A remote rainfall visualization monitoring method and system for reducing video transmission traffic

Abstract

The application discloses a remote rainfall visualization monitoring method and system for reducing video transmission flow, and the method comprises the following steps: multi-modal data acquisition, video preprocessing, key video frame extraction, multi-source data space-time synchronization encapsulation, low-flow remote transmission and generation of a visualization picture. The application belongs to the technical field of rainfall information monitoring, and specifically relates to a remote rainfall visualization monitoring method and system for reducing video transmission flow. In the scheme, a rain streak feature distinguishing model is constructed, and a video frame is subjected to semantic recognition and adaptive enhancement processing in combination with a rainfall state vector. Time stamp is used as a benchmark to perform multi-source data space-time alignment, and cross-modal state level alignment is performed based on the rainfall state vector. A multi-edge collaborative rainfall prediction model, an utility function and a gate fusion mechanism are combined to perform priority scheduling and hierarchical transmission on transmission data, so that low-flow adaptive transmission is realized.

Description

Technical Field

[0001] This invention belongs to the field of rainfall information monitoring technology, specifically a remote rainfall visualization monitoring method and system that reduces video transmission traffic. Background Technology

[0002] The remote rainfall visualization monitoring method is a method that uses ground-based cameras, radar, satellites, and IoT sensors to collect raw images, videos, and rainfall data on-site. This data is then transmitted in its entirety to a server for multi-source data fusion and processing to generate strong heat maps, echo maps, and real-scene images, thus remotely visually presenting the dynamic process of rainfall.

[0003] However, existing remote rainfall visualization monitoring methods suffer from several technical problems. First, raindrops adhering to the camera lens cause blurring, or dense rain lines obscure the background, making it impossible for the server to distinguish rain streaks from fog, dust, or other interference, leading to visualization failure and an inability to identify rainfall intensity and real-time conditions. Second, traditional network transmission latency and jitter, coupled with the independent acquisition of data by both sensors and cameras, result in data and image misalignment, making it difficult to verify quantitative rainfall data using video footage. Third, the real-time performance of rainfall monitoring and data validity are difficult to balance. Rainfall monitoring requires real-time or high-frequency video transmission, generating massive amounts of video data that consume significant network bandwidth, leading to transmission interruptions in remote areas with weak signals, high storage pressure, and significantly increased hardware costs. Even if rainfall remains unchanged for extended periods, the system continues to transmit redundant data, resulting in wasted bandwidth and power. Summary of the Invention

[0004] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides a remote rainfall visualization monitoring method and system that reduces video transmission traffic. It addresses the technical problems of raindrops blurring the camera lens or dense rain lines obscuring the background during rainfall, making it impossible for the server to distinguish rain patterns from fog, dust, and other interferences, thus rendering visualization observations ineffective and failing to identify rainfall intensity and real-time conditions. The invention employs a rain pattern feature differentiation model, combines rainfall state vectors for semantic recognition and adaptive enhancement processing of video frames, and uses continuous temporal frame difference and dynamic feature analysis to achieve secondary discrimination between rainfall and various interferences. It also combines sensor data and visual recognition results for cross-modal consistency verification, outputting rainfall credibility, thereby effectively distinguishing real rainfall from environmental interference and improving the visualization observation capability. Furthermore, it addresses the technical problems of latency jitter in traditional network transmission and the inability to synchronize data and images due to independent acquisition by dual systems of sensors and cameras, making it difficult to verify quantitative rainfall data using video footage. This paper addresses the challenge of balancing real-time rainfall monitoring with data validity. It employs a multi-source data spatiotemporal alignment based on timestamps and cross-modal state-level alignment based on rainfall state vectors. This enables the fusion of video and sensor data in a unified state space, allowing video footage to effectively corroborate corresponding rainfall data. Furthermore, it addresses the technical problem of the difficulty in achieving both real-time rainfall monitoring and data validity. Rainfall monitoring requires real-time or high-frequency video transmission, resulting in massive video data consumption, high network bandwidth usage, transmission interruptions in remote areas with weak signals, and significant storage pressure, leading to substantial hardware costs. Even when rainfall remains unchanged for extended periods, the system continues to transmit redundant data, resulting in wasted bandwidth and power. The paper employs a multi-party collaborative rainfall prediction model, utility functions, and a gating fusion mechanism to prioritize and hierarchically transmit data, achieving low-traffic adaptive transmission. Simultaneously, a feedback mechanism dynamically optimizes sampling and transmission strategies, significantly reducing bandwidth consumption and storage pressure while ensuring the integrity of critical rainfall information.

[0005] The technical solution adopted by this invention is as follows: This invention provides a remote rainfall visualization monitoring method that reduces video transmission traffic, the method comprising the following steps: Step S1: Multimodal data acquisition; Step S2: Video preprocessing; Step S3: Extraction of key video frames; Step S4: Multi-source data spatiotemporal synchronization encapsulation; Step S5: Low-bandwidth remote transmission; Step S6: Generate a visualization.

[0006] Further, in step S1, the multimodal data acquisition specifically involves real-time acquisition of rainfall and rainfall intensity through a rain sensor, acquisition of environmental parameters through a humidity sensor, a barometric pressure sensor, a wind speed sensor, an environmental sensor, and a visibility sensor, generating a rainfall state vector; acquisition of raw video frames by a camera; and preset of a rainfall trigger threshold.

