CN-122022495-A - Intelligent health environment monitoring and early warning system based on artificial intelligence and unmanned aerial vehicle images

CN122022495ACN 122022495 ACN122022495 ACN 122022495ACN-122022495-A

Abstract

The invention discloses an intelligent health environment monitoring and early warning system based on artificial intelligence and unmanned aerial vehicle images, and particularly relates to the field of monitoring and early warning, comprising an unmanned aerial vehicle cluster module, an edge computing module, a central analysis platform and an intelligent early warning and decision support module; according to the invention, images are acquired according to a preset path through the unmanned aerial vehicle cluster, parameters are dynamically adjusted according to environmental data, and three specialized unmanned aerial vehicles work cooperatively; the intelligent early warning and decision support module receives the analysis result, evaluates the risks and carries out grading early warning, generates a treatment scheme, records early warning events, is also provided with a relative intervention mechanism and ensures the safety of the health and maintenance environment.

Inventors

GAO XUEQIANG
WANG JUN
WANG MENGXIA
LIU XING
RU CHEN
LAI WEIGUO
WANG GUODONG

Assignees

山东协和学院

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (8)

1. Intelligent health and maintenance environment monitoring and early warning system based on artificial intelligence and unmanned aerial vehicle image, its characterized in that includes: the unmanned aerial vehicle cluster module is used for carrying out multi-dimensional image acquisition on the healthy and nursed area according to a preset path and comprises visible light, infrared and multispectral images; the edge computing module is used for being deployed at the unmanned aerial vehicle end, preprocessing image data in real time and extracting dynamic targets and environmental characteristics; a central analysis platform, comprising: The environment risk recognition unit is used for detecting a static environment risk target by adopting an improved YOLOv model, introducing an attention enhancement mechanism to optimize small target recognition, analyzing continuous frame images by using a 3D-CNN, predicting risk evolution trend, integrating double-flow features by using a self-adaptive weight fusion module, and outputting a dynamic risk probability map; The personnel behavior analysis unit is used for identifying key points of the human body from the overlook angle image based on MobileViT networks of knowledge distillation training, and then merging the proper aging behavior knowledge patterns through the behavior judgment logic to define a typical health care scene behavior pattern; the ecological parameter monitoring unit is used for inverting environmental parameters based on multispectral data, including temperature and humidity distribution, vegetation coverage rate and ultraviolet intensity; And the intelligent early warning and decision support module is used for receiving an analysis result of the central analysis platform and calling a corresponding model in the health environment risk model library to perform risk assessment and grading.
2. The intelligent health environment monitoring and early warning system based on artificial intelligence and unmanned aerial vehicle images, which is characterized in that in the unmanned aerial vehicle cluster module, the flying height and the shooting angle are dynamically adjusted according to real-time environment data, wherein a 5000-ten-thousand-pixel global shutter sensor is adopted for visible light images to collect detailed information at the ground resolution of 0.5cm/px, an infrared thermal imaging unit synchronously detects an abnormal ground surface temperature area, and a multispectral camera acquires vegetation physiological state data at the spectral resolution of 10 nm.
3. The intelligent health environment monitoring and early warning system based on artificial intelligence and unmanned aerial vehicle images, which is characterized in that in the edge calculation module, the real-time processing of image data is realized through a three-stage pipeline architecture, wherein the method comprises the steps of firstly aligning sensor data, establishing a pixel-level mapping relation of visible light, infrared and multispectral images, registering by adopting an affine transformation model, and specifically comprises the following steps: Wherein, the Expressed as source image coordinates, i.e. the position of the target in the source image; expressed as post-registration coordinates, i.e., the position of the object in the new image after affine transformation; 、、、 elements represented as affine transformation matrices; 、 Represented as translation vectors; Then, dynamic target detection is executed, a double-threshold detection mechanism is built based on a modified YOLOv-tiny model (parameter quantity is 6.3M), and the calculation method specifically comprises the following steps: Wherein, the Expressed as a probability of the final detection result, Expressed as a probability of a category, Expressed as the probability of the existence of the object, Indicated as a high threshold value, Represented as a low threshold; And for a confirmed target, calculating a motion track through an optical flow-feature point joint tracking algorithm, adopting an improved NDVI vegetation index aiming at multispectral data in an environmental feature extraction stage, and transmitting the processed feature data to a central analysis platform through an entropy coding compression algorithm.
