Foundation multi-parameter intelligent environment monitoring remote sensing diagnosis method and system

Abstract

The invention relates to a foundation multi-parameter intelligent environment monitoring remote sensing diagnosis method and system. The method comprises the steps that meteorological parameters, air quality parameters and all-sky data of a current target monitoring area are acquired; integrating the meteorological parameters and the air quality parameters to obtain environmental parameters of the target monitoring area; all-sky data is acquired, and the all-sky data is preprocessed to acquire processed all-sky data; and obtaining an environment monitoring remote sensing diagnosis result of the target monitoring area according to the environment parameters and the processed all-sky data. Through efficient processing of multi-source data, comprehensive monitoring of the environment is realized, and accuracy and real-time performance of a remote sensing diagnosis result are improved.

Description

Technical Field

[0001] This application relates to the field of environmental monitoring and remote sensing diagnosis, and particularly to a ground-based multi-parameter intelligent environmental monitoring remote sensing diagnosis method and system. Background Art

[0002] Currently, similar environmental monitoring instruments to this application collect fewer data types and can only cover the monitoring requirements of some meteorological and air quality data. In particular, they lack the extremely important multi-modal data source of cloud map cloud amount data, which limits the comprehensive monitoring of the overall environmental situation and results in low accuracy of environmental monitoring results. Moreover, due to the limitations of orbital period and observation frequency, existing large-platform remote sensing technologies have deficiencies in time resolution and spatial resolution. The data sources of large-platform remote sensing diagnosis technologies are single, and the diagnosis technologies have limitations. Summary of the Invention

[0003] In view of this, this application proposes a ground-based multi-parameter intelligent environmental monitoring remote sensing diagnosis method and system, which can perform environmental monitoring and remote sensing diagnosis based on multi-source data in real time and all-weather.

[0004] According to one aspect of this application, there is provided a ground-based multi-parameter intelligent environmental monitoring remote sensing diagnosis method, including:

[0005] Obtaining meteorological parameters, air quality parameters, and all-sky data corresponding to the current target monitoring area;

[0006] Integrating the meteorological parameters and the air quality parameters to obtain the environmental parameters of the target monitoring area; and preprocessing the all-sky data to obtain the processed all-sky data;

[0007] Obtaining the environmental monitoring and remote sensing diagnosis results of the target monitoring area according to the environmental parameters and the processed all-sky data.

[0008] In a possible implementation method, when obtaining the meteorological parameters and air quality parameters corresponding to the current target monitoring area, the meteorological parameters include at least one of temperature, humidity, wind speed, wind direction, atmospheric pressure, noise, ultraviolet intensity, and rainfall; the air quality parameters include common chemical gases (NH 3 、H 2 S、SO 2 、NO 2 、O 3 、CO), volatile organic compounds (TVOC, HCHO), greenhouse gases (CO 2 、eCO 2 ), and the concentration of respiratory health factors (PM 10 、PM 2.5 、PM 1.0 、negative oxygen ions) of at least one.

[0009] In a possible implementation method, when integrating the meteorological parameters and the air quality parameters, it includes the step of filtering out the outliers in the meteorological parameters and the air quality parameters to obtain the filtered regional environmental parameters.

[0010] In a possible implementation method, when obtaining the all-sky data corresponding to the current target monitoring area, visible light all-sky data is obtained through a visible light all-sky instrument; by controlling the rotation angle and speed of the pan-tilt, multi-angle images are collected at a predetermined angle to obtain multiple infrared all-sky data with different shooting angles.

[0011] In a possible implementation method, when preprocessing the all-sky data, it includes the step of stitching multiple infrared all-sky images collected and performing transitional processing on the stitching boundaries to obtain the processed all-sky data.

[0012] In a possible implementation method, when obtaining the environmental monitoring and remote sensing diagnosis results of the current target monitoring area based on the environmental parameters and the processed all-sky data, it includes:

[0013] Obtaining meteorological prediction data according to the environmental parameters;

[0014] Generating cloud amount and cloud types according to the processed all-sky data;

[0015] Obtaining environmental monitoring results according to the environmental parameters, meteorological prediction data, and the cloud amount and cloud types;

[0016] Obtaining remote sensing diagnosis results according to the processed all-sky data and the environmental monitoring results;

[0017] Among them, the remote sensing diagnosis results are jointly characterized by the correlation of meteorological parameters, cloud displacement vectors, the correlation between cloud displacement vectors and wind vectors, and the prediction results of cloud cluster distribution.

[0018] In a possible implementation method, when stitching multiple infrared sky images, it includes: taking the infrared images as subgraphs and using the visible light all-sky image as a reference benchmark to determine the stitching positions of each infrared image;

[0019] After determining the stitching positions of each infrared image, the infrared images are stitched according to the stitching positions and the stitching boundaries are subjected to transitional processing.

[0020] In a possible implementation method, when generating the cloud amount and cloud types, it includes:

[0021] Capturing the features in the processed all-sky data;

[0022] Classify the cloud based on the captured features to obtain the cloud classification result;

[0023] Identify the cloud from the all-sky data and count the proportion of the identified cloud in the all-sky data to obtain the cloud coverage in the sky as the cloud amount result.

[0024] According to another aspect of the present application, there is also provided a ground-based multi-parameter intelligent environmental monitoring remote sensing diagnosis system, including a data acquisition module, a data processing module, and a result diagnosis module;

[0025] The data acquisition module is configured to acquire meteorological parameters, air quality parameters, and all-sky data corresponding to the current target monitoring area;

[0026] The data processing module is configured to integrate the meteorological parameters and the air quality parameters to obtain the environmental parameters of the target monitoring area; preprocess the all-sky data to obtain the processed all-sky data; and obtain the environmental monitoring and remote sensing diagnosis result of the target monitoring area according to the environmental parameters and the processed all-sky data;

[0027] The result diagnosis module is configured to display the environmental monitoring and remote sensing diagnosis result and make a response according to the diagnosis result.

[0028] According to another aspect of the present application, there is also provided a non-volatile computer-readable storage medium, on which computer program instructions are stored, characterized in that when the computer program instructions are executed by a processor, the method described in any one of the above is implemented.

[0029] According to the following detailed description of the exemplary embodiments with reference to the accompanying drawings, other features and aspects of the present application will become clear. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings included in the specification and constituting a part of the specification show the exemplary embodiments, features, and aspects of the present application together with the specification and are used to explain the principles of the present application.

