CN-120071198-B - Unmanned aerial vehicle video stream analysis method based on high-frequency digital information analysis

CN120071198BCN 120071198 BCN120071198 BCN 120071198BCN-120071198-B

Abstract

The invention relates to the technical field of unmanned aerial vehicle video stream analysis and geological disaster monitoring, in particular to an unmanned aerial vehicle video stream analysis method based on high-frequency digital information analysis, which comprises the following steps of generating a dynamic risk heat map through a multi-mode data fusion technology, carrying out time sequence modeling on humidity change, crack expansion and heat radiation intensity by utilizing a time convolution network, and predicting the future change trend of a target area; and the unmanned aerial vehicle acquires newly-added data in real time and is used for dynamically updating a heat map and a volume data model to form a monitoring closed loop. According to the method, the high-risk area labeling is optimized through the multi-mode interaction attention mechanism, the resolution and the precision of the heat map are improved through the generation countermeasure network, the efficiency and the accuracy of unmanned aerial vehicle monitoring are improved through combining real-time path optimization, and the method is suitable for dynamic geological disaster monitoring in a complex environment.

Inventors

CHENG LI
LUO WEILI
LI HAITAO
Wei Fengsha
HAO YU
DAI DONG

Assignees

内蒙古警察职业学院
天津星河天玑数字科技有限公司

Dates

Publication Date: 20260512
Application Date: 20250208

Claims (9)

1. The unmanned aerial vehicle video stream analysis method based on high-frequency digital information analysis is characterized by comprising the following steps of: Based on a dynamic partitioning strategy, acquiring multi-mode data in a target area by using an unmanned aerial vehicle, wherein the multi-mode data comprises multi-spectrum video frame data and point cloud depth data, and performing data denoising and space alignment processing to generate preliminary three-dimensional model data; The method comprises the steps of generating a low-resolution volume data model by utilizing a three-dimensional modeling algorithm based on preliminary three-dimensional model data, generating a high-resolution volume data model by adopting refining treatment to a heavy point region, mapping humidity characteristics and thermal radiation characteristics of multispectral video frame data into the high-resolution volume data model by utilizing a nonlinear interpolation method according to a coordinate relation of multispectral data and point cloud depth data in a three-dimensional space, associating material characteristics with geometric characteristics to generate a multidimensional material three-dimensional model with humidity gradient and thermal radiation intensity information, and extracting characteristics of the humidity characteristics and the thermal radiation characteristics in the multidimensional material three-dimensional model by utilizing a characteristic extraction algorithm so as to extract high-frequency characteristic data and low-frequency contour data of a target region; Constructing a causal network based on a high-resolution volume data model and high-frequency characteristic data to define causal graph nodes, quantifying causal relations among the nodes by adopting a causal reasoning method to generate causal contribution data, modeling dynamic characteristic data by utilizing time sequence modeling, predicting future change trend of a target area, and generating extended prediction data; The method comprises the steps of inputting extension prediction data, generating a dynamic risk heat map through a feature fusion algorithm, generating a risk label of an optimized target area of an antagonism network, optimizing the flight path of the unmanned aerial vehicle through a reinforcement learning path planning algorithm in combination with the dynamic risk heat map, preferentially monitoring a high-risk part of the target area, feeding back newly-increased multi-mode acquisition data, and dynamically updating a volume data model and the risk heat map to realize monitoring closed loop.
2. The method of claim 1, wherein the multispectral video frame data in the multimodal data includes a visible light band for capturing surface texture features of a target area, a near infrared band for monitoring humidity changes of soil and vegetation, and a short wave infrared band for detecting local anomalies in surface thermal radiation intensity.
3. The method according to claim 2, wherein the multi-modal video frame data is denoised by an adaptive filter, the adaptive filter dynamically adjusts filtering parameters according to band characteristics of the multi-modal video frame data to eliminate noise caused by light changes, the point cloud depth data eliminates outliers by statistical filtering, and the statistical filtering removes depth data points deviating from an expected range based on local distribution characteristics of depth information.
4. The method of claim 1, wherein the three-dimensional modeling algorithm comprises generating a low-resolution volume data model by using a nerve radiation field model, wherein the nerve radiation field model performs distributed encoding on geometric and spectral information of preliminary three-dimensional model data in a ray sampling mode to generate a low-resolution three-dimensional representation, and performing refinement processing on cracks and subsidence areas of key areas by using a depth super-resolution network on the basis of the low-resolution volume data model, wherein the depth super-resolution network reconstructs detail information layer by using a convolution layer to generate a high-resolution volume data model.
5. The method of claim 1, wherein the feature extraction algorithm comprises performing a two-dimensional fast fourier transform on the moisture and thermal radiation features of the high resolution volumetric data model, extracting high frequency components for capturing details of crack propagation and local subsidence in the target region, and further processing the extracted high frequency components with a continuous wavelet transform that refines the spatial representation of crack edges and subsidence features by decomposing local variations of the high frequency components.
6. The method of claim 1, wherein the causal network constructs a causal graph based on humidity changes, thermal radiation intensity, crack propagation amplitude and surface subsidence in the high resolution volumetric data model, wherein causal relationships of the causal graph are quantified by a bayesian network reasoning method, and wherein the bayesian network calculates causal influence weights between nodes according to prior probabilities and conditional probabilities of each node to generate causal contribution data.
7. The method of claim 6, wherein the time series modeling models dynamic feature data through a time convolution network, a convolution kernel of the time convolution network is dynamically adjusted according to the length of the time series and the feature change rate, a time dependency relationship among humidity change, crack expansion and heat radiation intensity is captured, and an output result of the time convolution network is used for predicting future change trend of a target area and generating expansion prediction data.
8. The method of claim 1, wherein the dynamic risk heat map fuses humidity features, heat radiation features and geometric features from the high-resolution volume data model through a multi-modal interactive attention mechanism, and wherein the interactive attention mechanism dynamically adjusts contributions of the humidity features and the heat radiation features in the high-risk region labeling by calculating a weight matrix between modal features to generate a dynamic risk heat map with enhanced resolution and highlighted emphasis.
9. The method of claim 8, wherein the reinforcement learning path planning algorithm calculates an optimal flight path based on the priority of the high risk area of the dynamic risk heat map in combination with the current position and flight parameters of the unmanned aerial vehicle, wherein the path planning algorithm optimizes the path covering the high risk area by a reward function, adjusts the flight mission of the unmanned aerial vehicle in real time, and generates mission feedback data for updating the dynamic risk heat map.