4. The intelligent health environment monitoring and early warning system based on artificial intelligence and unmanned aerial vehicle images, which is characterized in that in the environment risk recognition unit, the environment risk recognition unit adopts a heterogeneous double-flow architecture to realize multi-scale risk perception, a spatial flow branch is constructed based on an improved YOLOv model, a cross-stage local attention module (CSLA) is added in a backbone network, the detection capability of small targets such as a wet-skid reflective area is obviously improved, and the calculation method specifically comprises the following steps: wherein X is represented as an input feature map, Represented as an activation function, Represented as a 3 x 3 convolution in a depth separable convolution, Represented as an element-by-element multiplication, Represented as a gaussian error linear unit, Represented as a layer normalization, Represented as a1 x 1 convolution operation; the time sequence flow branch adopts a space-time separation 3D-ResNet architecture to process a video flow of 10 frames/second, and the time dimension convolution kernel calculation method specifically comprises the following steps: Wherein, the Represented as the output of the time-dimensional convolution, Represented as a feature map of the current frame, The weight of the convolution kernel of the time dimension, Represented as the radius of time of the convolution kernel, Represented as a bias term, Represented as a feature map of adjacent frames.
5. The intelligent health environment monitoring and early warning system based on artificial intelligence and unmanned aerial vehicle images, according to claim 4, is characterized in that the double-flow features realize dynamic weighting through a gating fusion mechanism, and the calculation method specifically comprises the following steps: Wherein, the Represented as a feature after the fusion, Expressed as gating weight, dynamically adjusting the contribution ratio of the spatial stream and the time stream characteristics; Expressed as spatial stream features, capturing static information; represented as a time-stream feature; and finally, carrying out space smoothing on the output risk probability map by adopting a thermodynamic diffusion model.
6. The intelligent health environment monitoring and early warning system based on artificial intelligence and unmanned aerial vehicle images, which is disclosed in claim 1, is characterized in that the personnel behavior analysis unit adopts a multi-mode fusion architecture, a three-dimensional behavior analysis system is constructed through unmanned aerial vehicle collaborative observation of overlooking view angles and side view angles, a space-time diagram convolution network introduces an improved space-time attention mechanism, a time dimension convolution kernel of the space-time diagram convolution network is expanded to 9 frames, acceleration characteristics of falling actions are captured, and the calculation method specifically comprises the following steps: Wherein, the Represented as a total loss function for keypoint detection, The weight coefficient expressed as a loss of coordinates, The weight coefficient, denoted as angle loss, k is denoted as index of the keypoint, The coordinates of the kth keypoint of the model prediction, The coordinates of the kth key point expressed as true, S expressed as a set of key angles, j expressed as an index of key angles, A direction vector denoted as the jth critical angle of model prediction, Expressed as true first Direction vectors of the key angles; The noninductive monitoring mode for the solitary old people adopts a multistage trigger mechanism, wherein the milliwave radar is used for monitoring micro-motion signals, when effective activities are not detected for 2 continuous hours, the unmanned aerial vehicle is started to approach and observe, and when the effective activities are not detected for 8 continuous hours, the intelligent bracelet is linked to detect vital signs, and meanwhile, three-stage alarms are sent to the nursing station.
7. The intelligent health environment monitoring and early warning system based on the artificial intelligence and the unmanned aerial vehicle image, which is disclosed in claim 1, is characterized in that in the ecological parameter monitoring unit, inversion of environment parameters is carried out through a multispectral imaging system, and the spectral information collected by the unmanned aerial vehicle and the actually measured data of a weather station are calibrated by adopting a multisensor data fusion technology; the vegetation coverage analysis adopts an improved mixed pixel decomposition algorithm, combines the spectral characteristics of a visible light wave band and a red edge wave band, distinguishes artificial lawns from natural vegetation, introduces a shadow compensation mechanism to eliminate the shielding influence of a building, calculates the surface ultraviolet radiation distribution in real time by establishing an empirical model of multispectral reflectivity and UV index in combination with a solar altitude angle in ultraviolet intensity monitoring, and automatically marks a region exceeding a WHO safety threshold.
8. The intelligent health environment monitoring and early warning system based on the artificial intelligence and unmanned aerial vehicle images is characterized in that in the intelligent early warning and decision support module, a hierarchical early warning engine adopts a multi-mode self-adaptive decision framework to construct a complete risk assessment-threshold optimization-response triggering closed loop system, a gating multi-mode fusion network of an engine core adopts a three-level feature processing mechanism, firstly, heterogeneous data are unified to a common space-time reference through a feature alignment module, a dynamic time regularization algorithm is used for solving the time asynchronism problem among sensors, then a feature selector based on an attention mechanism is used for weighting importance of all the modal data, a sliding window self-adaptive algorithm is used for the early warning threshold, a dynamic baseline model is built by taking seasons as units, and the calculation method is as follows: Wherein, the Expressed as the current time Is used for detecting the early warning threshold value of the (a), Expressed as a mean of the historical data, m expressed as a coefficient of sensitivity, Represented as the standard deviation of the historical data, Expressed as a seasonal adjustment factor; and when the composite risk is detected, calculating a comprehensive risk value through fuzzy logic reasoning, outputting an optimal disposal scheme by combining the facility state database through a response strategy generating unit, and pushing through a hierarchical communication protocol.

