range, and the flash flood risk level is determined. The risk grading module reads the upload sampling strategy table based on the flash flood risk level, outputs the image upload granularity and sampling interval adjustment instructions, and records the missing fields, pending confirmation status and the previous valid flash flood risk level as abnormal flow direction records to ensure that the risk grading still has a clear subsequent call relationship in the case of missing images or delayed fluctuations.

[0060] In this embodiment, the perspective verification module first forms a perspective offset benchmark and station time-series image records, the water body identification module then filters valid river image frames, the overflow delineation module outputs candidate overflow water areas, the advancement judgment module confirms the actual overflow water areas, and the risk rating module generates flash flood risk levels and adjustment instructions. This ensures that the data input and output between the modules are continuously transmitted along the image time sequence of the same flash flood station, thereby enabling the urban waterlogging monitoring and risk analysis process to have a complete closed loop of perspective verification, time sequence correction, overflow confirmation, and risk feedback.

[0061] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for urban flooding monitoring and risk analysis based on artificial intelligence, characterized in that, include: The river image frames, image generation time, station identification and image transmission delay formed at the same flash flood station at continuous sampling time are obtained, and the view offset reference is determined based on the riverbank position, bridge pier bottom position and road edge position obtained by installation calibration. The effective river image frame that meets the effective viewpoint condition is determined by calculating the position difference between the fixed reference position in the river image frame and the viewpoint offset benchmark. Based on the effective river image frames, the normal water body response, flood response, exposed riverbank response, and carrying capacity response are determined, and candidate overflow water body areas are divided based on the boundary extension distance between the normal water body response and the flood response. Based on the candidate overflow water body region corresponding to continuous sampling time, the water body boundary advance speed is analyzed, and the water body boundary advance speed is compared with the rise speed benchmark generated by the average boundary displacement of adjacent sampling time to determine the real overflow water body region. The flash flood risk level is determined based on the overlap between the actual overflow water area and the calibrated position of the carrier, the advance speed of the water boundary, the response of floating objects and the response of the collapsed body, and the flash flood risk level is used to generate instructions for adjusting the image upload granularity and sampling interval.

2. The method for urban flooding monitoring and risk analysis based on artificial intelligence according to claim 1, characterized in that, The step of determining the viewpoint offset reference includes: Obtain river image frames corresponding to the same station identifier, write the image generation time into the frame sequence table in ascending order of sampling time, and bind the image transmission delay with the corresponding sampling time to generate a station time-series image record; The positions of the riverbank, the bottom of the bridge pier, and the edge of the road are read from the installation and calibration results. The pixel coordinate ranges of the riverbank, the bottom of the bridge pier, and the edge of the road in the reference image are extracted respectively. A fixed reference coordinate group is established based on the relative distance between each pixel coordinate range. The fixed reference coordinate group is associated with the time-series image records of the site, and the fixed reference coordinate group is determined as the viewpoint offset reference.

3. The method for urban flooding monitoring and risk analysis based on artificial intelligence according to claim 2, characterized in that, The generation process of the site time-series image records includes: Obtain the image generation time written by the edge camera terminal and the image reception time written by the upload server, calculate the time difference between the image generation time and the image reception time, and determine the time difference as the image transmission delay; When the image transmission delay is greater than the sampling interval between adjacent sampling times, the corresponding river image frame is marked as a delayed frame, and the original sampling time of the delayed frame is retained in the frame sequence table to generate a delay mark field; Write the delay flag field, image generation time, image transmission delay, and site identifier into the same time series record, and update the site time series image record.

4. The method for urban flooding monitoring and risk analysis based on artificial intelligence according to claim 1, characterized in that, The step of determining a valid river channel image frame that satisfies the valid viewpoint condition includes: In each river image frame, the fixed reference detection positions corresponding to the riverbank calibration section, the bridge pier bottom calibration section, and the road edge calibration section are extracted, and the fixed reference detection positions are matched with the corresponding pixel coordinate intervals in the view offset reference to obtain the reference matching relationship. Based on the reference object matching relationship, calculate the position offset of each fixed reference object detection position relative to the corresponding pixel coordinate interval, and write each position offset into the offset record according to the reference object type to generate a view offset group; When the offset of each position in the view offset group does not exceed the corresponding view offset threshold, the corresponding river image frame is determined as a valid river image frame.

5. The method for urban flooding monitoring and risk analysis based on artificial intelligence according to claim 4, characterized in that, The process of generating the permissible viewpoint offset threshold includes: Acquire calibration repeat sampling images formed during installation and calibration, calculate the pixel position dispersion of the same riverbank calibration section, the same bridge pier bottom calibration section, and the same road edge calibration section in different calibration repeat sampling images, and determine the pixel position dispersion as the calibration stability parameter; Based on the calibration stabilization parameters and the resolution of the camera terminal, the allowable offset base value corresponding to the reference object type is determined, and the allowable offset base value is written into the threshold configuration table to establish the mapping relationship between the reference object type and the viewpoint allowable offset threshold. When calculating the viewpoint offset group, the viewpoint allowable offset threshold corresponding to the detection position of the fixed reference object is read from the threshold configuration table.