[0007] Further, in step S2, the video preprocessing includes the following steps: Step S21: Construct a rain pattern feature library and establish a rain pattern feature differentiation model in advance to distinguish between real rainfall and various environmental disturbances, taking into account the differentiated features of rain monitoring scenarios. The smart gateway reads the original video frames at the edge and, combined with the rainfall state vector, extracts shallow texture features, dynamic pixel features, edge features, and temporal motion features frame by frame using a lightweight AI model, outputting the semantic recognition result of a single frame. The smart gateway is deployed as an edge node near the data acquisition end to perform video preprocessing, semantic analysis, key frame extraction, and transmission strategy scheduling, thereby realizing adaptive video transmission based on edge computing. Step S22: Based on the semantic recognition results of a single frame, the rain line density level, lens contamination level, image visibility level, fogging degree, and dust interference level are normalized and quantized, and an adaptive enhancement coefficient is calculated. The processing intensity of the rain removal, fog removal, noise reduction, water stain removal, and detail enhancement modules is dynamically adjusted according to the adaptive enhancement coefficient to obtain the enhanced video frame. Subsequently, the enhancement quality of the enhanced video frame is quantitatively evaluated, and the restored rain scene is quantitatively evaluated through image enhancement quality scoring. Finally, the optimized enhanced video frame is output. Step S23: Continuous temporal frame buffer comparison. The smart gateway buffers multiple consecutive video frames and performs inter-frame difference and temporal feature analysis on the consecutive video frames before and after enhancement. It uses the temporal characteristics of continuous displacement, dynamic flickering and random changes of rain patterns to make a differential comparison with the steady-state characteristics of slow gradual change of fog, static distribution of dust and fixed position blur of lens water stains. Secondary discrimination of dynamic and static interference involves fusing the semantic recognition results of a single frame with temporal change features to generate secondary discrimination results. If the texture between frames has continuous dynamic change characteristics, it is determined to be a real rainfall area. If the inter-frame changes are weak and the spatial position is fixed, it is determined to be a static interference area. Semantic recognition labels are obtained. Step S24: Perform cross-modal consistency verification based on rainfall sensor results, environmental sensor results, and visual recognition results to obtain rainfall confidence and anomaly markers.

[0008] Further, in step S3, the extraction of the key video frames includes the following steps: Step S31: Receive the optimized enhanced video frames, semantic recognition tags, image enhancement quality score and rainfall credibility, and set up a rainfall semantic event triggering mechanism to trigger the keyframe extraction operation; Step S32: Construct an adaptive frame rate control mechanism driven by rainfall information density. The sampling frame rate is dynamically and adaptively adjusted based on the rainfall information density, which is defined as the rate of change of the rainfall state vector. Based on the rainfall information density, a nonlinear mapping function from the degree of information change to the sampling frequency is constructed. Through this mapping relationship, the dynamic sampling frame rate is obtained, realizing an adaptive sampling strategy based on information changes. Step S33: Generate a semantic event frame sequence. Based on the dynamic sampling frame rate, control the frequency and number of key frame extractions, and deeply bind the extracted key frames, timestamps, rainfall state vectors, semantic recognition labels, enhanced quality scores, rainfall credibility, and anomaly markers to generate a semantic event frame sequence.

[0009] Further, in step S4, the spatiotemporal synchronization encapsulation of multi-source data includes the following steps: Step S41: The smart gateway receives the semantic event frame sequence and, based on the high-precision timestamp in the semantic event frame as a unified time axis reference, performs cross-modal state-level alignment of the semantic event frame, rain sensor data, environmental sensor data, device status data, and network status data. Step S42: Perform Kalman filtering on the key state variables in the rainfall state vector to generate an optimized rainfall state vector; Step S43: Unify the encapsulation of rainfall status packets. The optimized rainfall status vector, semantic event frame, enhanced quality score, rainfall credibility, anomaly marker, device status data, and network status data are uniformly encapsulated to generate a standardized rainfall status packet.

[0010] Further, in step S5, the low-bandwidth remote transmission includes the following steps: Step S51: Deploy an attention-based multi-sided collaborative rainfall prediction model on the smart gateway, receive standardized rainfall status packets, and construct a time-series feature vector; A global temporal collaborative analysis module based on an attention mechanism is introduced to dynamically capture the spatiotemporal correlation patterns of rainfall between cross-regional edge nodes. The attention weights between nodes are calculated by combining temporal feature vectors to quantify the relative influence of each node in global temporal rainfall prediction. The multilateral collaborative rainfall prediction model outputs prediction results based on the attention-weighted features, including expected rainfall intensity, expected rainfall duration, trend of change, and prediction confidence. Step S52: Perform cross-node collaborative storage and pre-scheduling based on the prediction results. When an upstream node detects heavy rainfall and its movement direction is downstream, it sends a recording preparation instruction to the smart gateway of the downstream node in advance, so that the downstream node allocates storage space in advance and enters the recording state. The capacity allocation of the local circular buffer is dynamically adjusted according to the expected rainfall intensity level and expected rainfall duration. Step S53: Construct a utility function. Based on transmission delay and image quality perception, construct a utility function to drive dynamic decision-making on video upload timing and compression rate. Normalize the rainfall information density and construct a dynamic weight function based on the normalized rainfall information density. Use this function to adjust the weights of each item in the utility function to obtain a comprehensive utility value. Step S54: Adaptive priority adjustment, introducing a gating fusion mechanism, inputting the comprehensive utility value and the normalized rainfall information density into the gating fusion model, calculating the gating fusion weight, and dynamically adjusting the transmission priority based on the gating fusion weight to obtain the final transmission priority; Step S55: The standardized rainfall status packets, keyframes, short video clips, and status vectors are sorted and transmitted in a hierarchical manner according to the final transmission priority; The system supports breakpoint resume and tiered confirmation mechanisms. After a transmission interruption, the upload resumes from the last interruption point without retransmitting the successfully transmitted portion. The smart gateway can only clear its local cache after the server confirms receipt. When the network recovers, the system automatically resumes the transmission of backlogged data packets according to the final transmission priority, ensuring reliable data transmission.

[0011] Further, in step S6, generating the visualization includes the following steps: Step S61: The server receives the rainfall status packet sequence, and reconstructs the complete rainfall process based on the timestamp, optimized rainfall status vector, semantic recognition label, optimized enhanced video frame and prediction result, generating dynamic rainfall intensity curve, visibility change curve, rainfall status change map and abnormal event timeline, and obtaining a visualized rainfall process interface; Step S62: Application-side interactive presentation. The application and the server establish network communication. The application sends a request to the server to retrieve video data for a specific rainfall period. After receiving the request, the server returns the corresponding video data to the application according to the requested time period. The application provides timeline backtracking, semantic search, and data-video linkage functions, allowing users to view corresponding video evidence and rainfall curves based on rainfall events, thus realizing interactive linkage between video and data. Step S63: The server generates feedback parameters based on video reconstruction quality, transmission packet loss rate, latency, and user viewing frequency, and returns the feedback parameters to the smart gateway for subsequent dynamic adjustment of keyframe sampling frequency, compression rate, buffer capacity, enhancement parameters, and transmission priority to achieve closed-loop optimization.

