Low-altitude unmanned aerial vehicle environment monitoring data acquisition method

Abstract

The invention relates to the technical field of environment monitoring, and discloses a low-altitude unmanned aerial vehicle environment monitoring data acquisition method, which comprises the following steps: selecting a multi-rotor unmanned aerial vehicle as a flight platform, and carrying a thermal infrared sensor, a hyperspectral sensor, a data acquisition module and a positioning module; planning the flight height and speed of the unmanned aerial vehicle and the sampling time interval of the sensor according to the underlying surface type of the monitoring area and a preset route overlapping rate; controlling the unmanned aerial vehicle to fly according to a planned route, synchronously acquiring thermal infrared image data and hyperspectral image data, and transmitting the data to a ground control system in real time; a convolutional neural network is used to process the thermal infrared image to generate a temperature distribution map, and a U-Net network is used to process the hyperspectral image to generate a ground feature classification map. The low-altitude unmanned aerial vehicle environment monitoring data acquisition method aims at solving the problems that in a traditional urban thermal environment monitoring method, the monitoring range is limited, data updating is not timely, and the influence of weather and traffic is large.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental monitoring, and specifically to a method for collecting environmental monitoring data by a low-altitude unmanned aerial vehicle (UAV). Background Art

[0002] UAV environmental monitoring is an advanced technology that uses UAVs equipped with various sensors and devices to collect and analyze environmental data. By flying at high altitudes, UAVs can quickly cover large areas and conduct real-time monitoring of multi-dimensional environmental indicators such as air quality, water quality, soil pollution, vegetation coverage, and ecological diversity. This technology not only improves the efficiency and accuracy of data collection but also greatly reduces the risk of personnel entering complex or dangerous environments. Devices such as high-definition cameras and multi-spectral imagers carried by UAVs can visually display environmental changes and provide timely and accurate data support for environmental protection departments.

[0003] Traditional urban thermal environment monitoring methods have significant drawbacks. First, the monitoring range is limited, mainly relying on ground fixed stations or vehicle-mounted devices, making it difficult to cover all areas of the city, especially inaccessible corners and areas with dense high-rise buildings, resulting in potentially incomplete monitoring results. In addition, traditional methods are affected by various factors such as weather and traffic, which may lead to monitoring interruptions or data loss, affecting the continuity and accuracy of monitoring. Therefore, a method for collecting environmental monitoring data by a low-altitude UAV is proposed to solve the above-mentioned problems. Summary of the Invention

[0004] The purpose of the present invention is to address the problems of limited monitoring range, untimely data update, and high susceptibility to weather and traffic in traditional urban thermal environment monitoring methods, and to propose a method for collecting environmental monitoring data by a low-altitude UAV.

[0005] The technical solution for the present invention to solve the above technical problems is as follows:[[]]

[0006] A method for collecting environmental monitoring data by a low-altitude UAV includes the following steps:[[]]

[0007] S10. UAV platform construction: Select a multi-rotor UAV as the flight platform and equip it with a thermal infrared sensor, a hyperspectral sensor, a data acquisition module, and a positioning module.[[]]

[0008] S20. Flight route planning: According to the underlying surface type of the monitoring area and the preset flight line overlap rate, plan the flight altitude, speed of the UAV, and the sampling time interval of the sensors.[[]]

[0009] S30. Synchronous data collection: Control the UAV to fly according to the planned route, synchronously collect thermal infrared image data and hyperspectral image data, and transmit the data to the ground control system in real time.[[]]

[0010] S40. Deep learning data processing: Use a convolutional neural network to process thermal infrared images to generate a temperature distribution map, and use a U-Net network to process hyperspectral images to generate a land cover classification map;

[0011] S50. Thermal environment parameter analysis: Combine the temperature distribution map and the land cover classification map to calculate the heat island intensity distribution, vegetation coverage, and normalized difference vegetation index;

[0012] S60. Visualization and optimized output: Display the analysis results through a visualization tool and generate an optimized urban thermal environment plan.

[0013] Based on the above technical solutions, the present invention can also be improved as follows.