[0031] Figure 1 A flowchart showing the ground-based multi-parameter intelligent environmental monitoring remote sensing diagnosis method according to an embodiment of the present application;

[0032] Figure 2 A schematic diagram showing the user page design according to an embodiment of the present application;

[0033] Figure 3 A schematic diagram showing a sub-page of the meteorological parameter monitoring module according to an embodiment of the present application;

[0034] Figure 4Schematic diagram of a sub - page of the air quality monitoring module according to an embodiment of the present application;

[0035] Figure 5 Schematic diagram of the vertical distribution of cloud types in Beijing, Daihai, and Lhasa according to an embodiment of the present application;

[0036] Figure 6 Schematic diagram of a sub - page of the all - sky monitoring module according to an embodiment of the present application;

[0037] Figure 7 Schematic diagram of the cloud displacement vector according to an embodiment of the present application;

[0038] Figure 8 Schematic diagram of the analysis result of the correlation between the cloud displacement vector and the wind displacement vector according to an embodiment of the present application;

[0039] Figure 9 Schematic diagram of the analysis result of the correlation between cloud amount and meteorological parameters according to an embodiment of the present application;

[0040] Figure 10 Schematic diagram of the time series of cloud cluster distribution according to an embodiment of the present application;

[0041] Figure 11 Meteorological parameter data set stored in time series according to an embodiment of the present application;

[0042] Figure 12 Schematic diagram of the loss calculation process for cloud map classification and cloud amount calculation according to an embodiment of the present application;

[0043] Figure 13 Schematic diagram of the loss calculation process for cloud map prediction according to an embodiment of the present application;

[0044] Figure 14 Schematic diagram of the loss calculation process for meteorological prediction according to an embodiment of the present application;

[0045] Figure 15 Schematic diagram of the working flow of the ground - based multi - parameter intelligent environmental monitoring remote sensing diagnosis system according to an embodiment of the present application;

[0046] Figure 16 Schematic diagram of the working flow of the cloud - based display system according to an embodiment of the present application;

[0047] Figure 17 Schematic diagram of the main page of the mobile - end user according to an embodiment of the present application. Detailed implementation manners

[0048] The following will detail various exemplary embodiments, features, and aspects of the present application with reference to the accompanying drawings. The same reference numerals in the drawings denote elements with the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings do not have to be drawn to scale unless otherwise specified.

[0049] As used herein, the term "exemplary" means "serving as an example, embodiment, or illustration". Any embodiment described as "exemplary" herein need not be construed as superior to or better than other embodiments.

[0050] In addition, for a better description of the present application, numerous specific details are given in the following detailed implementation manners. Those skilled in the art should understand that the present application can also be implemented without certain specific details. In some instances, methods, means, elements, and circuits well-known to those skilled in the art are not described in detail so as to highlight the gist of the present application.

[0051] The present application is applicable to a ground-based multi-parameter intelligent environmental monitoring remote sensing diagnosis method, which obtains meteorological parameters, air quality parameters, and all-sky data corresponding to the current target monitoring area; integrates the meteorological parameters and the air quality parameters to obtain the environmental parameters of the area; obtains all-sky data by converting the large-platform remote sensing diagnosis technology to a ground-based small platform; integrates and processes the environmental parameters and the all-sky data to obtain an environmental monitoring remote sensing diagnosis result. Specifically, the all-sky data is preprocessed based on an image stitching algorithm to obtain processed all-sky data; features are extracted from the processed all-sky data and environmental parameters based on a deep neural network algorithm to obtain cloud amount and cloud type, meteorological prediction results, and cloud cluster distribution prediction results, and the result data is integrated and calculated to obtain an environmental monitoring and remote sensing diagnosis result and store it in the backend database of the cloud visualization system; a front-end user page is designed based on the Vue.js framework to realize the visualization of the environmental monitoring and remote sensing diagnosis result, providing an intuitive user page and a convenient operation process, thereby reducing the usage and maintenance difficulty; based on this page, the environmental monitoring remote sensing diagnosis result is jointly characterized by environmental parameter results, processed all-sky data, meteorological prediction data, cloud amount and cloud type results, cloud type vertical distribution, cloud displacement vector, cloud displacement vector and wind displacement vector correlation analysis results, and cloud amount and meteorological parameter correlation analysis results, and cloud cluster distribution prediction results, so as to obtain the result of remote environmental monitoring remote sensing diagnosis at any time and place.

[0052] Figure 1 The flowchart of the ground-based multi-parameter intelligent environmental monitoring remote sensing diagnosis method according to an embodiment of the present application is shown. As Figure 1 shown, the method includes:

[0053] Step S100, obtaining meteorological parameters, air quality parameters, and all-sky data corresponding to the current target monitoring area.

[0054] Step S200, integrating the meteorological parameters and the air quality parameters to obtain the environmental parameters of the area; preprocessing the all-sky data to obtain processed all-sky data.

[0055] Step S300: Obtain the environmental monitoring remote sensing diagnosis result of the current target monitoring area based on the environmental parameters and the processed all-sky data.

[0056] Specifically, in the method of the embodiment of the present application, step S100 obtains the meteorological parameters, air quality parameters, and all-sky data corresponding to the target detection area based on the meteorological perception module, the air quality perception module, and the all-sky perception module; among them, the meteorological parameters obtained by the meteorological perception module include at least one environmental basic data such as temperature, humidity, wind speed, wind direction, atmospheric pressure, noise, ultraviolet intensity, and rainfall; the air quality parameters corresponding to the target detection area obtained by the air quality perception module include common chemical gases (NH 3 、H 2 S、SO 2 、NO 2 、O 3 ,CO), volatile organic compounds (TVOC, HCHO), greenhouse gases (CO 2 、eCO 2 ), and at least one air quality parameter that can reflect the air quality situation, such as the concentration of respiratory health factors (PM 10 、PM 2.5 、PM 1.0 、negative oxygen ions); the all-sky data obtained by the all-sky perception module includes at least one of visible light all-sky images and infrared images. The above perception modules use sensors to measure and record data, that is, the monitoring results, to obtain the parameters of the monitoring area.

[0057] The meteorological perception module consists of an optical rain gauge meteorograph (e.g., HQWS-6P), a noise sensor, and an ultraviolet radiation sensor. The optical rain gauge meteorograph is used to collect the temperature, humidity, wind speed, wind direction, atmospheric pressure, and rainfall of the environment; the noise sensor is used to collect environmental noise; the ultraviolet radiation sensor is used to detect the ultraviolet radiation intensity.

[0058] The air quality perception module consists of an electrochemical gas sensor module (e.g., ZE03, ZH03B), a carbon dioxide gas sensor (e.g., MH-411D), an alcohol sensor (e.g., MS-VOC-V4), and a negative (oxygen) ion monitoring sensor. The electrochemical gas sensor module is used to detect common chemical gases (NH 3 、H 2 S、SO 2 、NO 2 、O 3 ,CO) and the concentration of particulate matter such as PM 2.5 in the air; the alcohol sensor is mainly used to monitor formaldehyde and total volatile organic compounds (TVOC, HCHO). The carbon dioxide sensor is used to measure greenhouse gases (CO2 , eCO 2 ) concentration; the negative (oxygen) ion monitoring sensor is used to detect the concentration of negative ions in the air.