[0012] The present invention provides a remote rainfall visualization monitoring system with reduced video transmission traffic, comprising a multimodal data acquisition module, a video preprocessing module, a key video frame extraction module, a multi-source data spatiotemporal synchronization encapsulation module, a low-traffic remote transmission module, and a visualization generation module; The multimodal data acquisition module specifically acquires rainfall and environmental parameters through sensors and constructs a rainfall state vector, while simultaneously acquiring raw video frames and preset rainfall trigger thresholds; The video preprocessing module specifically performs semantic recognition and adaptive enhancement processing on the original video frames, constructs a rain pattern feature discrimination model and extracts semantic features by combining rainfall state vectors, performs secondary discrimination of dynamic and static interference through continuous temporal frame difference and temporal feature analysis, and outputs optimized enhanced video frames, semantic recognition labels, rainfall credibility and anomaly markers by combining cross-modal consistency verification. The key video frame extraction module specifically involves setting up a rainfall semantic event triggering mechanism, constructing an adaptive frame rate adjustment mechanism driven by rainfall information density to obtain a dynamic sampling frame rate, and generating a semantic event frame sequence based on the dynamic sampling frame rate. The multi-source data spatiotemporal synchronization encapsulation module specifically involves the smart gateway receiving a sequence of semantic event frames and performing cross-modal state-level alignment, performing Kalman filtering on the rainfall state vector to generate an optimized rainfall state vector, performing unified encapsulation, and generating a standardized rainfall state package. The low-traffic remote transmission module specifically predicts rainfall trends based on a multilateral collaborative rainfall prediction model and an attention mechanism. It performs cross-node collaborative storage and pre-scheduling based on the prediction results and adaptively schedules transmission priorities by combining a utility function and a gating fusion mechanism to achieve hierarchical sorting and low-traffic transmission of data. The module for generating visualizations specifically reconstructs and visualizes the rainfall process on the server side, and achieves closed-loop adaptive optimization of system parameters through a feedback mechanism.

[0013] The beneficial results achieved by the present invention using the above solution are as follows: (1) To address the technical problem that raindrops adhere to the camera lens during rain, causing the image to become blurry, or dense rain lines obscuring the background, making it impossible for the server to distinguish rain patterns from other interferences such as fog and dust, resulting in the failure of visual observation and the inability to identify the rain intensity and the actual situation on site, a rain pattern feature differentiation model is constructed. The video frames are semantically recognized and adaptively enhanced by combining the rainfall state vector. The secondary identification of rainfall and various interferences is achieved through continuous temporal frame difference and dynamic feature analysis. Cross-modal consistency verification is performed by combining sensor data and visual recognition results to output the credibility of rainfall, thereby effectively distinguishing real rainfall from environmental interference and improving the visual observation capability of the image. (2) To address the technical problems of time delay jitter in traditional network transmission and the inability to synchronize and align data and images due to independent acquisition by dual systems of sensors and cameras, making it difficult to verify quantitative rainfall data with video footage, we adopt a method of spatiotemporal alignment of multi-source data based on timestamps and cross-modal state-level alignment based on rainfall state vectors to achieve fusion expression of video data and sensor data in a unified state space, thereby enabling video footage to effectively verify the corresponding rainfall data. (3) To address the technical problem that it is difficult to balance the real-time performance and data validity of rainfall monitoring, since rainfall monitoring requires real-time or high-frequency images, the 24 / 7 transmission will generate massive amounts of video data, resulting in a large amount of network bandwidth being occupied. In remote areas with weak signals, transmission is prone to interruption, and storage pressure is high, significantly increasing hardware costs. Even if there is no change in rainfall in the image for a long time, the system is still transmitting redundant data, resulting in ineffective waste of traffic and power. To address this technical problem, a combination of a multilateral collaborative rainfall prediction model, utility function and gating fusion mechanism is adopted to prioritize and hierarchically transmit the transmitted data, achieving low-traffic adaptive transmission. At the same time, the sampling and transmission strategy is dynamically optimized through a feedback mechanism, thereby significantly reducing bandwidth occupation and storage pressure while ensuring the integrity of key rainfall information. Attached Figure Description

[0014] Figure 1 A flowchart illustrating a remote rainfall visualization monitoring method for reducing video transmission traffic provided by the present invention; Figure 2 This is a schematic diagram of a remote rainfall visualization monitoring system that reduces video transmission traffic, as provided by the present invention.

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0017] Example 1, see Figure 1 The present invention provides a remote rainfall visualization monitoring method that reduces video transmission traffic, the method comprising the following steps: Step S1: Multimodal data acquisition, specifically, acquiring rainfall and environmental parameters through sensors and constructing a rainfall state vector, while simultaneously acquiring raw video frames and setting a rainfall trigger threshold; Step S2: Video preprocessing, specifically, semantic recognition and adaptive enhancement processing are performed on the original video frames, a rain pattern feature discrimination model is constructed and semantic features are extracted by combining the rainfall state vector, dynamic and static interference is identified by continuous temporal frame difference and temporal feature analysis, and cross-modal consistency verification is combined to output the optimized enhanced video frames, semantic recognition labels, rainfall credibility and anomaly markers. Step S3: Key video frame extraction, specifically, setting up a rainfall semantic event triggering mechanism, constructing an adaptive frame rate adjustment mechanism driven by rainfall information density to obtain a dynamic sampling frame rate, and generating a semantic event frame sequence based on the dynamic sampling frame rate; Step S4: Multi-source data spatiotemporal synchronization encapsulation, specifically, the smart gateway receives the semantic event frame sequence and performs cross-modal state-level alignment, performs Kalman filtering on the rainfall state vector to generate an optimized rainfall state vector, performs unified encapsulation, and generates a standardized rainfall state package; Step S5: Low-volume remote transmission, specifically, rainfall trend prediction is performed based on a multilateral collaborative rainfall prediction model and attention mechanism, cross-node collaborative storage and pre-scheduling are performed based on the prediction results, and the transmission priority is adaptively scheduled by combining utility function and gating fusion mechanism to achieve hierarchical sorting and low-volume transmission of data. Step S6: Generate a visualization, specifically by reconstructing and visualizing the rainfall process on the server side, and achieving closed-loop adaptive optimization of system parameters through a feedback mechanism.