[0014] Furthermore, in S10, the wavelength range of the thermal infrared sensor is 7.5 - 13.5 μm, and the wavelength range of the hyperspectral sensor is 400 - 1000 nm. And the data acquisition module includes a radiation correction unit and a timestamp synchronization unit. The radiation correction unit is a component for performing radiometric calibration and correction on the acquired thermal infrared images and hyperspectral images. The radiation correction unit performs pixel-by-pixel radiometric calibration and correction on the acquired image data through preset calibration parameters and correction algorithms. The timestamp synchronization unit records the timestamps of the image data acquired by each sensor and synchronizes them. The positioning module is a dual-frequency RTK positioning module with a positioning accuracy of centimeter level, and the thermal infrared sensor and the hyperspectral sensor are fixed by a rigid bracket to ensure spatial consistency during data acquisition.

[0015] Furthermore, in S20, the underlying surface types of the monitoring area include, but are not limited to, urban construction areas, vegetation-covered areas, water areas, bare areas, and agricultural land. Among them, the planned flight altitude range is 50 - 150 meters, the flight line overlap rate is not less than 70%, the sampling time interval of the thermal infrared sensor is 1 - 5 seconds, and the observation angle of the hyperspectral sensor is vertically downward ±15°.

[0016] Furthermore, S30 specifically includes the following steps:

[0017] S301. Start the drone and execute the flight plan

[0018] The operator starts the drone through the ground control system and inputs the parameters of the pre-planned flight route, flight altitude, and speed. After receiving the instructions, the drone starts to fly autonomously according to the planned route;

[0019] S302. Synchronously start the sensors to collect data

[0020] During the flight of the drone, the ground control system synchronously activates the thermal infrared sensor and the hyperspectral sensor. The sensors start to collect data according to the preset sampling time interval and observation angle. The thermal infrared sensor is used to capture the surface temperature information, while the hyperspectral sensor is used to obtain detailed information on the composition and structure of surface substances.

[0021] S303. Data preprocessing and packaging

[0022] The raw data collected by the sensors is preprocessed, including steps such as denoising and calibration, to ensure the accuracy and reliability of the data. The preprocessed data is packaged into a transmission format and transmitted to the ground control system in real time.

[0023] S304. Real-time data transmission to the ground control system

[0024] The drone transmits the packaged data to the ground control system in real time through wireless communication technology. After receiving the data, the ground control system will perform decompression and parsing processes for subsequent data analysis and processing.

[0025] Furthermore, in the conversion process from the radiation intensity of the thermal infrared image to the surface temperature in S40, the temperature inversion is achieved through the following formula:

[0026] T = f(λ, I)

[0027] where T is the underlying surface temperature, λ is the wavelength collected by the thermal infrared sensor, I is the radiation intensity of the thermal infrared image, and f is a non-linear mapping function obtained by training with a CNN model.

[0028] The generation of the land cover classification map includes the following steps:

[0029] After performing band fusion and noise suppression on the hyperspectral image, it is input into the U-Net network for pixel-level classification, and the classification results including buildings, vegetation, water bodies, and bare land are output. The classification accuracy is evaluated through the following formula:

[0030] Accuracy = (TP + TN) / (TP + TN + FP + FN)

[0031] where TP is the true positive, TN is the true negative, FP is the false positive, and FN is the false negative.

[0032] Furthermore, in the calculation process of the heat island intensity distribution in step S50, the highest temperature concentration area is accurately identified and extracted from the temperature distribution map and defined as the heat island core area. Subsequently, combined with the land cover classification map, especially the distribution information of building density, a quantitative analysis of the heat island intensity is carried out. The calculation of the heat island intensity gradient value UHI adopts the following formula:

[0033]

[0034] Among them, T i and T j respectively represent the temperature values of each pixel in the heat island core area and the overall monitoring area; A i and A j correspond to the areas represented by these pixels respectively; represents the weighted average temperature of the heat island core area, that is, T max ; represents the weighted average temperature of the monitoring area, that is, T avg , to quantitatively evaluate the intensity of the heat island effect. The heat island intensity gradient value is further combined with wind speed and humidity data to predict the future 6-hour heat island diffusion trend through a long short-term memory network model.

