CN-122024132-A - Real-time tracking method and system for water pollution diffusion based on video stream time sequence analysis

CN122024132ACN 122024132 ACN122024132 ACN 122024132ACN-122024132-A

Abstract

The application provides a real-time tracking method and a system for water pollution diffusion based on video stream time sequence analysis, which relate to the technical field of data processing, and the method comprises the steps of performing scene self-adaptive enhancement processing on a real-time video stream; introducing a pollution visual feature detection network to identify relevant features of water pollution, carrying out time sequence analysis oriented to pollution diffusion evolution according to time sequence consistency constraint, carrying out pollution diffusion situation modeling according to a water pollution diffusion time sequence change chain, carrying out water pollution pseudo-target characteristic mining under time attention association according to historical video streams, carrying out pseudo-target inhibition correction according to a water area pseudo-target characteristic space, and establishing a second diffusion tracking model. The application solves the technical problems of low reliability and high false alarm rate of the visual recognition result of the water pollution caused by the multi-source and complex interference noise in the optical monitoring scene in the prior art, and improves the accuracy of the visual recognition result of the pollution by combining the time sequence analysis of the video stream and the suppression correction of the false target.

Inventors

LI HAINI
SHI JIACHENG
LI XIONGWEI
YUAN SHUIXING
YU YONG
SHI WEI
HUANG LIUQUN
QIU PEIPEI
CAO CHENCHEN

Assignees

无锡零碳环境管理有限公司

Dates

Publication Date: 20260512
Application Date: 20260121

Claims (10)