[0018] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the multimodal data acquisition specifically involves real-time acquisition of rainfall and rainfall intensity through a rain sensor, and acquisition of environmental parameters through a humidity sensor, a barometric pressure sensor, a wind speed sensor, an environmental sensor, and a visibility sensor to generate a rainfall state vector. The rainfall state vector includes at least rainfall intensity, raindrop density, visibility, wind speed, wind direction, humidity, barometric pressure, and ambient light intensity. At the same time, the camera captures raw video frames and synchronously writes the capture time, device number, geographical location, sensor status, battery status, and network status into the captured data. A preset rainfall trigger threshold is set. When the rain sensor detects that the rainfall or rain intensity exceeds the preset rainfall trigger threshold, or when suspected rain streak motion features are detected in the video footage, the smart gateway generates a rainfall trigger signal.

[0019] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the video preprocessing includes the following steps: Step S21: Construct a rain pattern feature library and establish a rain pattern feature differentiation model in advance for different features of rain monitoring scenarios, based on the real rainfall and various environmental interferences such as fog, dust, and lens water stains. Among them, real rain patterns have significant dynamic motion characteristics, including the oblique motion trajectory of raindrops, discrete rain line shape, high-frequency pixel flickering change pattern, and random density texture distribution characteristics; while fog is characterized by uniform gray-scale gradient across the entire area and no dynamic motion texture; dust is characterized by local static particle spots; lens water stains are characterized by static blurring of fixed areas; environmental interference characteristics all have significant dimensional differences from rain pattern characteristics. The smart gateway reads the original video frames at the edge and, combined with the rainfall state vector, extracts shallow texture features, dynamic pixel features, edge features, and temporal motion features frame by frame using a lightweight AI model, and outputs the semantic recognition results for a single frame. The single-frame semantic recognition results include at least the rain line density level, lens contamination level, image visibility level, fogging degree, dust interference level, and visual rainfall credibility. Step S22: Based on the semantic recognition results of a single frame, normalize and quantize the rain line density level, lens contamination level, image visibility level, fogging degree, and dust interference level, and calculate the adaptive enhancement coefficient. The processing intensity of the rain removal, defogging, noise reduction, water stain removal, and detail enhancement modules is dynamically adjusted based on the adaptive enhancement coefficient. For fixed blurry areas caused by raindrops attached to the lens, the raindrop removal intensity is adaptively enhanced according to the lens contamination level to eliminate lens water stains and other false interference. For background occlusion caused by dense rain streaks, background interpolation is performed by combining the information of preceding and following frames to restore the scene details obscured by the rain streaks. For overall whitening and noise blurring caused by fog and dust, defogging enhancement and noise reduction processing are adaptively enabled to improve image clarity and obtain enhanced video frames. Subsequently, the enhanced video frames were subjected to a quantitative evaluation of enhancement quality, which was achieved through image enhancement quality scoring. The restored rainfall footage is quantitatively evaluated. The image enhancement quality score comprehensively considers image cleanliness, detail retention, and residual noise level. The enhancement parameter range is constrained in reverse to avoid over-enhancement distortion or under-enhancement residual interference. Finally, the optimized enhanced video frame is output. Step S23: Continuous temporal frame buffer comparison. The smart gateway buffers multiple consecutive video frames and performs inter-frame difference and temporal feature analysis on the consecutive video frames before and after enhancement. It uses the temporal characteristics of continuous displacement, dynamic flickering and random changes of rain patterns to make a differential comparison with the steady-state characteristics of slow gradual change of fog, static distribution of dust and fixed position blur of lens water stains. Secondary discrimination of dynamic and static interference involves fusing the semantic recognition results of a single frame with temporal change features to generate secondary discrimination results. If the texture between frames has continuous dynamic change characteristics, it is determined to be a real rainfall area. If the inter-frame changes are weak and the spatial position is fixed, it is determined to be a static interference area. Semantic recognition labels are obtained. Step S24: Perform cross-modal consistency verification based on the results of the rain sensor, the environmental sensor, and the visual recognition, and calculate the rainfall credibility; when the rain sensor detects rainfall and the video rain pattern features match, increase the rainfall credibility; when the rain sensor detects an anomaly but the vision does not detect rain patterns, or the vision detects rain patterns but the rain sensor does not respond, generate an anomaly marker.

[0020] By performing the above operations, a rain pattern feature differentiation model is constructed. Combined with the rainfall state vector, semantic recognition and adaptive enhancement processing are performed on video frames. Secondary identification of rainfall and various interferences is achieved through continuous temporal frame difference and dynamic feature analysis. Cross-modal consistency verification is performed by combining sensor data and visual recognition results to output the credibility of rainfall. This effectively distinguishes real rainfall from environmental interference, improves the visual observation capability of the image, and solves the technical problem that when it rains, raindrops adhere to the camera lens and cause the image to be blurred, or dense rain lines cause the background to be obscured, making it impossible for the server to distinguish rain patterns from other interferences such as fog and dust, resulting in the failure of visual observation and the inability to identify the rainfall intensity and the actual situation on site.