[0035] Furthermore, the visualization tool in the S60 is a geographic information system platform, and the optimization plan includes specific suggestions for increasing the green area, adjusting the building layout, and improving the material reflectivity.

[0036] Furthermore, the input of the U-Net network is the result of dimensionality reduction by band principal component analysis of the hyperspectral image, the output layer uses the Softmax activation function, and the classification result is spatially registered and verified with the lidar point cloud data. The convolutional neural network adopts the ResNet-50 architecture, the training data includes historical thermal infrared images and corresponding ground measured temperature data, and the loss function is the mean square error.

[0037] Furthermore, the data transmission module in the S30 communicates with the ground control system using a 5G network, and the data compression algorithm is lossy compression based on wavelet transform, and the compression ratio is not less than 10:1.

[0038] Compared with the prior art, the technical solution of the present application has the following beneficial technical effects:

[0039] The present invention selects a multi-rotor unmanned aerial vehicle (UAV) as the flight platform. With its flexibility and low-altitude flight ability, the multi-rotor UAV can easily cover complex terrains and high-rise building-intensive areas in the city, achieving a great expansion of the monitoring range. At the same time, a thermal infrared sensor, a hyperspectral sensor, a data acquisition module, and a positioning module are carried on the UAV platform. These high-precision sensors can collect multi-dimensional environmental data in real time, ensuring the comprehensiveness and accuracy of the data. According to the underlying surface type of the monitoring area and the preset flight line overlap rate, the UAV can fly and collect data according to the planned altitude, speed, and sampling time interval. During the flight process, the UAV can synchronously collect thermal infrared image data and hyperspectral image data, and transmit these data to the ground control system in real time. The thermal infrared images are processed by a convolutional neural network to generate a temperature distribution map, and the hyperspectral images are processed by a U-Net network to generate a land cover classification map. This deep learning method can efficiently process a large amount of data, extract valuable information, and improve the accuracy and efficiency of data processing. By combining the temperature distribution map and the land cover classification map, key thermal environment parameters such as the heat island intensity distribution, vegetation coverage, and normalized difference vegetation index are calculated. These parameters provide a scientific basis for the evaluation and optimization of the urban thermal environment, helping to formulate more precise environmental protection measures. Finally, the analysis results are displayed through a visualization tool, and an urban thermal environment optimization plan is generated. This visualization output method makes the analysis results more intuitive and understandable, providing clearer decision-making support for environmental protection departments. At the same time, the proposed optimization plan also provides feasible paths and measures for improving the urban thermal environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flowchart of a method for collecting environmental monitoring data by a low-altitude UAV according to the present invention;

[0041] Figure 2 is a flowchart of a method for synchronous data collection in a method for collecting environmental monitoring data by a low-altitude UAV according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0043] Combined with Figure 1 - Figure 2 as shown, a method for collecting environmental monitoring data by a low-altitude UAV according to the present invention includes the following steps:

[0044] S10. UAV Platform Setup: Select a multi-rotor UAV as the flight platform and carry a thermal infrared sensor, a hyperspectral sensor, a data acquisition module, and a positioning module;

[0045] S20. Flight Route Planning: According to the underlying surface type of the monitoring area and the preset flight line overlap rate, plan the flight altitude, speed of the UAV, and the sampling time interval of the sensor;

[0046] S30. Data Synchronous Acquisition: Control the UAV to fly according to the planned route, synchronously acquire thermal infrared image data and hyperspectral image data, and transmit the data to the ground control system in real time;

[0047] S40. Deep Learning Data Processing: Use a convolutional neural network to process the thermal infrared image to generate a temperature distribution map, and process the hyperspectral image through a U-Net network to generate a land cover classification map;

[0048] S50. Thermal Environment Parameter Analysis: Combine the temperature distribution map and the land cover classification map to calculate the heat island intensity distribution, vegetation coverage, and normalized difference vegetation index;

[0049] S60. Visualization and Optimization Output: Display the analysis results through a visualization tool and generate an urban thermal environment optimization plan.

