CN-121999442-A - Off-site supervision method and system for air monitoring station based on intelligent vision and multi-source data fusion

CN121999442ACN 121999442 ACN121999442 ACN 121999442ACN-121999442-A

Abstract

The application discloses an off-site supervision method and system for an air monitoring station based on intelligent vision and multi-source data fusion. The method comprises the steps of obtaining time sequence data of a target monitoring station and a reference station and calculating a differential sequence to strip background fluctuation, determining an abnormal occurrence reference time point based on the differential sequence, calculating a physical delay time parameter, carrying out reverse tracing from the abnormal reference time point to generate a tracing time window, analyzing video segments in the tracing time window to extract video feature vectors, and carrying out cross-modal alignment and similarity calculation on the time sequence features and the video feature vectors to judge the causal relationship between data abnormality and video behaviors. The application can solve the problem of evidence obtaining time dislocation through accurate reverse time tracing, reduce calculation power consumption through dynamic visual analysis driven by data gradient, and realize accurate, efficient and low-cost off-site supervision on the interference behavior of the monitoring station.

Inventors

ZHENG YAO
XING YU
LI MING
ZHANG TINGTING
HAN JIAN
JI HONGKUN
DUAN RAN
SHAN YIPENG

Assignees

上海第翼信息科技有限公司
河南省生态环境监测和安全中心
上海图视智能科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260409

Claims (10)