[0021] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, the extraction of key video frames includes the following steps: Step S31: Receive the optimized enhanced video frames, semantic recognition tags, image enhancement quality score, and rainfall credibility, and set up a rainfall semantic event triggering mechanism; Keyframe extraction is triggered when rainfall begins, stops, rainfall intensity changes, visibility changes abruptly, lens contamination changes abruptly, anomaly markers are generated, rainfall credibility changes abruptly, or the image enhancement quality score decreases. Step S32: Construct an adaptive frame rate control mechanism driven by rainfall information density, quantify the intensity of rainfall state changes as rainfall information density, and dynamically and adaptively adjust the sampling frame rate based on the rainfall information density. The rainfall information density is defined as the rate of change of the rainfall state vector. ,in, This represents the rainfall information density at time t. A larger value indicates that the rainfall changes more rapidly, requiring a higher sampling frame rate. This represents the normalized rainfall state vector. Indicates the sampling interval; Based on the rainfall information density, a nonlinear mapping function from the degree of information change to the sampling frequency is constructed. ,in, Indicates the dynamic sampling frame rate. Indicates the minimum sampling frame rate. Indicates the highest sampling frame rate. Represents the information density scale parameter, Represents the hyperbolic tangent function; By mapping the relationship, the changes in rainfall status are transformed from a physical quantity space into a sampling control space, thereby realizing an adaptive sampling strategy based on information changes. When the rainfall information density is low, the rainfall status is determined to be stable, and the sampling frame rate is automatically reduced to a low-frequency sampling mode of 1 frame per minute to reduce the generation of redundant data. When the density of rainfall information increases rapidly, it is determined to be a sudden change in rainfall. The sampling frame rate is increased to 1-2 frames per second to capture the key change process. When an abnormal marker or a sudden change in rainfall confidence is detected, a short-time video continuous sampling mode is triggered to generate a 1-5 second short video clip. Rainfall information density is not only used to drive the keyframe sampling frequency adjustment, but also as a unified control parameter for subsequent data encapsulation density and transmission strategy optimization. It is used to influence the semantic event frame generation density and data transmission load, thereby achieving coordinated optimization of acquisition, processing and transmission resources. It transforms from traditional fixed-period sampling or bandwidth-driven sampling to information-driven sampling based on rainfall state changes, thereby ensuring the integrity of key rainfall change information while significantly reducing data transmission volume. Step S33: Generate a semantic event frame sequence. Based on the dynamic sampling frame rate, control the frequency and number of key frame extractions. Deeply bind the extracted key frames, timestamps, rainfall state vectors, semantic recognition labels, enhanced quality scores, rainfall credibility, and anomaly markers to generate a semantic event frame sequence, so that each frame of data has interpretable, quantifiable, and traceable rainfall state attributes.

[0022] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the spatiotemporal synchronization encapsulation of multi-source data includes the following steps: Step S41: The smart gateway receives the semantic event frame sequence and, based on the high-precision timestamp in the semantic event frame as a unified time axis reference, performs cross-modal state-level alignment of the semantic event frame, rain sensor data, environmental sensor data, device status data, and network status data. For data with different sampling frequencies, interpolation, resampling, or filtering algorithms are used to uniformly map them to the same time axis to ensure that each component in the rainfall state vector corresponds precisely to the video footage at the corresponding moment in the time dimension. Step S42: Perform Kalman filtering or adaptive sliding filtering on the key state variables in the rainfall state vector to smooth sensor noise and instantaneous jumps, eliminate false jumps caused by raindrops randomly hitting the sensor, and generate an optimized rainfall state vector. Step S43: Unify the encapsulation of rainfall status packets. The optimized rainfall status vector, semantic event frame, enhanced quality score, rainfall credibility, anomaly marker, device status data, and network status data are uniformly encapsulated to generate a standardized rainfall status packet.

[0023] By performing the above operations, multi-source data is spatiotemporally aligned based on timestamps, and cross-modal state-level alignment is performed based on rainfall state vectors. This enables the fusion expression of video data and sensor data in a unified state space, allowing video footage to effectively verify the corresponding rainfall data. This solves the technical problems of latency jitter in traditional network transmission and the inability to synchronize data and images due to independent acquisition by dual systems of sensors and cameras, making it difficult to verify quantitative rainfall data with video footage.

[0024] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the low-traffic remote transmission includes the following steps: Step S51: Deploy an attention-based, multi-sided collaborative rainfall prediction model on the smart gateway; The system receives standardized rainfall status packets, reads optimized rainfall status vectors, semantic event frames, enhanced quality scores, rainfall confidence, anomaly markers, equipment status data, and network status data, and constructs a temporal feature vector. The temporal feature vector includes at least the rainfall intensity change rate, raindrop density fluctuation, visibility decline trend, humidity change rate, wind speed change rate, enhanced quality change trend, anomaly event frequency, and network status change trend. A global temporal collaborative analysis module based on an attention mechanism is introduced to dynamically capture the spatiotemporal correlation patterns of rainfall between cross-regional edge nodes, including the rainfall propagation delay patterns between upstream and downstream stations, the similarity patterns of rainfall intensity changes in adjacent areas, and the movement trends of regional weather systems. Attention weights between nodes are calculated to quantify the relative influence of each node in global temporal rainfall prediction, enabling edge nodes to dynamically adjust predictions based on cross-regional information. The formula used is as follows: ; In the formula, This represents the attention weight of node j towards node i at time t. , Let represent the temporal feature vectors of the i-th and j-th edge nodes, respectively; W represents the learnable weight matrix; a represents the learnable attention vector; and || represents the vector concatenation operation. Let represent the set of neighboring nodes of node i, k represent the index of the neighboring node, and T represent the matrix transpose operation. Represents a nonlinear activation function; The prediction output of the multi-party collaborative rainfall prediction model is based on attention-weighted features, which outputs prediction results for the next 5-15 minutes, including expected rainfall intensity, expected rainfall duration, trend, and prediction confidence. Through multi-party collaborative analysis, the model utilizes the spatiotemporal correlation of rainfall across sites to significantly improve the accuracy of single-point rainfall prediction, providing a basis for forward-looking video acquisition and transmission scheduling. Step S52: Perform cross-node collaborative storage and pre-scheduling based on the prediction results. When the upstream node detects heavy rainfall and the direction of movement is downstream, it sends a recording preparation instruction to the smart gateway of the downstream node in advance, so that the downstream node can allocate storage space in advance and enter the recording state. Based on the expected rainfall intensity and duration, the capacity allocation of the local ring buffer zone is dynamically adjusted to ensure sufficient storage space during critical periods. The formula used is as follows: ; In the formula, This indicates the capacity of the dynamically allocated buffer. Indicates the basic buffer capacity. This represents the maximum normalized baseline value for rainfall intensity level, which is fixed at 1. This indicates the system's preset baseline rainfall duration. This represents the adjustment coefficient, and its value range is... , This indicates the normalized expected rainfall intensity level. Indicates the expected duration of rainfall; The design employs a tiered storage strategy: for forecasts of stable light rain, the minimum storage quota is allocated; for forecasts of moderate rain, the standard storage quota is allocated; and for forecasts of sudden heavy rain, the highest priority storage is allocated. Through cross-node collaborative pre-scheduling, the linkage storage between upstream and downstream sites is achieved to avoid the loss of critical data due to insufficient local storage. Step S53: Construct a utility function. Based on transmission delay and image quality perception, construct a utility function to drive dynamic decisions regarding video upload timing and compression rate. Normalize the rainfall information density and construct a dynamic weight function based on the normalized rainfall information density. This dynamic weight function is used to adjust the weights of each component of the utility function. The formula used is as follows: ; In the formula, This represents the overall utility value of the data to be transmitted at time t, used to characterize the overall transmission value of the data under the current system state. This represents the image enhancement quality score, with a value ranging from 0 to 1. This indicates the actual transmission latency required for data to be transmitted from the edge node to the server. This indicates the maximum tolerable transmission delay that the system can tolerate. This represents the normalized delay cost. This indicates the predicted data volume or bitstream size of the data to be transmitted. This indicates the maximum amount of data or bandwidth that the system allows to be transmitted. This represents the normalized bandwidth usage cost. This represents the normalized rainfall information density. , , These represent the image quality weighting coefficient, latency penalty weighting coefficient, and bandwidth constraint weighting coefficient, respectively. All three are greater than 0 and satisfy the following conditions: This is used to ensure the normalization of the contributions of each term in the utility function; Step S54: Adaptive priority adjustment. A gating fusion mechanism is introduced, where the comprehensive utility value and the normalized rainfall information density are input into the gating fusion model. The gating fusion weights are calculated, and the transmission priority is dynamically adjusted based on these weights. The formula used is as follows: ; ; In the formula, Indicates the gating fusion weights, This indicates the current network signal-to-noise ratio. This represents the Sigmoid activation function. and This represents the learnable weight matrix and bias, where T denotes the matrix transpose operation. Indicates the final transmission priority. This represents the baseline output corresponding to the image quality priority strategy. This represents the baseline output corresponding to the low-latency-first strategy; When the gating fusion weight is greater than 0.7, the system favors a quality-first strategy, transmitting high-resolution, low-compression video frames. When the gating fusion weight is less than 0.3, the system favors a low-latency priority strategy, transmitting low-resolution, highly compressed video frames or even only the state vector. When the gating fusion weight is between 0.3 and 0.7, a balancing strategy is adopted; Step S55: The standardized rainfall status packets, keyframes, short video clips, and status vectors are sorted and transmitted in a hierarchical manner according to the final transmission priority; When the network is good, complete semantic event frames and key video segments are transmitted; when the network is weak, rainfall status vectors, semantic tags, and low bitrate key frames are transmitted first; when the network is interrupted, no data is uploaded from the local cache. The system supports breakpoint resume and tiered confirmation mechanisms. After a transmission interruption, the upload resumes from the last interruption point without retransmitting the successfully transmitted portion. The smart gateway can only clear its local cache after the server confirms receipt. When the network recovers, the system automatically resumes the transmission of backlogged data packets according to the final transmission priority, ensuring reliable data transmission.