[0059] The all-sky perception module consists of a visible light all-sky imager (e.g., DJ-CGQ-QTK-VIS-V1) and an infrared all-sky imager (e.g., FTII640). The visible light all-sky imager is used to collect visible light all-sky images; the infrared all-sky imager is used to collect infrared images.

[0060] It should be noted here that, as an alternative implementation of this application, the above-mentioned models of each sensor are used. It can be adjusted to other models of sensors for detecting the same parameters according to the specific implementation situation, and no specific limitation is made in this application.

[0061] The infrared all-sky imager is installed on the pan-tilt and controlled based on the pan-tilt control algorithm to achieve the coverage of the infrared observation range over the entire sky;

[0062] Specifically, the pan-tilt control algorithm uses C++ as the programming language to control the rotation angle and speed of the pan-tilt; first, according to the specific task requirements, the target angle of the pan-tilt is set, and by obtaining the current angle in the real-time data, the error between the current angle of the pan-tilt and the target angle is calculated. The error calculation formula is:

[0063] e(t) = target angle - current angle. The control signal is calculated using the PID formula and used to drive the motor to adjust the position of the pan-tilt. The control signal calculation formula is as follows, where u(t) represents the output signal of the PID controller:

[0064]

[0065] Among them, the PID parameters are set as follows: proportional gain (Kp): 1.2, which is used to control the amplification factor of the current error; integral gain (Ki): 0.5, which represents the magnitude of the cumulative error; derivative gain (Kd): 0.05, which can predict future errors through the change rate of the error; it should be noted that the PID parameters can be adjusted according to the actual situation.

[0066] Furthermore, based on the closed-loop feedback mechanism, continuously obtain the real-time position information of the current pan-tilt and compare it with the preset position information; according to the dynamic change of the error, adjust the output of the motor and gradually approach the target position.

[0067] It should also be noted here that the control algorithm enables the camera to automatically take pictures at a predetermined angular interval (the angular interval is 360° / the number of sub - images. For example, if the preset number of sub - images is 12, then the angular interval is 30°) during the rotation of the pan - tilt head. Among them, the angular interval can be flexibly adjusted according to the actual situation and is not specifically limited in this application; the preset position information is the position information that the pan - tilt head should be at the current predetermined angle. By comparing with the real - time position, the motor output is adjusted, and then the actual position is approximated to the target position.

[0068] In a possible implementation, the acquisition of real - time infrared image data is realized based on an angle sensor, an incremental encoder, and an accelerometer. Specifically, the angle sensor and the accelerometer can directly obtain real - time angle and acceleration data; the incremental encoder converts displacement into a periodic electrical signal, then converts the electrical signal into pulses, and records the cumulative rotation angle of the motor by counting the number of rotating pulses; tracks the relative position of the motor shaft, thereby calculating the current position of the pan - tilt head based on the initial position and calculating the speed. The speed and position of the motor are collected every 10 milliseconds, and a low - pass filter is used to process the real - time parameters to reduce the influence of high - frequency noise.

[0069] It should be further noted that the pan - tilt head control algorithm is also equipped with a fault detection and recovery system to ensure that when an abnormality occurs in the device, it can quickly return to the normal working state; during normal operation, the fault detection mechanism monitors the rationality of sensor data in real time. If the sensor data appears abnormal (such as sudden jumps or remaining unchanged within a preset time period), there may be a fault; if the pan - tilt head does not move as expected or moves unstably after receiving a control signal, there may be a motor or controller fault; when the error exceeds the preset error threshold range of 0.1 - 0.2, the alarm mechanism is triggered.

[0070] After a fault is detected, in the case of a minor fault, the system can automatically recalibrate the sensor or restart the motor; in the case of a serious fault, the system can automatically switch to the backup system to ensure that the pan - tilt head continues to work normally; if the fault cannot be recovered within the set time, the system will safely stop the operation of the pan - tilt head and record the fault log for subsequent analysis.

[0071] Specifically, the degree of the fault is judged as a minor fault or a serious fault according to the following situations; among them, the minor fault includes at least one of the following situations:

[0072] Abnormal fluctuation of sensor data: The sensor data has a short - term jump or fluctuation, but the data returns to the normal range within 3 - 5 acquisition cycles (such as 3 - 5 minutes). For example, the data of the angle sensor or the accelerometer shows a deviation, but it does not last;

[0073] Control error fluctuation: The error value of the PID controller exceeds the normal range (such as an error between 0.01° and 0.05°) within 2s - 3s, but is within the alarm area (the area that exceeds the normal range but does not exceed the threshold), and gradually returns to normal after adjustment.

[0074] Motor jitter or short-term stall: The motor operation shows instantaneous instability, such as the change gap between the current speed and the set speed is more than 5%, or the position deviates from the preset track by 3% - 7%, but returns to the normal state within 3 - 5 cycles.

[0075] Communication delay or packet loss: There is a short delay or a small amount of data loss in the communication between the system and the pan-tilt head, but it does not cause the overall operation to interrupt, and it recovers within the set time (10s).

[0076] Serious faults include at least one of the following situations:

[0077] Continuous abnormal sensor data: The sensor data continuously exceeds the reasonable range, or does not change for more than the set time (10s). For example, the angle sensor does not update the data or shows errors (such as angle data jumps).

[0078] Control signal failure: The PID controller cannot control the error within the normal range (0.01° - 0.05°), the error continuously remains in the alarm area, and the system cannot normally adjust the position of the pan-tilt head. Among them, the value range of the threshold can be set between 0.1° and 0.2°.

[0079] Motor does not work or is out of control: The motor has no response after receiving the control signal, or there is a stall, weakness, or movement instability for more than 10s during operation. The motor fails to reach the target position within the set time (10s), or the operation trajectory significantly deviates from the predetermined track by 3% - 7%.

[0080] Communication interruption: The communication with the pan-tilt head is interrupted for more than 10s (including 10s), resulting in the system being unable to obtain real-time data or send control instructions.

[0081] System self-check failure: The system cannot pass the self-check during startup or operation (such as the self-check of sensors, encoders, motors, etc. fails), indicating hardware faults or software problems.

[0082] Error exceeds the upper limit: When the error exceeds the preset maximum error threshold (such as 0.2) or higher, it indicates that the system cannot return to the normal state through automatic adjustment, and further manual intervention or the intervention of the backup system is required.

[0083] It should be noted that during the process of determining the degree of the above faults, parameters such as cycles, set time, error thresholds, etc. can be adjusted according to the actual situation, and specific parameter limitations are not carried out in this application.