Description

Intelligent health environment monitoring and early warning system based on artificial intelligence and unmanned aerial vehicle images Technical Field The invention relates to the technical field of monitoring and early warning, in particular to an intelligent health environment monitoring and early warning system based on artificial intelligence and unmanned aerial vehicle images. Background With the profound changes in global population structure, aging has become an irreversible major social trend in the 21 st century, and united nations related reports indicate that the population worldwide, 60 years and older, is growing at a rate of about 3% per year, and it is expected that this population will account for nearly one quarter of the total population by 2050. The aging rapid development brings great health and maintenance requirements, the aged has higher requirements on the quality and safety of living environment due to the reduced physical functions, and the aged needs comfortable and convenient living conditions, and further needs an environment capable of monitoring health conditions in real time and early warning potential hazards in time. The safety management of the health care institutions and community care environments faces unprecedented challenges, the traditional manual inspection method has the inherent defects of large monitoring blind area, response lag and the like, and the fixed monitoring system is limited by the problems of single visual angle, insufficient analysis capability and the like. Disclosure of Invention In order to overcome the above-mentioned drawbacks of the prior art, embodiments of the present invention provide an intelligent health environment monitoring and early warning system based on artificial intelligence and unmanned aerial vehicle images, so as to solve the problems set forth in the above-mentioned background art. In order to achieve the above purpose, the present invention provides the following technical solutions: the unmanned aerial vehicle cluster module is used for carrying out multi-dimensional image acquisition on the healthy and nursed area according to a preset path and comprises visible light, infrared and multispectral images; the edge computing module is used for being deployed at the unmanned aerial vehicle end, preprocessing image data in real time and extracting dynamic targets and environmental characteristics; a central analysis platform, comprising: The environment risk recognition unit is used for detecting a static environment risk target by adopting an improved YOLOv model, introducing an attention enhancement mechanism to optimize small target recognition, analyzing continuous frame images by using a 3D-CNN, predicting risk evolution trend, integrating double-flow features by using a self-adaptive weight fusion module, and outputting a dynamic risk probability map; The personnel behavior analysis unit is used for identifying key points of the human body from the overlook angle image based on MobileViT networks of knowledge distillation training, and then merging the proper aging behavior knowledge patterns through the behavior judgment logic to define a typical health care scene behavior pattern; the ecological parameter monitoring unit is used for inverting environmental parameters based on multispectral data, including temperature and humidity distribution, vegetation coverage rate and ultraviolet intensity; And the intelligent early warning and decision support module is used for receiving an analysis result of the central analysis platform and calling a corresponding model in the health environment risk model library to perform risk assessment and grading. Preferably, in the unmanned aerial vehicle cluster module, the flying height and the shooting angle are dynamically adjusted according to real-time environmental data, the pitch angle is +/-30 degrees, wherein a 5000-thousand-pixel global shutter sensor is adopted for visible light images to acquire detail information with the ground resolution of 0.5cm/px, an infrared thermal imaging unit synchronously detects an abnormal ground surface temperature area, and a multispectral camera) acquires vegetation physiological state data with the spectral resolution of 10 nm; Three types of specialized unmanned aerial vehicles are deployed by adopting a heterogeneous collaborative architecture, wherein a Sony ILME-FR7 camera and an RTK positioning unit are equipped in the first type, a centimeter-level precision daytime panoramic scanning is performed according to a preset grid path, a second type of integrated Livox Mid-360 laser radar and FLIR A8580 thermal imager are automatically switched to a spiral descent mode to perform sub-meter fine detection after receiving an early warning coordinate issued by a center platform, dangerous object three-dimensional characteristics are identified through three-dimensional modeling of a point cloud density of >200pts/cm 3, and the third type is that