Description

Unmanned aerial vehicle video stream analysis method based on high-frequency digital information analysis Technical Field The invention relates to the technical field of unmanned aerial vehicle video stream analysis and geological disaster monitoring, in particular to an unmanned aerial vehicle video stream analysis method based on high-frequency digital information analysis. Background The unmanned aerial vehicle video stream analysis has important significance in the unmanned aerial vehicle monitoring field, and utilizes the video stream acquired at high frequency and the intelligent analysis technology to process the dynamic change scene in real time, and particularly provides high-precision data support in the fields of geological disaster monitoring, emergency response, environmental protection and the like. By high-frequency digital information analysis, abnormal changes can be identified more quickly, and an efficient early warning mechanism is provided. In the prior art (Chinese patent, publication No. CN117854256B, geological disaster monitoring method based on unmanned aerial vehicle video stream analysis), an unmanned aerial vehicle is adopted to collect video streams and monitor geological disasters through an optical flow calculation method. The main technical means comprises the steps of collecting image frames in real time, graying, calculating optical flow and generating a motion amplitude distribution map. However, the technical drawbacks are mainly represented by: In the prior art, optical flow calculation is carried out by processing gray images frame by frame, high-frequency change characteristics in a complex environment are difficult to capture, particularly in a region with complex crack expansion or humidity abnormality, accuracy bottleneck exists in optical flow calculation, the prior art is mainly based on single video flow data, other modal characteristics such as humidity and heat radiation cannot be fused, so that description on complex geological disasters is insufficient, a real-time path adjustment strategy based on risk priority is lacking in the prior art, and monitoring task allocation of an unmanned aerial vehicle is difficult to optimize, so that resource waste or a monitoring blind area is caused. Disclosure of Invention Aiming at a plurality of problems existing in the prior art, the invention provides an unmanned aerial vehicle video stream analysis method based on high-frequency digital information analysis, the invention is based on multi-mode data fusion and high-frequency digital information analysis, A set of geological disaster monitoring method integrating dynamic risk heat map generation, time sequence modeling and reinforcement learning path planning is constructed. By fusing humidity, thermal radiation and geometric characteristics, risk priority is dynamically calculated, unmanned plane path planning is optimized, and accurate monitoring and risk early warning of a target area are achieved. The method and the system remarkably improve the real-time performance, flexibility and precision of unmanned aerial vehicle monitoring, can quickly mark high-risk areas in complex environments, and provide a high-efficiency solution for early warning and response of geological disasters. A high-frequency digital information analysis-based unmanned aerial vehicle video stream analysis method comprises the following steps: Based on a dynamic partitioning strategy, acquiring multi-mode data in a target area by using an unmanned aerial vehicle, wherein the multi-mode data comprises multi-spectrum video frame data and point cloud depth data, and performing data denoising and space alignment processing to generate preliminary three-dimensional model data; Combining multispectral video frame data with point cloud depth data through a multi-mode data fusion technology, mapping material characteristics of a target area into the three-dimensional model, and extracting high-frequency characteristic data and low-frequency contour data of the target area through a characteristic extraction algorithm; Constructing a causal network based on a high-resolution volume data model and high-frequency characteristic data to define causal graph nodes, quantifying causal relations among the nodes by adopting a causal reasoning method to generate causal contribution data, modeling dynamic characteristic data by utilizing time sequence modeling, predicting future change trend of a target area, and generating extended prediction data; The method comprises the steps of inputting extension prediction data, generating a dynamic risk heat map through a feature fusion algorithm, generating a risk label of an optimized target area of an antagonism network, optimizing the flight path of the unmanned aerial vehicle through a reinforcement learning path planning algorithm in combination with the dynamic risk heat map, preferentially monitoring a high-risk part of the target area, feeding back newly-increa