[0084] In a possible implementation, the meteorological and air quality sensing module transmits measurement values at a first preset frequency (e.g., a frequency of 1 min / time), the visible light all-sky imager transmits all-sky cloud images at a second preset frequency (e.g., a frequency of 5 min / time), and the infrared all-sky imager transmits infrared sub-images at a third preset frequency (e.g., a frequency of 25 s / time). Moreover, multiple sub-images can be stitched together to form an infrared all-sky cloud image. Among them, in a specific embodiment of the present application, the number of sub-images is preset to 12. There is the following quantitative correspondence relationship between the acquisition frequency of the infrared all-sky imager and the acquisition frequency of the visible light all-sky imager:

[0085]

[0086] When the above-mentioned sensing module transmits measurement values, the number of sub-images and the frequency of transmitting measurement data can be adjusted according to the actual situation, and no specific limitation is made in the present application.

[0087] Based on the collected data, step S200 of the ground multi-parameter intelligent environmental monitoring remote sensing diagnosis method integrates environmental parameters and preprocesses all-sky data. Specifically, the meteorological parameters and air quality parameters collected by the data acquisition module in step S100 are integrated to obtain the environmental parameters of the region. Similarly, the all-sky data collected by the data acquisition module in step S100 is preprocessed to obtain the preprocessed all-sky data.

[0088] The data in step S100 is transmitted to step S200 through a customized serial port protocol

[0089] Two coprocessors are set during the data transmission process. Preferably, in a specific embodiment of the present application, the S5P4418 coprocessor is used. The measurement values collected by the meteorological sensing module and the air quality sensing module are respectively sent to the two S5P4418 coprocessors through a customized serial port protocol. The two coprocessors are connected to a switch through a network cable and transmit the measurement values in the form of a serial information stream through the MQTT protocol. The visible light all-sky imager and the infrared all-sky imager are directly connected to the switch through a network cable and use the MQTT protocol to transmit image information. Finally, the edge computing platform (such as NVIDIA TX2) connected to the switch receives and integrates all the data collected in step S100 through the MQTT protocol. Among them, the use of the coprocessor can be replaced with other specifications of processors according to the actual situation, and no specific limitation is made in the present application.

[0090] Further, after the edge computing platform receives the data transmitted by the meteorological sensing module and the air quality sensing module, for the meteorological parameters and air quality parameters, the 3σ principle is adopted to filter abnormal values (such as overflow values, negative values), and then the processed environmental parameters are directly uploaded to the database;

[0091] Specifically, first calculate the mean μ and standard deviation σ of the received meteorological parameters and air quality parameters respectively. When calculating, calculate the mean and standard deviation for at least one of temperature, humidity, wind speed, wind direction, atmospheric pressure, noise, and rainfall in the meteorological parameters respectively.

[0092] For common chemical gases (NH 3 、H 2 S, SO 2 、NO 2 、O 3 、CO), volatile organic compounds (TVOC, HCHO), greenhouse gases (CO2, eCO2), and at least one of the concentrations of respiratory health factors (PM 10 、PM 2.5 、PM 1.0 、negative oxygen ions) in the air quality parameters, calculate the mean and standard deviation respectively.

[0093] It should be noted that during the collection process of meteorological parameters and air quality parameters, the meteorological sensing module and the air quality sensing module transmit measurement values at the first preset frequency (e.g., a frequency of 1 min / time), and calculate the mean and standard deviation using the measurement value data received in the previous 60 minutes.

[0094] Among them, calculate the mean and standard deviation for each data type in the meteorological parameters and air quality parameters respectively.

[0095] For each data type in the meteorological parameters and air quality parameters, calculate the ±3σ standard range of their respective μ, and use this standard range to screen the data;

[0096] Based on the normal distribution, regard the data falling outside the standard range as abnormal data, and exclude the abnormal values and boundary values in the data to achieve the filtering of abnormal data values.

[0097] It should be noted that the method of filtering abnormal values using the 3σ principle adopted in this step can be replaced by other anomaly detection algorithms, and no specific limitation is made in this application.

[0098] Furthermore, visualize the data of the above meteorological parameters and air quality parameters on the page. The main page is as Figure 2 shown. The meteorological parameter monitoring module includes the current monitoring data and meteorological prediction data of temperature, humidity, wind speed, wind direction, atmospheric pressure, noise, and rainfall, etc.

[0099] The air quality monitoring module includes common chemical gases (NH 3 、H 2 S, SO 2 、NO 2 、O 3, CO), volatile organic compounds (TVOC, HCHO), greenhouse gases (CO 2 , eCO 2 ), respiratory health factors (PM 10 , PM 2.5 , PM 1.0 , concentration of negative oxygen ions). There is a sub - page corresponding to each module on the main page. As Figure 3 shown in the schematic diagram of the sub - page of the meteorological parameter monitoring module, including temperature, humidity, wind force, air pressure, rainfall, environmental noise, dividing the corresponding data according to time and retaining historical data. As Figure 4 shown in the schematic diagram of the sub - page of the air quality monitoring module, including the content of common gases, the measurement of chemical substances, and the concentration of atmospheric negative oxygen ions and PM 2.5 , PM 10 concentration. By selecting data of different dates and different scales, formal data (i.e., original collected data) and predicted data generated by neural network can be viewed, and data of different dates can also be selected for dynamic comparison. Moreover, in the sub - page, a table can be downloaded, and the data can be pulled out to generate a table form convenient for monitoring, providing a more intuitive display for users.

[0100] Pre - process the all - sky data. Among them, the visible - light all - sky image is directly obtained by a visible - light all - sky imager; the infrared all - sky image is formed by stitching multiple images collected by an infrared all - sky instrument. In nature, due to the disturbance of natural factors such as wind force, terrain, temperature and humidity changes, the position of the pan - tilt may shift, affecting the stability and accuracy of image acquisition. The pan - tilt control algorithm of this application compares the preset position of the pan - tilt with the actual position, and realizes the assistance and correction of the pan - tilt position by calculating the deviation and making adjustments.

[0101] Among them, the stitching of infrared images is obtained based on the image stitching algorithm. During the image stitching process, infrared image data and visible - light image data are synchronously collected. Multiple infrared images collected at a preset time interval (e.g., 12 infrared images) and the corresponding one visible - light image are used as a group of data for transmission. Among them, according to the acquisition frequencies of visible - light images and infrared images, the quantity correspondence relationship between visible - light images and infrared images is determined, and the number of infrared images corresponding to one visible - light image can be adjusted according to the actual situation, and no specific quantity limit is set here;

[0102] Then, taking the collected group of infrared images as sub - graphs and using the corresponding one visible - light image as a reference benchmark, determine the position where the average similarity result of the features of each infrared image in the group and the features of the corresponding visible - light image is the maximum value, that is, the stitching position of the infrared images, and stitch the sub - graphs according to the position information.

[0103] Specifically, the angle and position information of the pan-tilt are obtained from the data transmitted back by the pan-tilt sensor as the current position and rotation angle information of the sub-image, which is input into the reinforcement learning model SAC as the current state for feature extraction, including spatial features, edge features, texture features, and frequency domain features; a feature vector for comparison is generated; the result of the comparison is characterized by the average similarity between the sub-image and the corresponding visible light image feature vector.