1. An off-site supervision method of an air monitoring station based on intelligent vision and multi-source data fusion is characterized by comprising the following steps: Acquiring pollutant concentration time sequence data of a target monitoring station, and acquiring a pollutant concentration average value sequence of at least one reference station around the target monitoring station; Calculating a differential sequence between the pollutant concentration time sequence data and the pollutant concentration mean value sequence of the target monitoring station so as to strip regional background concentration fluctuation, and extracting time sequence characteristics of the differential sequence; determining an abnormality occurrence reference time point based on the differential sequence ; According to the abnormal occurrence reference time point Calculating to obtain physical delay time parameter ; Based on the physical delay time parameter From the abnormal occurrence reference time point Reverse tracing to generate a tracing time window ; Intercepting video segments in the tracing time window from a video cache, and analyzing the video segments to extract video feature vectors; the differential sequence is processed at the abnormal occurrence reference time point And mapping the nearby time sequence features and the video feature vectors to the same feature space, calculating similarity scores between the two features, and judging that the data abnormality of the target monitoring station has causal relation with the behavior in the video segment when the similarity scores exceed a preset similarity threshold.
2. The air monitoring station offsite supervisory method of claim 1, wherein: Determining an abnormal occurrence reference time point based on the differential sequence, wherein the abnormal occurrence reference time point is determined based on the fact that the absolute value of the first derivative of the differential sequence exceeds a preset fluctuation threshold value; The step of calculating physical delay time parameters according to the abnormal occurrence reference time point comprises the steps of combining the physical distance between the sensor of the target monitoring station and a preset potential intervention point according to the abnormal occurrence reference time point Real-time ambient wind speed Intrinsic retention constant of wind direction and equipment sampling pipeline Calculating to obtain physical delay time parameter ; The step of generating a tracing time window by reversely tracing from the abnormal occurrence reference time point based on the physical delay time parameter comprises the step of generating a tracing time window by reversely tracing from the abnormal occurrence reference time point based on the physical delay time parameter; The analysis of the video segment to extract the video feature vector comprises the steps of dynamically adjusting a sampling method of a slow-fast double-flow visual neural network according to the absolute value of the first derivative of the differential sequence, and analyzing the video segment to extract the video feature vector.
3. The air monitoring station offsite supervisory method of claim 2 wherein said calculating results in a physical delay time parameter Specifically comprises the following steps: the physical delay time parameter Defined as contaminant diffusion time Intrinsic retention constant with the sampling tubing of the device The sum of, i.e ; Wherein the contaminant diffusion time Is determined as the physical distance Effective diffusion rate of pollutants in real-time wind environment The effective diffusion rate of Based on the real-time ambient wind speed And wind direction.
4. The air monitoring station offsite supervisory method of claim 3 wherein said effective diffusion rate The calculation mode of (a) is as follows: Acquiring a direction vector pointing from the preset potential intervention point to the sensor ; Acquiring a wind direction vector corresponding to the real-time wind direction ; Calculating the direction vector And the wind direction vector Included angle between ; The effective diffusion rate By the formula Performing a calculation, wherein when When the molecular free diffusion effect is smaller than a preset diffusion influence lower limit value, taking the diffusion influence lower limit value to represent the molecular free diffusion effect under upwind or crosswind conditions; Further, the contaminant diffusion time By passing through And (5) performing calculation.
5. The off-site supervision method of an air monitoring station according to claim 2, wherein the step of dynamically adjusting the sampling method of the slow-fast double-flow visual neural network specifically comprises the following steps: Setting a first derivative threshold And a second derivative threshold Wherein ; When the absolute value of the first derivative of the differential sequence Less than When the method is used, a low-power consumption sampling mode is adopted, the rapid flow path sampling rate of the slow-speed dual-flow optical neural network is reduced to a preset lower limit, and the method mainly relies on a slow flow path to perform environmental static texture analysis; When (when) When the method is used, a standard sampling mode is adopted, the rapid flow path sampling rate is increased to a reference value, and meanwhile, a slow flow path is activated for context analysis; When (when) And when the high-frequency capturing mode is adopted, the sampling rate of the fast flow path is increased to a preset upper limit to capture instantaneous action details, and the channel weight of the slow flow path is increased to enhance space-time feature fusion.
6. An off-site supervisory system for an air monitoring station based on intelligent vision and multi-source data fusion, comprising a processor, a memory, and a program stored on the memory and executed by the processor, the program when executed implementing the method of any one of claims 1 to 5, the system comprising: The time sequence abnormality recognition module is used for acquiring pollutant concentration time sequence data of the target monitoring station and pollutant concentration average value sequences of the peripheral reference stations, calculating a differential sequence between the pollutant concentration time sequence data and the peripheral reference stations, extracting time sequence characteristics, and determining an abnormality occurrence reference time point based on the differential sequence ; A reverse time window definition module for determining a reference time point according to the abnormality occurrence Calculating a physical delay time parameter Generating a tracing time window; The dynamic visual feature extraction module is used for intercepting the video segments in the tracing time window and extracting video feature vectors; And the cross-modal causal judgment module is used for mapping the time sequence characteristics of the differential sequence and the video characteristic vector to the same characteristic space to perform similarity calculation, and judging that causal relation exists between the data abnormality of the target monitoring station and the behavior in the video segment when the similarity score exceeds a preset similarity threshold.
7. The air monitoring station offsite supervisory system as set forth in claim 6 wherein, The time sequence abnormality identification module is further configured to determine an abnormality occurrence reference time point based on an absolute value of a first derivative of the differential sequence exceeding a preset fluctuation threshold; the reverse time window definition module is further configured to combine the physical distance between the sensor of the target monitoring station and the preset potential intervention point according to the abnormal occurrence reference time point Real-time ambient wind speed Intrinsic retention constant of wind direction and equipment sampling pipeline , Calculating to obtain physical delay time parameter The amount of redundant time is included in generating the traceback time window; The dynamic visual feature extraction module is further configured to dynamically adjust sampling logic of the slow-fast dual-flow visual neural network according to an absolute value of a first derivative of the differential sequence to extract video feature vectors.
8. The air monitoring station offsite supervisory system of claim 7, wherein the inverse time window definition module is configured to: the physical delay time parameter Calculated as contaminant diffusion time Intrinsic retention constant with the sampling tubing of the device And (3) summing; and by taking the physical distance Divided by the real-time ambient wind speed Effective diffusion rate calculated from wind direction To determine the contaminant diffusion time 。
9. The air monitoring station offsite supervisory system of claim 8 wherein the inverse time window definition module is operative to calculate the effective diffusion rate And, when further configured to: Acquiring a direction vector pointing from the preset potential intervention point to the sensor Wind direction vector corresponding to real-time wind direction ; Calculating the direction vector And the wind direction vector Included angle between ; Applying the formula Calculating the effective diffusion rate And is arranged as When the diffusion influence lower limit value is lower than the diffusion influence lower limit value, taking the lower limit value for compensating diffusion under upwind and crosswind conditions; ultimately based on the effective diffusion rate Calculating the contaminant diffusion time 。
10. The air monitoring station offsite supervisory system of claim 7, wherein the dynamic visual feature extraction module is configured to: based on a preset first derivative threshold And a second derivative threshold Absolute value of the first derivative of the differential sequence Dividing into three sections; According to And in different intervals, a low-power sampling mode, a standard sampling mode or a high-frequency capturing mode is respectively executed, and the self-adaptive allocation of video processing resources is realized by adjusting the fast flow sampling rate and the channel weight of the slow flow path of the slow fast flow and double flow optical neural network.