6. The method for urban flooding monitoring and risk analysis based on artificial intelligence according to claim 1, characterized in that, The step of delineating candidate overflow water body areas includes: The pixel regions located within the river calibration range in the effective river image frames are obtained. The normal water body response is determined based on the continuity of pixel brightness, the smoothness of water surface texture, and the intensity of edge reflection. The flood response is determined based on the wet diffusion area that is outside the calibration range of the riverbank and is continuously adjacent to the normal water body response. Obtain the bank slope area located outside the riverbank calibration range and with texture roughness reaching the riverbank texture threshold in the effective river channel image frame, determine the response of the exposed riverbank, and obtain the stable contour area that coincides with the calibration position of the bridge pier, road surface or guardrail, to determine the response of the load-bearing body. Calculate the continuous boundary position between the normal water body response and the flood response, and count the boundary expansion distance of the flood response in the direction away from the normal water body response. When the boundary expansion distance reaches the overflow judgment distance threshold, the area where the corresponding flood response is located is divided into candidate overflow water body areas.

7. The method for urban flooding monitoring and risk analysis based on artificial intelligence according to claim 1, characterized in that, The steps for determining the actual overflow water area include: Multiple candidate overflow water bodies corresponding to the same station identifier within adjacent sampling times are obtained. A temporal correspondence relationship of overflow areas is established based on the area of ​​overlap and the direction of boundary connectivity. Isolated candidate areas that do not form a correspondence relationship in adjacent sampling times are eliminated to generate a continuous candidate area sequence. Extract the outer boundary of each candidate region in the continuous candidate region sequence along the direction away from the normal water body response, calculate the displacement distance of the outer boundary between adjacent sampling times, and determine the water body boundary advancement speed by combining the time interval between the corresponding image generation times. The water body boundary advance speed is compared with the rise speed benchmark. When the water body boundary advance speed reaches the rise speed benchmark and the corresponding candidate region remains outwardly connected during continuous sampling, the corresponding candidate region is determined as the actual overflow water body region.

8. The method for urban flooding monitoring and risk analysis based on artificial intelligence according to claim 1, characterized in that, The time correction process for the water body boundary advance velocity includes: Obtain the image generation time and image transmission delay corresponding to adjacent sampling times, use the image generation time as the time base corresponding to the boundary displacement, and write the image transmission delay into the delay verification field of the corresponding boundary displacement record to generate a boundary displacement time record. When the difference in image transmission delay between adjacent boundary displacement records exceeds the delay fluctuation threshold, the sampling order of the corresponding candidate overflow water body area is rearranged according to the image generation time, and boundary displacement records with reversed time order are removed to obtain a corrected boundary displacement sequence. The water body boundary advance velocity is recalculated based on the corrected boundary displacement sequence.

9. The method for urban flooding monitoring and risk analysis based on artificial intelligence according to claim 8, characterized in that, The steps for determining the flash flood risk level and generating instructions for adjusting image upload granularity and sampling interval include: Obtain the overlap area, overlap duration, and overlap location type between the actual overflow water area and the calibrated location of the carrier, and determine the flooding risk value of the carrier based on the overlap area, overlap duration, and overlap location type; The water body boundary advance velocity, floating object response area, and collapse body response area are obtained. The advance risk value, floating object blockage risk value, and collapse impact risk value are determined respectively. The flooding risk value of the bearing body, the advance risk value, floating object blockage risk value, and collapse impact risk value are jointly determined as the flash flood risk assessment parameters. The flash flood risk assessment parameters are compared with the preset risk level range to determine the flash flood risk level. Based on the flash flood risk level, the image upload granularity and sampling interval are read from the upload sampling strategy table to generate image upload granularity and sampling interval adjustment instructions.

10. An artificial intelligence-based urban flooding monitoring and risk analysis system, characterized in that, The system is used to implement the artificial intelligence-based urban flooding monitoring and risk analysis method according to any one of claims 1-9, and the system includes: The perspective calibration module acquires river image frames, image generation time, station identifier and image transmission delay formed at the same flash flood station at continuous sampling time, and determines the perspective offset benchmark based on the riverbank position, bridge pier bottom position and road edge position obtained by installation calibration. The water body identification module calculates the position difference between the fixed reference position in the river image frame and the view offset benchmark to determine the valid river image frame that meets the view validity condition. The overflow delineation module determines the normal water body response, flood response, exposed riverbank response, and carrying capacity response based on the effective river channel image frames, and delineates candidate overflow water body areas based on the boundary expansion distance between the normal water body response and the flood response. The advancement determination module analyzes the advancement speed of the water body boundary based on the candidate overflow water body region corresponding to continuous sampling time, and compares the advancement speed of the water body boundary with the rising speed benchmark generated by the average boundary displacement of adjacent sampling time to determine the real overflow water body region. The risk rating module determines the flash flood risk level based on the overlap between the actual overflow water area and the calibrated position of the carrier, the advance speed of the water boundary, the response of floating objects and the response of the collapsed body, and generates instructions for adjusting the image upload granularity and sampling interval based on the flash flood risk level.

Images