[0104] Among them, the similarity calculation is carried out in the following way. First, at least one of the spatial feature similarity, edge feature similarity, texture feature similarity, and frequency domain feature similarity between the infrared image and the visible light image is calculated. Then, based on the above several calculated similarities, a weighted calculation is performed to obtain the final average similarity result.

[0105] Specifically, the spatial feature similarity calculation is as follows: use convolution to extract the spatial features of the visible light image and the infrared image, and obtain the spatial similarity result by calculating the inner product of the spatial feature vectors;

[0106] The edge feature similarity calculation is as follows: after using the Sobel edge detection algorithm to extract the edge features of the visible light image and the infrared image, obtain the edge similarity result by calculating the inner product of the edge feature vectors;

[0107] The texture feature similarity calculation is as follows: after using the Gabor filter method to extract the texture features of the visible light image and the infrared image, obtain the texture similarity result by calculating the inner product of the texture feature vectors;

[0108] The frequency domain feature similarity calculation is as follows: extract the features of the visible light image and the infrared image in the frequency domain through Fourier transform, and calculate the inner product of the frequency domain feature vectors to obtain the frequency domain similarity result.

[0109] In the above process of similarity calculation, the spatial features, edge features, and texture features of the visible light and infrared images are extracted through convolution operators; through Fourier transform, the image can be transformed from the spatial domain to the frequency domain to extract its frequency domain features. First, perform a discrete Fourier transform (DFT) on the image, then centralize the spectrum, move the low-frequency components to the center position, and calculate the amplitude spectrum and phase spectrum of the image to form a frequency domain feature vector. The corresponding similarity results are obtained by calculating the inner product of each feature vector, and the calculated spatial similarity result, edge similarity result, texture similarity result, and frequency domain similarity result are averaged to obtain the final average similarity result.

[0110] Furthermore, the obtained final average similarity result is fed back to the reinforcement learning model as a reward function, and the model neural network uses the average similarity result to approximate the target value, and then outputs updated actions (tiny movements and scalings in the horizontal and vertical directions in the stitched image), where the target value is the maximum similarity;

[0111] After each action is executed, the system will update the position of the sub - map, and compare the updated sub - map with the visible - light image again. Through multiple iterations, the reinforcement learning model gradually adjusts the position of the sub - map to maximize the average similarity between its image feature vector and the visible - light image feature vector.

[0112] For each sub - map, the above method is used to compare it with the corresponding visible - light image to obtain the maximum average similarity, and the current position is determined;

[0113] It should be noted that when the obtained average similarity reaches the maximum value, it indicates that the current position of the sub - map is the best position for image stitching. According to the position information at this time, this group of sub - maps is stitched to obtain the stitched infrared all - sky image.

[0114] Due to differences in brightness, hue, contrast, etc. between different sub - maps, even if the position has been adjusted to the optimal, obvious boundary lines or seams may still appear in the stitched image, and the stitching boundary needs to be processed for transition;

[0115] Among them, first, by calling the image - processing functions inside OpenCV, the method of adaptive histogram equalization is used to normalize the image illumination.

[0116] Specifically, the stitched infrared all - sky image is used as the input, and it is converted into a grayscale image. The grayscale image of the infrared all - sky image is divided into blocks, and then histogram equalization is performed on each block. Among them, the block size in the process of image block division can be adjusted according to the actual situation, and specific parameters are not limited in this application.

[0117] Furthermore, the motion blur of the infrared all - sky image after processing the illumination change is eliminated. By estimating the noise power spectrum of the image and combining the known motion - blur point - spread function (PSF) to construct a Wiener filter, the Wiener filter is applied to de - blur the image to reduce the influence of motion blur.

[0118] Even further, the infrared all - sky image after the above - mentioned illumination change and motion - blur processing is color - corrected. The color space of the image is linearly transformed to the absolute RGB color space through the CCM (color correction matrix), thereby realizing color correction.

[0119] It should be noted that the above methods of illumination normalization, eliminating motion blur, and image color correction are all implemented by calling the open - source functions in OpenCV. The processing methods and function - calling methods are well - known in the art and will not be elaborated here.

[0120] After improving the overall consistency of the image through the above image processing technology, perform smoothing processing on the edges of the processed image;

[0121] Specifically, for the blank part of the splicing, bilinear interpolation is used; for the large difference between sub-images, an edge detection algorithm is used to detect the significant edges in the splicing boundary area, and bilateral filtering (or median filtering) is applied to smooth the detected edges and reduce the mutation at the edges; non-local means filtering is used to reduce the noise at the splicing boundary and improve the overall smoothness of the image, ensuring that the spliced image is seamless and continuous, and obtaining the spliced infrared all-sky cloud image.

[0122] In a possible implementation method of all-sky data integration, first preprocess the input infrared all-sky image and visible light all-sky image respectively, and synchronize the time of the visible light and the spliced infrared sub-image. Filter the collected outliers for subsequent multimodal fusion processing.

[0123] Specifically, for infrared image processing, first divide the image into a preset number of sub-images and use the corresponding visible light image as a reference. Calculate the feature similarity between the infrared image and the visible light image through spatial features, texture features, edge features or advanced features extracted by deep learning, and find the most matching splicing position. Determine the splicing position according to the maximum similarity, and align the pixels of the infrared image and the visible light image through scaling and interpolation to ensure that the pixel sizes of the fused images are the same. Reduce the random noise of the fused image through a denoising algorithm (such as Gaussian filtering) to ensure the image quality. Ensure the time synchronization of the visible light and infrared images, and align the shooting times of the two types of images through timestamp matching to ensure their consistency in time. If part of the image is damaged, blank data with the same pixel size and number of channels is generated to keep the algorithm running normally.

[0124] After obtaining the environmental parameters of the target monitoring area and the processed all-sky data through step S200, based on step S300, process the above data to obtain the environmental monitoring remote sensing diagnosis result;

[0125] Extract features from the cloud image based on ResNet, where the extracted images include infrared all-sky images and visible light all-sky images; upsample the extracted feature maps and perform multi-scale feature fusion on the upsampled feature maps to obtain the fused feature maps, and based on the UperNet structure, obtain the segmentation image and the classification result of the cloud image according to the fused feature maps; calculate the cloud amount for the segmentation image to obtain the cloud amount calculation result.

[0126] Specifically, the three channels (RGB) of the visible light image are fused with one channel of the infrared image to form a four-channel fused image as the model input, so that the model can simultaneously utilize the multimodal information of visible light and infrared. Next, the four-channel fused image is feature extracted by ResNet to generate a feature map containing multimodal information. UperNet combines the Pyramid Pooling Module to capture global context information to construct a multi-scale feature fused feature map and add a segmentation head to implement the segmentation task. At the same time, a classification head is added after the global pooling layer to implement the classification task. During the training process, a binary cross entropy loss (BCE Loss) is used for the segmentation task to classify each feature point in the image and obtain a pixel-level segmented image. The multi-class cross entropy loss (Cross-Entro py Loss) is used for the classification task, and the model is optimized by a joint loss function (weighted sum of the classification and segmentation losses). In terms of optimization strategy, the AdamW optimizer and the cosine annealing learning rate scheduler are used to significantly improve the overall performance of classification and segmentation.