Description

Off-site supervision method and system for air monitoring station based on intelligent vision and multi-source data fusion Technical Field The application relates to the field of artificial intelligence and environmental monitoring data processing, in particular to an off-site supervision method and system of an air monitoring station based on intelligent vision and multi-source data fusion and applied to an air quality automatic monitoring station data anti-fake and interference behavior check scene. Background The air quality automatic monitoring station is an important component of an environment protection data network, and the accuracy and the authenticity of data are the basic stones for environment decision-making and law enforcement work. Traditional monitoring station supervision mode mainly relies on periodic manual field inspection and a remote alarm system based on a single pollutant concentration threshold. However, with increasingly stringent regulatory requirements, the prior art has exposed several technical drawbacks in dealing with offsite regulations, particularly in protecting against jamming and data counterfeits. The problem of splitting of data and video is prominent. In the existing supervision system, the video monitoring system and the air quality data acquisition system are usually two sets of physically isolated systems, and only the appearance correspondence of the time stamp layer exists between the two systems. When the monitoring equipment encounters malicious interference such as 'fog gun vehicle spraying', 'local shielding of a sampling port', or 'artificial emission of specific pollutants around a monitoring point', deep causal relation between interference behavior and data pollution results cannot be established. There is a serious problem of evidence-taking time dislocation. The physical diffusion of the contaminants, the capture of the contaminants by the sensor, and the internal processing of the analyzer all require time, which results in a non-negligible physical lag from the actual occurrence of the disturbance behavior to the appearance of a significant anomaly in the sensor data. When the background system invokes the contemporaneous video recording according to the time point of the data abnormality, the enforcer of the interference behavior often leaves the site, so that the intercepted video clip cannot capture effective illegal behavior evidence, and the subsequent law enforcement work is in trouble. Conventional visual analysis schemes have inherent limitations. Some visual anomaly identification schemes based on background modeling or conventional target detection have the problem of high false alarm rate under extreme weather conditions such as outdoor complex natural illumination change, bird sweepness, tree shadow shaking, rain and snow and the like. Meanwhile, full-period and full-quantity intelligent behavior recognition is carried out on the 24-hour uninterrupted high-definition video stream of the monitoring station, huge computing resources are needed, and the computing resources are difficult to bear for computing equipment with limited computing power at the edge of the monitoring station, so that the computing equipment cannot be effectively deployed and responded in real time at the front end. Therefore, how to solve the problems of multi-mode data splitting, evidence obtaining time dislocation, limited edge calculation force and the like, and realize accurate, efficient and low-cost offsite supervision of an air monitoring station is a technical problem to be solved in the field. Disclosure of Invention The application aims to overcome the defects in the prior art, and provides an off-site supervision method and system for an air monitoring station based on intelligent vision and multi-source data fusion, which are used for solving the problem of multi-mode data splitting and evidence obtaining time dislocation and reducing the consumption of computing resources at the edge side. In order to achieve the above purpose, the application provides an off-site supervision method for an air monitoring station based on intelligent vision and multi-source data fusion, which specifically comprises the following steps: Acquiring pollutant concentration time sequence data of a target monitoring station, and acquiring a pollutant concentration average value sequence of at least one reference station around the target monitoring station; Calculating a differential sequence between the pollutant concentration time sequence data and the pollutant concentration mean value sequence of the target monitoring station so as to strip regional background concentration fluctuation; determining an abnormal occurrence reference time point based on the absolute value of the first derivative of the differential sequence exceeding a preset fluctuation threshold value ; According to the abnormal occurrence reference time pointAnd combining the physical distance between the s