1. The real-time tracking method for the water pollution diffusion based on the video stream time sequence analysis is characterized by comprising the following steps of: Acquiring real-time video stream and video acquisition scene data of a target water area, and performing scene self-adaptive enhancement processing on the real-time video stream according to the video acquisition scene data to generate a water area video sequence; Introducing a pollution visual feature detection network to perform water pollution related feature identification on the water area video sequence to obtain a water pollution feature sequence of each frame; according to the time sequence consistency constraint, carrying out time sequence analysis oriented to pollution diffusion evolution on the water pollution characteristic sequences of each frame, and constructing a water pollution diffusion time sequence change chain; modeling pollution diffusion situation according to the water pollution diffusion time sequence change chain, and establishing a first water pollution diffusion tracking model; Digging a water pollution pseudo-target characteristic under the time attention correlation according to a plurality of historical video streams of the target water area, and establishing a water pseudo-target characteristic space; And performing pseudo-target inhibition correction on the first diffusion tracking model of water pollution according to the characteristic space of the pseudo-target in the water area, and establishing a second diffusion tracking model of water pollution.
2. The method for real-time tracking of water pollution spread based on video stream timing analysis according to claim 1, wherein introducing a pollution visual feature detection network to perform water pollution related feature recognition on the water area video sequence, and obtaining each frame of water pollution feature sequence comprises: Invoking the polluted visual feature detection network, wherein the polluted visual feature detection network comprises a multidimensional visual feature capturing network and a water pollution detection network; inputting the water area video sequence into the multidimensional visual feature capturing network to obtain a water area feature sequence of each frame; And carrying out water pollution related characteristic discrimination and coding on the water area characteristic sequences of each frame through the water pollution detection network to generate the water pollution characteristic sequences of each frame.
3. The method for real-time tracking of water pollution diffusion based on video stream timing analysis according to claim 1, wherein performing time series analysis of pollution diffusion evolution-oriented on the water pollution characteristic sequences of each frame according to timing consistency constraint, constructing a water pollution diffusion timing change chain, comprises: Carrying out continuous frame identification of pollution characteristics according to the water pollution characteristic sequences of each frame, and determining a plurality of continuous trends of the pollution characteristics; performing time sequence consistency evaluation according to each pollution characteristic continuous trend to obtain time sequence consistency of each trend; Classifying the water pollution characteristic sequences of each frame according to the time sequence consistency constraint based on the time sequence consistency of each trend to obtain a continuous pollution characteristic group and a discontinuous pollution characteristic group; reconstructing a space-time related pollution diffusion evolution path according to the continuous pollution characteristic group to generate a space-time related pollution diffusion chain; and carrying out short-time discontinuous diffusion characteristic compensation on the space-time related pollution diffusion chain according to the discontinuous pollution characteristic group to generate the water pollution diffusion time sequence change chain.
4. The method for real-time tracking of water pollution diffusion based on video stream timing analysis according to claim 1, wherein modeling pollution diffusion situations according to the water pollution diffusion timing change chain, and establishing a first water pollution diffusion tracking model, comprises: Constructing a pollution diffusion tracking basic model according to the water pollution diffusion time sequence change chain; loading water boundary condition data and water barrier data of the target water area; Performing pollution diffusion situation influence fitting on the pollution diffusion tracking basic model according to the water area boundary condition data to obtain boundary diffusion situation influence characteristics; performing pollution diffusion situation influence fitting on the pollution diffusion tracking basic model according to the water area obstacle data to obtain obstacle diffusion situation influence characteristics; and carrying out fusion constraint correction on the pollution diffusion tracking basic model according to the boundary diffusion situation influence characteristic and the obstacle diffusion situation influence characteristic to generate the first diffusion tracking model of the water pollution.
5. The method for real-time tracking of water pollution spread based on video stream timing analysis according to claim 1, wherein the step of performing time attention-related water pollution pseudo-target characteristic mining according to a plurality of historical video streams of the target water area to establish a water area pseudo-target characteristic space comprises the steps of: performing time attention distribution according to the plurality of historical video streams, and acquiring time attention of each frame; performing water background feature mining on the plurality of historical video streams according to the time attentiveness of each frame to acquire a first pseudo-target characteristic domain; performing aquatic plant feature mining on the plurality of historical video streams according to the time attentiveness of each frame to obtain a second pseudo-target characteristic field; Performing aquatic animal feature mining on the plurality of historical video streams according to the time attentiveness of each frame to obtain a third pseudo-target feature domain; and integrating the first pseudo-target characteristic domain, the second pseudo-target characteristic domain and the third pseudo-target characteristic domain to generate the water area pseudo-target characteristic space.
6. The method for real-time tracking water pollution spread based on video stream timing analysis according to claim 5, wherein performing water background feature mining on the plurality of historical video streams according to the frame time attentions to obtain a first pseudo-target characteristic field comprises: carrying out water background color feature identification according to the plurality of historical video streams to obtain a first water background feature sequence; Carrying out water background brightness characteristic identification according to the plurality of historical video streams to obtain a second water background characteristic sequence; Carrying out water background texture feature identification according to the plurality of historical video streams to obtain a third water background feature sequence; And carrying out multidimensional association fusion on the first water body background characteristic sequence, the second water body background characteristic sequence and the third water body background characteristic sequence based on the time attention of each frame to generate the first pseudo-target characteristic domain.
7. The method for real-time tracking of water pollution diffusion based on video stream timing analysis according to claim 1, wherein performing pseudo-target inhibition correction on the first diffusion tracking model of water pollution according to the water pseudo-target characteristic space, and establishing a second diffusion tracking model of water pollution, comprises: carrying out multi-point position twin evaluation on the water pollution first diffusion tracking model according to a first pseudo-target characteristic domain to obtain a first pseudo-target twin evaluation set; Carrying out multi-point position twin evaluation on the first diffusion tracking model of water pollution according to a second pseudo-target characteristic domain to obtain a second pseudo-target twin evaluation set; carrying out multi-point position twin evaluation on the first diffusion tracking model of water pollution according to a third pseudo-target characteristic domain to obtain a third pseudo-target twin evaluation set; Checking the first pseudo-target twin evaluation set, the second pseudo-target twin evaluation set and the third pseudo-target twin evaluation set according to pseudo-target twin evaluation constraint conditions to obtain a multi-level pseudo-target evaluation checking sequence; And performing pseudo-target inhibition optimization on the first water pollution diffusion tracking model according to the multi-level pseudo-target evaluation and inspection sequence, and generating a second water pollution diffusion tracking model.
8. The method for real-time tracking of water pollution spread based on video stream timing analysis as recited in claim 1, wherein said video acquisition scene data includes lighting condition data, water surface state data, photographing device data, and weather environment data.
9. The method for real-time tracking of water pollution spread based on video stream timing analysis according to claim 1, wherein a water pollution spread alarm is generated based on the second water pollution spread tracking model.
10. A real-time tracking system for water pollution spread based on video stream timing analysis, characterized by comprising the steps of implementing the real-time tracking method for water pollution spread based on video stream timing analysis according to any one of claims 1 to 9, wherein the real-time tracking system for water pollution spread based on video stream timing analysis comprises: The scene self-adaptive enhancement processing module is used for acquiring real-time video stream and video acquisition scene data of a target water area, and carrying out scene self-adaptive enhancement processing on the real-time video stream according to the video acquisition scene data to generate a water area video sequence; The water pollution related characteristic identification module is used for introducing a pollution visual characteristic detection network to carry out water pollution related characteristic identification on the water area video sequence, and acquiring a water pollution characteristic sequence of each frame; The diffusion evolution time sequence analysis module is used for carrying out time sequence analysis oriented to pollution diffusion evolution on the water pollution characteristic sequences of each frame according to time sequence consistency constraint to construct a water pollution diffusion time sequence change chain; the pollution diffusion situation modeling module is used for modeling the pollution diffusion situation according to the water pollution diffusion time sequence change chain and establishing a first water pollution diffusion tracking model; The water pollution pseudo-target characteristic mining module is used for mining the water pollution pseudo-target characteristics under the time attention association according to a plurality of historical video streams of the target water area, and establishing a water area pseudo-target characteristic space; and the pseudo target inhibition and correction module is used for carrying out pseudo target inhibition and correction on the first diffusion tracking model of water pollution according to the characteristic space of the water area pseudo target and establishing a second diffusion tracking model of water pollution.