[0127] like Figure 5 , which is a schematic diagram of the vertical distribution of cloud types in Beijing, Lhasa and Daihai in one embodiment of the present application. The cloud types include at least one of cirrus, cumulus, stratus and hybrid (stratus and cumulus);

[0128] By counting the number of marked cloud pixel blocks, the ratio of the marked cloud pixel blocks to the number of pixel blocks in the entire segmented image is calculated, that is, the ratio of the marked cloud pixel blocks to the entire sky image is taken as the cloud amount result.

[0129] The full sky image input in the segmentation process is a sequence of pictures of the same size arranged in time.

[0130] In the above process, the UperNet model needs to be trained. In the embodiment of the present application, a large amount of labeled cloud image data under different weather conditions can be used to train the neural network model.

[0131] The segmented image and the corresponding cloud amount and cloud type result data are visualized in the all-sky monitoring module for display, such as Figure 2As shown in the left part of the user page. The images shown in the cloud amount annotation result part are respectively the visible light full-sky image (or infrared full-sky image) and the semantic segmentation result image of the corresponding visible light full-sky image (or the segmentation image of the infrared full-sky image), and also include the ultraviolet intensity and the cloud amount analysis result. As Figure 6 The schematic diagram of the sub-page of the full-sky monitoring module shown also specifically shows the historical data of the ultraviolet intensity at different times and the percentages and solar radiation values of the cloud amount analysis results at different times.

[0132] After obtaining the cloud amount result, the cloud displacement vector is calculated by using the cross-correlation method to calculate the two adjacent cloud amount segmentation results before and after in time, as Figure 7 shown in the schematic diagram of the cloud vector displacement of an embodiment of the present application; the wind displacement vector is calculated through vector calculation according to the wind direction and wind speed. As Figure 8 , the calculated cloud displacement vector and the wind displacement vector are used to calculate the remote sensing diagnosis result of the correlation between the cloud displacement vector and the wind displacement vector by using the Pearson correlation coefficient; the regression coefficients of rainfall (Rainfall) and cloud cover (Cloud_Cover), temperature (Temperature) and cloud amount, humidity (Humidity) and cloud amount, and humidity and rainfall are calculated by using univariate regression analysis to obtain the remote sensing diagnosis result, that is, the correlation of meteorological parameters. As Figure 9 shows the schematic diagram of the remote sensing diagnosis result of the correlation analysis between the cloud amount and meteorological parameters of the embodiment of the present application.

[0133] It should be noted that the cross-correlation method used in the above process to calculate the cloud cluster displacement and the method of using the regression coefficient to calculate the correlation are conventional technical means in the field.

[0134] The multi-parameter correlation provides important information for predicting the cloud cluster distribution in the future moment. This method adopts the Mask-RCNN model with Swin-Transformer as the backbone network, and combines multi-modal learning technology to achieve accurate prediction of the cloud cluster distribution. The core idea is to use the semantic segmentation time series results generated by UperNet as the input, integrate the meteorological parameter features, and construct a framework with both spatio-temporal modeling ability and multi-modal fusion ability to improve the accuracy of cloud cluster distribution prediction. The specific process is as follows:

[0135] Organize the UperNet semantic segmentation results at the current moment into time series data as the main input features of the model; at the same time, regard the segmentation results generated by UperNet at the next moment as the target label (Cover Label), as Figure 10As shown, it is used to measure the difference between the model prediction and the true distribution. While constructing the input data, external meteorological parameters (such as temperature, humidity, wind speed, etc.) are processed through normalization or standardization to make their distribution range consistent with the characteristics of the segmentation result and strictly aligned with the image sequence on the time axis to ensure the temporal consistency of the data. A small Transformer is used to encode the features of the meteorological parameter time series, converting the high-dimensional features of the meteorological parameter time series into a structured feature representation. The encoded meteorological features and the time series segmentation result are combined through feature concatenation to combine multiple groups of features in the channel dimension (or other specified dimensions), providing a comprehensive input containing multi-modal information, spatio-temporal information, and global context information for Mask-RCNN.

[0136] In model training, the binary cross-entropy loss (BCE Loss) function is used to measure the difference between the predicted cloud mass distribution of the model and the target segmentation result to optimize the parameters of the meteorological data encoding module and Mask-RCNN. The optimization strategy adopts an adaptive learning rate adjustment mechanism (AdamW optimizer combined with the cosine annealing learning rate strategy) to achieve stable and efficient prediction. The prediction results can be queried after selecting the future data option in the all-sky monitoring module on the selection display page.

[0137] Another part of the meteorological prediction uses past data to accurately predict future weather changes in local areas. Based on the requirements of ODL (online deep learning), the received meteorological data and air quality data are stored in csv format in the form of a time series, such as Figure 11As shown, it includes temperature, humidity, wind direction, wind speed, and air pressure. Then, the stored data is input into the time series prediction model PatchTST to obtain the prediction results. Each element in the multivariate time series is processed separately, that is, each dimension is input into the prediction model, and then the obtained prediction results are concatenated along the dimension direction. This model can learn the long-term dependencies in historical meteorological data and continuously update and fine-tune the parameters in real time as new data is supplemented on the platform. During the training process, a weighted mean squared error (MSE) loss function is used to independently optimize each dimension of the multivariate time series, where the loss of each dimension is assigned different weights according to its importance for meteorological prediction to improve the prediction accuracy of key meteorological factors. To adapt to the changes in real-time meteorological data, an incremental fine-tuning mechanism is introduced to perform mini-batch updates on the newly received data blocks, and only local training is performed on the latest data window. At the same time, the low-level feature extraction part of the network (such as the embedding layer or the first few layers of the encoder) is frozen, and only the parameters of the high-level attention module and the output layer are optimized, thereby significantly reducing the computational cost. The optimization of the online learning process is achieved through a dynamic learning rate adjustment strategy, adopting an adaptive learning rate adjustment mechanism (AdamW optimizer combined with the cosine annealing learning rate strategy). The learning rate is increased to accelerate convergence when the error drops rapidly, and the learning rate is decreased to refine parameter optimization when the error tends to be stable. In addition, combined with the model drift detection method, the difference between the new data distribution and the historical data distribution is evaluated by calculating the KL divergence of the distribution (taking a 1-hour preprocessing period as an example) to judge the degree of deviation of the data distribution. When a significant drift is detected, further global optimization of the model is triggered to ensure the adaptability of the model to rapidly changing meteorological conditions and the stability of long-term prediction.