[0025] Example 7, see Figure 1 This embodiment is based on the above embodiment. In step S6, generating the visualization includes the following steps: Step S61: The server receives the rainfall status packet sequence, and reconstructs the complete rainfall process based on the timestamp, optimized rainfall status vector, semantic recognition label, optimized enhanced video frame and prediction result, generating dynamic rainfall intensity curve, visibility change curve, rainfall status change map and abnormal event timeline, and obtaining a visualized rainfall process interface; When there are few video frames transmitted, the server performs frame interpolation or state-driven video compensation reconstruction between adjacent key frames based on key frame timestamps, optimized rainfall state vectors, and semantic recognition tags to generate transition frames and achieve pseudo-continuous video playback. Step S62: Application-side interactive presentation. The application and the server establish network communication. The application sends a request to the server to retrieve video data for a specific rainfall period. After receiving the request, the server returns the corresponding video data to the application according to the requested time period. The application provides timeline backtracking, semantic search, and data-video linkage functions, allowing users to view corresponding video evidence and rainfall curves based on rainfall events, thus realizing interactive linkage between video and data. The timeline backtracking refers to viewing the status curve and corresponding video evidence according to rainfall events; The semantic search means that entering "intensity level = rainstorm" will locate records for all rainstorm periods; The data-video linkage means that clicking on any point on the rainfall curve will automatically jump to a clear video view of the corresponding moment; Step S63: The server generates feedback parameters based on video reconstruction quality, transmission packet loss rate, latency, and user viewing frequency, and returns the feedback parameters to the smart gateway for subsequent dynamic adjustment of keyframe sampling frequency, compression rate, buffer capacity, enhancement parameters, and transmission priority to achieve closed-loop optimization.

[0026] By performing the above operations, a multi-party collaborative rainfall prediction model, utility function, and gating fusion mechanism are combined to prioritize and hierarchically transmit the transmitted data, achieving low-traffic adaptive transmission. At the same time, a feedback mechanism is used to dynamically optimize the sampling and transmission strategy, thereby significantly reducing bandwidth consumption and storage pressure while ensuring the integrity of key rainfall information. This solves the technical problems of the difficulty in balancing the real-time performance and data validity of rainfall monitoring, the need for real-time or high-frequency video transmission, the generation of massive video data due to 24 / 7 transmission, the easy interruption of transmission in remote areas with weak signals, the high storage pressure, the significant increase in hardware costs, and the ineffective waste of traffic and power even when the rainfall in the video does not change for a long time.

[0027] Example 8, see Figure 2 Based on the above embodiments, this embodiment provides a remote rainfall visualization monitoring system with reduced video transmission traffic, including a multimodal data acquisition module, a video preprocessing module, a key video frame extraction module, a multi-source data spatiotemporal synchronization encapsulation module, a low-traffic remote transmission module, and a visualization image generation module. The multimodal data acquisition module specifically acquires rainfall and environmental parameters through sensors and constructs a rainfall state vector, while simultaneously acquiring raw video frames and preset rainfall trigger thresholds; The video preprocessing module specifically performs semantic recognition and adaptive enhancement processing on the original video frames, constructs a rain pattern feature discrimination model and extracts semantic features by combining rainfall state vectors, performs secondary discrimination of dynamic and static interference through continuous temporal frame difference and temporal feature analysis, and outputs optimized enhanced video frames, semantic recognition labels, rainfall credibility and anomaly markers by combining cross-modal consistency verification. The key video frame extraction module specifically involves setting up a rainfall semantic event triggering mechanism, constructing an adaptive frame rate adjustment mechanism driven by rainfall information density to obtain a dynamic sampling frame rate, and generating a semantic event frame sequence based on the dynamic sampling frame rate. The multi-source data spatiotemporal synchronization encapsulation module specifically involves the smart gateway receiving a sequence of semantic event frames and performing cross-modal state-level alignment, performing Kalman filtering on the rainfall state vector to generate an optimized rainfall state vector, performing unified encapsulation, and generating a standardized rainfall state package. The low-traffic remote transmission module specifically predicts rainfall trends based on a multilateral collaborative rainfall prediction model and an attention mechanism. It performs cross-node collaborative storage and pre-scheduling based on the prediction results and adaptively schedules transmission priorities by combining a utility function and a gating fusion mechanism to achieve hierarchical sorting and low-traffic transmission of data. The module for generating visualizations specifically reconstructs and visualizes the rainfall process on the server side, and achieves closed-loop adaptive optimization of system parameters through a feedback mechanism.