Description

Real-time tracking method and system for water pollution diffusion based on video stream time sequence analysis Technical Field The application relates to the technical field of data processing, in particular to a real-time tracking method and system for water pollution diffusion based on video stream time sequence analysis. Background In the existing water pollution monitoring field, means such as optical remote sensing, fixed-point sensor network and the like are widely adopted for pollution diffusion tracking. However, in the water area monitoring process, the optical monitoring faces complex interference factors including water surface fluctuation, illumination change, meteorological conditions, water surface reflection and the like, so that a false target similar to real pollution is easily generated on the visual characteristic level, the visual recognition result is unstable, the false alarm rate is high, and the accuracy and the reliability of the pollution monitoring result are affected. In summary, in the prior art, because interference noise in an optical monitoring scene is multi-source and complex, the interference pollution monitoring of a pseudo target is easy to occur, and the technical problems of low reliability and high false alarm rate of a visual recognition result of water pollution are caused. Disclosure of Invention The application aims to provide a real-time tracking method and a real-time tracking system for water pollution diffusion based on video stream time sequence analysis, which are used for solving the technical problems of low reliability and high false alarm rate of a visual recognition result of water pollution caused by interference noise multi-source and complex interference pollution monitoring of a pseudo target in an optical monitoring scene in the prior art. In order to achieve the purpose, the application provides a real-time tracking method and a real-time tracking system for water pollution diffusion based on video stream time sequence analysis. The application provides a real-time tracking method of water pollution diffusion based on video flow time sequence analysis, which is realized by a real-time tracking system of water pollution diffusion based on video flow time sequence analysis, wherein the real-time tracking method of water pollution diffusion based on video flow time sequence analysis comprises the steps of obtaining real-time video flow and video acquisition scene data of a target water area, carrying out scene self-adaptive enhancement processing on the real-time video flow according to the video acquisition scene data to generate a water area video sequence, introducing a pollution visual characteristic detection network to carry out water pollution related characteristic identification on the water area video sequence to obtain each frame of water pollution characteristic sequence, carrying out time sequence analysis facing pollution diffusion evolution on each frame of water pollution characteristic sequence according to time sequence consistency constraint, constructing a water pollution diffusion time sequence change chain, carrying out pollution diffusion situation modeling according to the water pollution diffusion time sequence change chain, establishing a first diffusion tracking model of water pollution, carrying out time attention related water pollution pseudo-target characteristic mining according to a plurality of historical video flows of the target water area, establishing a pseudo-target characteristic space, carrying out water pollution pseudo-target characteristic space tracking on the water body under the condition of the pseudo-target water pollution-target space, and carrying out second pseudo-target diffusion inhibition on the first diffusion tracking model. Optionally, the video acquisition scene data includes lighting condition data, water surface state data, shooting equipment data and weather environment data. Optionally, the pollution visual feature detection network is called, the pollution visual feature detection network comprises a multidimensional visual feature capturing network and a water pollution detection network, the water area video sequence is input into the multidimensional visual feature capturing network to obtain each frame of water area feature sequence, and the water pollution detection network is used for carrying out water pollution related feature discrimination and coding on each frame of water area feature sequence to generate each frame of water pollution feature sequence. The method comprises the steps of carrying out continuous frame identification of pollution characteristics according to a water pollution characteristic sequence of each frame, determining a plurality of continuous pollution characteristics, carrying out time sequence consistency evaluation according to each continuous pollution characteristic trend, obtaining time sequence consistency of each trend, classifying the water poll