[0138] Furthermore, the above deep learning method can be widely deployed at different altitudes and regions, collect data in real time and converge the remote sensing diagnosis results to the cloud, and perform big data operations and construct a space-ground integrated large model in combination with the remote sensing raster data collected by the cloud platform to improve the time resolution in remote areas; generate multi-parameter statistical distributions, provide a precise meteorological dynamic vector model from local micro to overall macro, and give the dynamic change diagnosis results of meteorological all-sky and other parameters, that is, the prediction of future dynamic changes and the influencing factors among them.

[0139] Specifically, the knowledge distillation-based method can further save the computing resources required by the deep learning method to efficiently perform edge computing. Such as Figure 12The figure shows a schematic diagram of the loss calculation process for cloud map classification and cloud amount calculation in an embodiment of this application. Taking UperNet as the teacher model and MobileNet V3 as the student model. The student model shares the Backbone of MobileNetV3, designs a global average pooling layer (GAP) plus a fully connected layer as the classification head, and combines the DeepLab V3+ structure as the segmentation head to achieve multi-scale feature decoding. During the training process, the classification task is optimized by combining the multi-class cross-entropy loss (true label) and the KL divergence loss (soft label of the teacher model); the segmentation task is jointly optimized by the weighted sum of the binary cross-entropy loss (BCELoss) and the KL divergence loss. By adjusting the task loss weights and introducing the AdamW optimizer and the cosine annealing learning rate strategy, the learning efficiency and inference performance of the lightweight student model in classification and segmentation tasks are significantly improved, which is suitable for efficient intelligent task deployment in resource-constrained scenarios.

[0140] After the Mask-RCNN model based on Swin-Large (large-scale Swin-Transformer pre-trained model) as the backbone network is trained, the knowledge distillation method is adopted, and it is used as the teacher model to guide the training of the Mask-RCNN model (i.e., the student model) with Swin-Tiny (small-scale Swin-Transformer pre-trained model) as the backbone network. The student model is jointly optimized by the weighted sum of the supervised loss of the true label and the distillation loss of the teacher model, and gradually learns the knowledge of the teacher model. Specifically, the output of the student model and the true label are used to calculate the binary cross-entropy loss, and at the same time, the KL divergence or feature matching loss is calculated with the output of the teacher model, and the weighted sum of the two parts of the loss is used as the total loss of the student model, so as to achieve the efficient learning of the student model for detection and segmentation tasks under lightweight conditions, as Figure 13 shown.

[0141] Similarly, as Figure 14The schematic diagram of the meteorological prediction loss calculation process shown below uses PatchTST as the teacher model and Transformer as the student model for meteorological prediction. The output of the teacher model is used as a high-dimensional feature guidance signal, and the mean squared error (MSE) loss function is used to calculate the difference between the teacher model PatchTST and the lightweight Transformer of the student model in time series prediction. The meteorological prediction result of Patch TST at the next moment is used as the label y, and the loss calculation result is obtained by calculating the MSE. Then, the losses of the teacher model and the student model are weighted and fused by the weight coefficients α and (1 - α) respectively to balance the losses, ensuring that in the backpropagation process, the student model can not only learn the high-level feature representation of the teacher model but also retain its own characteristics, ultimately improving the prediction performance of the student model and achieving model compression, saving a large amount of space and then realizing lightweight deployment.

[0142] The environmental monitoring remote sensing diagnosis results obtained by integrating and calculating based on the above deep learning method have multiple uses. Based on the multi-faceted ground-based remote sensing data (all-sky data) and meteorological element analysis provided by site-based deployment, combined with information such as cloud cluster distribution, wind field, humidity, and temperature, accurately track the typhoon path and evaluate the intensity change, and generate a typhoon path prediction map. Real-time early warning of extreme weather events such as tornadoes and hailstorms, and provide an event impact prediction report. At the same time, by analyzing cloud water content, rainfall distribution, and surface runoff data, combined with topographic features and soil water absorption capacity, simulate flood risk areas, and generate a rainfall concentration analysis and flood risk assessment report. Combine information such as humidity, cloud cover, and precipitation to monitor the farmland water status in real time, diagnose drought stress areas, and generate a water monitoring report. Analyze the impact of cloud shadow changes and light distribution on crop growth and photovoltaic power generation, and generate a drought risk assessment report and a light optimization analysis map.

[0143] This application also provides a ground-based multi-parameter intelligent environmental monitoring remote sensing diagnosis system, including a data acquisition module, a data processing module, and a result diagnosis module. The working process is as Figure 15As shown. In an embodiment of the present application, after the instrument is started, each sensor of the data acquisition module starts to run, and at preset intervals (which can refer to the preset frequencies of each sensing module in the ground-based multi-parameter intelligent environmental monitoring remote sensing diagnosis method), the collected environmental parameters and all-sky data are continuously uploaded to the edge computing platform (such as: NVIDIA TX2) connected to the same switch through a coprocessor. Specifically, with the help of the MQTT communication protocol, the data is sent to the data processing module deployed on the edge computing platform, and the corresponding data processing is performed by the deep learning model of the data processing module. The environmental monitoring results and remote sensing diagnosis results obtained through the deep learning model are synchronized to the database of the cloud display system of the result diagnosis module through the HTTP protocol. The cloud display system based on the front-end and back-end architecture design integrates the processed data and displays the final result.

[0144] The data acquisition module integrates a variety of sensors to obtain meteorological parameters, air quality parameters and all-sky data corresponding to the current target monitoring area, collects data in real time and sends the data to the data processing module via the MQTT communication protocol for further processing.

[0145] After receiving the environmental monitoring data of the target monitoring area, the data processing module processes the all-sky image based on the image stitching algorithm and UperNet and its distilled model to obtain the cloud cover analysis and cloud map classification results; predicts the meteorological data based on PatchTST and its distilled model to obtain the meteorological prediction results, and obtains the cloud cluster prediction results based on Mask-RCNN and its distilled model. The deep learning algorithm of the data processing module adopts the deep learning model optimization and deployment method based on the Jetson platform, comprehensively uses TorchScript serialization, TensorRT inference acceleration, mixed precision calculation (FP16) and quantization technology (INT8), significantly improving the inference speed; configures a running environment compatible with CUDA, cuDNN and PyTorch to optimize the system environment configuration; combines small model design or pruning technology to achieve model lightweight; uses multi-threaded parallelism and efficient data loading process to further improve the inference efficiency. This method significantly improves the performance and stability of model deployment on the Jetson platform through comprehensive optimization design. Finally, the data processing module synchronizes the integrated environmental monitoring remote sensing diagnosis results output based on the deep learning algorithm to the back-end database of the cloud display system of the result diagnosis module through the HTTP protocol. The data processing module can also be deployed on a cloud server; when processing real-time meteorological data, the cloud server uses a cloud large model combined with multi-source cloud data for instant calculation, which can effectively improve the accuracy and efficiency of the system.