[0028] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0029] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0030] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A remote rainfall visualization monitoring method that reduces video transmission traffic, characterized in that: The method includes the following steps: Step S1: Multimodal data acquisition, specifically, acquiring rainfall and environmental parameters through sensors and constructing a rainfall state vector, while simultaneously acquiring raw video frames and setting a rainfall trigger threshold; Step S2: Video preprocessing, specifically, semantic recognition and adaptive enhancement processing are performed on the original video frames, a rain pattern feature discrimination model is constructed and semantic features are extracted by combining the rainfall state vector, dynamic and static interference is identified by continuous temporal frame difference and temporal feature analysis, and cross-modal consistency verification is combined to output the optimized enhanced video frames, semantic recognition labels, rainfall credibility and anomaly markers. Step S3: Key video frame extraction, specifically, setting up a rainfall semantic event triggering mechanism, constructing an adaptive frame rate adjustment mechanism driven by rainfall information density to obtain a dynamic sampling frame rate, and generating a semantic event frame sequence based on the dynamic sampling frame rate; Step S4: Multi-source data spatiotemporal synchronization encapsulation, specifically, the smart gateway receives the semantic event frame sequence and performs cross-modal state-level alignment, performs Kalman filtering on the rainfall state vector to generate an optimized rainfall state vector, performs unified encapsulation, and generates a standardized rainfall state package; Step S5: Low-volume remote transmission, specifically, rainfall trend prediction is performed based on a multilateral collaborative rainfall prediction model and attention mechanism, cross-node collaborative storage and pre-scheduling are performed based on the prediction results, and the transmission priority is adaptively scheduled by combining utility function and gating fusion mechanism to achieve hierarchical sorting and low-volume transmission of data. Step S6: Generate a visualization, specifically by reconstructing and visualizing the rainfall process on the server side, and achieving closed-loop adaptive optimization of system parameters through a feedback mechanism.

2. The remote rainfall visualization monitoring method for reducing video transmission traffic according to claim 1, characterized in that: In step S5, the low-bandwidth remote transmission includes the following steps: Step S51: Deploy an attention-based multi-sided collaborative rainfall prediction model on the smart gateway, receive standardized rainfall status packets, and construct a time-series feature vector; A global temporal collaborative analysis module based on an attention mechanism is introduced to dynamically capture the spatiotemporal correlation patterns of rainfall between cross-regional edge nodes. The attention weights between nodes are calculated by combining temporal feature vectors to quantify the relative influence of each node in global temporal rainfall prediction. The multilateral collaborative rainfall prediction model outputs prediction results based on the attention-weighted features, including expected rainfall intensity, expected rainfall duration, trend of change, and prediction confidence. Step S52: Perform cross-node collaborative storage and pre-scheduling based on the prediction results. When an upstream node detects heavy rainfall and its movement direction is downstream, it sends a recording preparation instruction to the smart gateway of the downstream node in advance, so that the downstream node allocates storage space in advance and enters the recording state. The capacity allocation of the local circular buffer is dynamically adjusted according to the expected rainfall intensity level and expected rainfall duration. Step S53: Construct a utility function. Based on transmission delay and image quality perception, construct a utility function to drive dynamic decision-making on video upload timing and compression rate. Normalize the rainfall information density and construct a dynamic weight function based on the normalized rainfall information density. Use this function to adjust the weights of each item in the utility function to obtain a comprehensive utility value. Step S54: Adaptive priority adjustment, introducing a gating fusion mechanism, inputting the comprehensive utility value and the normalized rainfall information density into the gating fusion model, calculating the gating fusion weight, and dynamically adjusting the transmission priority based on the gating fusion weight to obtain the final transmission priority; Step S55: The standardized rainfall status packets, keyframes, short video clips, and status vectors are sorted and transmitted in a hierarchical manner according to the final transmission priority; The system supports breakpoint resume and tiered confirmation mechanisms. After a transmission interruption, the upload resumes from the last interruption point without retransmitting the successfully transmitted portion. The smart gateway can only clear its local cache after the server confirms receipt. When the network recovers, the system automatically resumes the transmission of backlogged data packets according to the final transmission priority, ensuring reliable data transmission.

3. The remote rainfall visualization monitoring method for reducing video transmission traffic according to claim 1, characterized in that: In step S2, the video preprocessing includes the following steps: Step S21: Construct a rain pattern feature library and establish a rain pattern feature differentiation model in advance to distinguish between real rainfall and various environmental disturbances, taking into account the differentiated features of rain monitoring scenarios. The smart gateway reads the original video frames at the edge and, combined with the rainfall state vector, extracts shallow texture features, dynamic pixel features, edge features, and temporal motion features frame by frame using a lightweight AI model, and outputs the semantic recognition results for a single frame. Step S22: Based on the semantic recognition results of a single frame, the rain line density level, lens contamination level, image visibility level, fogging degree, and dust interference level are normalized and quantized, and an adaptive enhancement coefficient is calculated. The processing intensity of the rain removal, fog removal, noise reduction, water stain removal, and detail enhancement modules is dynamically adjusted according to the adaptive enhancement coefficient to obtain the enhanced video frame. Subsequently, the enhancement quality of the enhanced video frame is quantitatively evaluated, and the restored rain scene is quantitatively evaluated through image enhancement quality scoring. Finally, the optimized enhanced video frame is output. Step S23: Continuous temporal frame buffer comparison. The smart gateway buffers multiple consecutive video frames and performs inter-frame difference and temporal feature analysis on the consecutive video frames before and after enhancement. It uses the temporal characteristics of continuous displacement, dynamic flickering and random changes of rain patterns to make a differential comparison with the steady-state characteristics of slow gradual change of fog, static distribution of dust and fixed position blur of lens water stains. Secondary discrimination of dynamic and static interference involves fusing the semantic recognition results of a single frame with temporal change features to generate secondary discrimination results. If the texture between frames has continuous dynamic change characteristics, it is determined to be a real rainfall area. If the inter-frame changes are weak and the spatial position is fixed, it is determined to be a static interference area. Semantic recognition labels are obtained. Step S24: Perform cross-modal consistency verification based on rainfall sensor results, environmental sensor results, and visual recognition results to obtain rainfall confidence and anomaly markers.