[0146] The result diagnosis module displays the remote sensing diagnosis results of environmental monitoring in real time and monitors the conditions when the parameters exceed the threshold. For example, some possible parameter threshold setting cases: temperature exceeds 50℃; humidity is less than 20%; cloud coverage exceeds 90%.

[0147] Users can customize parameter thresholds in the system according to their needs. When a parameter exceeds the set threshold, the page will trigger a warning prompt and activate the sound alarm function equipped in the result diagnosis module to attract the user's attention. At the same time, the system background stores the user's mobile phone number and other communication methods, and sends relevant notifications via SMS to ensure that users can understand abnormal situations in the first place.

[0148] In terms of data storage, the result diagnosis module uses MySQL to support data persistence. Combined with Laravel's Eloquent ORM (object-relational mapping) mechanism, data mapping objects can be directly called to simplify database operations and improve development efficiency.

[0149] The result diagnosis module implements functions based on the cloud visualization system. The cloud visualization system uses the Vue.js framework. It effectively improves the rendering efficiency of the page, and combines the fragment display algorithm and the database filtering algorithm to ensure accurate information display while optimizing data loading speed. The workflow based on the cloud display system is as follows: Figure 16 As shown, the user initiates a request as a client and transmits the user's interactive actions, such as clicking, typing, etc. The display system deployed on the server starts working after receiving the request. The front end receives the action from the user based on Vue.js and sends the processed request to the back end. The back end processes the request sent by the front end based on the PHP language, communicates with the database according to the request, obtains or updates the data, and then sends the result data back to the front end. Among them, the front-end display page consists of a main page and three sub-pages, which are the same as the pages mentioned in the ground-based multi-parameter intelligent environmental monitoring remote sensing diagnosis method. Based on the above design, the data visualization system can share cloud data and realize real-time display functions on remote portable devices, such as Figure 17 The main page for mobile users is shown.

[0150] Through the above steps, the intelligent environmental monitoring remote sensing diagnosis system for multi-parameters of the foundation in an embodiment of the present application can collect various types of environmental data in real time, and process environmental parameters and all-sky data through the integrated image stitching algorithm and deep neural network algorithm to obtain environmental monitoring remote sensing diagnosis results such as environmental parameters, meteorological prediction parameters, correlation of meteorological parameters, processed all-sky data (all-sky cloud map), cloud amount and cloud type in the corresponding image, correlation between cloud displacement vector and wind vector, and cloud cluster distribution prediction results, and store the result data in the back-end database of the cloud visualization system; by integrating a variety of sensor technologies and data processing methods, it realizes comprehensive and real-time monitoring of the environment, and at the same time can efficiently process multi-source data, improving the overall efficiency and accuracy of environmental monitoring remote sensing diagnosis; it also designs a front-end user page based on the Vue.js framework to realize the visualization of the system, providing an intuitive user page and a convenient operation process, and reducing the difficulty of use and maintenance.

[0151] The embodiments of the present application have been described above. The above description is exemplary and not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art in the technical field without departing from the scope and spirit of the described embodiments. The selection of the terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other ordinary skill in the art in the technical field to understand the embodiments disclosed herein.

Claims

1. A ground-based multi-parameter intelligent environmental monitoring remote sensing diagnosis method, characterized in that: The steps include: Obtain meteorological parameters, air quality parameters and all-sky data corresponding to the current target monitoring area; Integrating the meteorological parameters and the air quality parameters to obtain environmental parameters of the target monitoring area; and preprocessing the full sky data to obtain processed full sky data; The environmental monitoring and remote sensing diagnosis results of the target monitoring area are obtained according to the environmental parameters and the processed all-sky data.

2. The method according to claim 1, characterized in that The meteorological parameters include at least one of temperature, humidity, wind speed, wind direction, atmospheric pressure, noise, ultraviolet intensity and rainfall; the air quality parameters include common chemical gases (NH3, H2S, SO2, NO2, O3, CO), volatile organic compounds (TVOC, HCHO), greenhouse gases (CO2, eCO2), respiratory health factors (PM 10 、PM 2.5 、PM 1.0 , negative oxygen ions) concentration.

3. The method according to claim 1, characterized in that When the meteorological parameters and the air quality parameters are integrated, the step of filtering abnormal values ​​in the meteorological parameters and the air quality parameters to obtain the filtered regional environmental parameters is included.

4. The method according to claim 1, characterized in that: When obtaining the full sky data corresponding to the current target monitoring area, it includes: Obtain visible light all-sky data through a visible light all-sky instrument; By controlling the rotation angle and speed of the gimbal, multi-angle images are collected according to the preset angles to obtain multiple infrared full-sky data at different shooting angles.

5. The method according to claim 1, characterized in that When the full sky data is preprocessed, it includes the steps of splicing the multiple infrared sky images collected and performing transition processing on the splicing boundaries to obtain the processed full sky data.

6. The method according to claim 1, characterized in that When the environmental monitoring and remote sensing diagnosis results of the current target monitoring area are obtained according to the environmental parameters and the processed all-sky data, it includes: Obtaining weather forecast data according to the environmental parameters; generating cloud amount and cloud type according to the processed all-sky data; Obtaining environmental monitoring results based on the environmental parameters, meteorological forecast data, and the cloud amount and cloud type; Obtaining a remote sensing diagnosis result according to the processed all-sky data and the environmental monitoring result; The remote sensing diagnosis results are jointly characterized by meteorological parameter correlation, cloud displacement vector, correlation between cloud displacement vector and wind vector, and cloud distribution prediction results.

7. The method according to claim 5, characterized in that When stitching multiple infrared sky images, including: Taking the infrared image as a sub-image and the visible light full sky image as a reference benchmark, determining the stitching position of each of the infrared images; After the stitching positions of the infrared images are determined, the infrared images are stitched according to the stitching positions and the stitching boundaries are transition processed.

8. The method according to claim 6, characterized in that When generating the cloud amount and cloud type, include: capturing features in the processed all-sky data; Classifying clouds based on the captured features to obtain cloud classification results; Clouds are identified from the full sky data, and the proportion of the identified clouds in the full sky data is counted to obtain the coverage of clouds in the sky as the cloud amount result.

9. A ground-based multi-parameter intelligent environment monitoring remote sensing diagnosis system, characterized in that: include: Data acquisition module, data processing module, result diagnosis module; The data acquisition module is configured to acquire meteorological parameters, air quality parameters and full sky data corresponding to the current target monitoring area; The data processing module is configured to integrate the meteorological parameters and the air quality parameters to obtain the environmental parameters of the target monitoring area; and pre-process the full sky data to obtain processed full sky data; and obtain the environmental monitoring and remote sensing diagnosis results of the target monitoring area according to the environmental parameters and the processed full sky data; The result diagnosis module is configured to display the environmental monitoring and remote sensing diagnosis results and respond according to the diagnosis results.

10. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images