4. The remote rainfall visualization monitoring method for reducing video transmission traffic according to claim 1, characterized in that: In step S4, the spatiotemporal synchronization encapsulation of multi-source data includes the following steps: Step S41: The smart gateway receives the semantic event frame sequence and, based on the high-precision timestamp in the semantic event frame as a unified time axis reference, performs cross-modal state-level alignment of the semantic event frame, rain sensor data, environmental sensor data, device status data, and network status data. Step S42: Perform Kalman filtering on the key state variables in the rainfall state vector to generate an optimized rainfall state vector; Step S43: Unify the encapsulation of rainfall status packets. The optimized rainfall status vector, semantic event frame, enhanced quality score, rainfall credibility, anomaly marker, device status data, and network status data are uniformly encapsulated to generate a standardized rainfall status packet.

5. The remote rainfall visualization monitoring method for reducing video transmission traffic according to claim 1, characterized in that: In step S1, the multimodal data acquisition specifically involves real-time acquisition of rainfall and rainfall intensity using a rain sensor, acquisition of environmental parameters using a humidity sensor, a barometric pressure sensor, a wind speed sensor, an environmental sensor, and a visibility sensor, generating a rainfall state vector, acquisition of raw video frames by a camera, and preset of a rainfall trigger threshold.

6. The remote rainfall visualization monitoring method for reducing video transmission traffic according to claim 1, characterized in that: In step S3, the extraction of key video frames includes the following steps: Step S31: Receive the optimized enhanced video frames, semantic recognition tags, image enhancement quality score and rainfall credibility, and set up a rainfall semantic event triggering mechanism to trigger the keyframe extraction operation; Step S32: Construct an adaptive frame rate control mechanism driven by rainfall information density. The sampling frame rate is dynamically and adaptively adjusted based on the rainfall information density, which is defined as the rate of change of the rainfall state vector. Based on the rainfall information density, a nonlinear mapping function from the degree of information change to the sampling frequency is constructed. Through this mapping relationship, the dynamic sampling frame rate is obtained, realizing an adaptive sampling strategy based on information changes. Step S33: Generate a semantic event frame sequence. Based on the dynamic sampling frame rate, control the frequency and number of key frame extractions, and deeply bind the extracted key frames, timestamps, rainfall state vectors, semantic recognition labels, enhanced quality scores, rainfall credibility, and anomaly markers to generate a semantic event frame sequence.

7. The remote rainfall visualization monitoring method for reducing video transmission traffic according to claim 1, characterized in that: In step S6, generating the visualization includes the following steps: Step S61: The server receives the rainfall status packet sequence, and reconstructs the complete rainfall process based on the timestamp, optimized rainfall status vector, semantic recognition label, optimized enhanced video frame and prediction result, generating dynamic rainfall intensity curve, visibility change curve, rainfall status change map and abnormal event timeline, and obtaining a visualized rainfall process interface; Step S62: Application-side interactive presentation. The application and the server establish network communication. The application sends a request to the server to retrieve video data during the rainfall period. After receiving the request, the server returns the corresponding video data to the application according to the requested time period. The application provides timeline backtracking, semantic search and data-video linkage functions, allowing users to view the corresponding video evidence and rainfall curves based on the rainfall event, realizing interactive linkage between video and data. Step S63: The server generates feedback parameters based on video reconstruction quality, transmission packet loss rate, latency, and user viewing frequency, and returns the feedback parameters to the smart gateway to achieve closed-loop optimization.

8. A remote rainfall visualization monitoring system for reducing video transmission traffic, used to implement the remote rainfall visualization monitoring method for reducing video transmission traffic as described in any one of claims 1-7, characterized in that: It includes a multimodal data acquisition module, a video preprocessing module, a key video frame extraction module, a multi-source data spatiotemporal synchronization encapsulation module, a low-traffic remote transmission module, and a visualization generation module.

9. A remote rainfall visualization monitoring system for reducing video transmission traffic according to claim 8, characterized in that: The multimodal data acquisition module specifically acquires rainfall and environmental parameters through sensors and constructs a rainfall state vector, while simultaneously acquiring raw video frames and preset rainfall trigger thresholds; The video preprocessing module specifically performs semantic recognition and adaptive enhancement processing on the original video frames, constructs a rain pattern feature discrimination model and extracts semantic features by combining rainfall state vectors, performs secondary discrimination of dynamic and static interference through continuous temporal frame difference and temporal feature analysis, and outputs optimized enhanced video frames, semantic recognition labels, rainfall credibility and anomaly markers by combining cross-modal consistency verification. The key video frame extraction module specifically involves setting up a rainfall semantic event triggering mechanism, constructing an adaptive frame rate adjustment mechanism driven by rainfall information density to obtain a dynamic sampling frame rate, and generating a semantic event frame sequence based on the dynamic sampling frame rate. The multi-source data spatiotemporal synchronization encapsulation module specifically involves the smart gateway receiving a sequence of semantic event frames and performing cross-modal state-level alignment, performing Kalman filtering on the rainfall state vector to generate an optimized rainfall state vector, performing unified encapsulation, and generating a standardized rainfall state package. The low-traffic remote transmission module specifically predicts rainfall trends based on a multilateral collaborative rainfall prediction model and an attention mechanism. It performs cross-node collaborative storage and pre-scheduling based on the prediction results and adaptively schedules transmission priorities by combining a utility function and a gating fusion mechanism to achieve hierarchical sorting and low-traffic transmission of data. The module for generating visualizations specifically reconstructs and visualizes the rainfall process on the server side, and achieves closed-loop adaptive optimization of system parameters through a feedback mechanism.

Images