[0050] In a preferred embodiment of the present invention, it can be further configured as follows: in S10, the wavelength range of the thermal infrared sensor is 7.5 - 13.5 μm, and the wavelength range of the hyperspectral sensor is 400 - 1000 nm. The data acquisition module includes a radiation correction unit and a timestamp synchronization unit. The radiation correction unit is a component for performing radiometric calibration and correction on the acquired thermal infrared images and hyperspectral images. When the unmanned aerial vehicle conducts environmental monitoring data acquisition, due to factors such as atmospheric absorption, scattering, and the non-linear response of the instrument itself, there will be certain radiation errors in the acquired image data. If these errors are not corrected, it will seriously affect the generation accuracy of the subsequent temperature distribution map and ground object classification map. The radiation correction unit performs pixel-by-pixel radiometric calibration and correction on the acquired image data through preset calibration parameters and correction algorithms to reduce these errors. This process ensures that the acquired image data can truly and accurately reflect the actual radiation characteristics of the ground surface, providing a reliable data basis for subsequent thermal environment parameter analysis and the formulation of optimization schemes. The timestamp synchronization unit is a component that ensures the consistency of the timestamps of the thermal infrared images and hyperspectral images during acquisition. During the flight of the unmanned aerial vehicle, due to factors such as flight speed, sensor sampling frequency, and data transmission delay, there may be slight differences in the time of the image data acquired by different sensors. If these time differences are not synchronized, it will lead to mismatches between the image data, thereby affecting subsequent data fusion and analysis. The timestamp synchronization unit records the timestamps of the image data acquired by each sensor and synchronizes them, thus ensuring the temporal consistency of the thermal infrared images and hyperspectral images. This process provides a basis for subsequent data fusion, enabling the image data acquired by different sensors to accurately correspond to each other and jointly reflect the actual situation of the ground surface. The positioning module is a dual-frequency RTK positioning module with a positioning accuracy of centimeter level. The thermal infrared sensor and the hyperspectral sensor are fixed by a rigid bracket to ensure spatial consistency during data acquisition. By specifying the wavelength ranges of the thermal infrared sensor and the hyperspectral sensor, the pertinence and accuracy of data acquisition are improved. The thermal infrared sensor can more sensitively capture the long-wave radiation information of the ground surface within this wavelength range, which is crucial for monitoring environmental parameters such as ground surface temperature distribution and thermal anomalies. The hyperspectral sensor has a higher spectral resolution within the range of 400 - 1000 nm and can obtain rich spectral characteristics of ground objects, which helps in ground object classification and identification;

[0051] A radiation correction unit and a timestamp synchronization unit are introduced. The radiation correction unit performs pixel-by-pixel radiometric calibration and correction on the collected thermal infrared images and hyperspectral images through preset calibration parameters and correction algorithms, effectively eliminating the radiation errors caused by factors such as atmospheric absorption, scattering, and instrument nonlinear response, ensuring that the collected image data can truly and accurately reflect the actual radiation characteristics of the ground surface, greatly improving the reliability and accuracy of the data. At the same time, the timestamp synchronization unit records the timestamps of the image data collected by each sensor and performs synchronization processing, solving the problem of possible small time differences in the image data collected by different sensors during the flight of the unmanned aerial vehicle, ensuring the temporal consistency of the thermal infrared images and hyperspectral images, and providing strong support for subsequent data fusion and analysis.

[0052] In a preferred embodiment of the present invention, it can be further configured as follows: The underlying surface types in the monitoring area in S20 include, but are not limited to, urban construction areas, vegetation-covered areas, water bodies, bare areas, and agricultural land. The planned flight altitude range is 50 - 150 meters, the flight line overlap rate is not less than 70%, the sampling time interval of the thermal infrared sensor is 1 - 5 seconds, and the observation angle of the hyperspectral sensor is ±15° vertically downward. By clarifying the monitoring values, the adaptability and accuracy of environmental monitoring data collection are improved. For different types of underlying surfaces, the unmanned aerial vehicle can be flexibly adjusted according to the planned flight altitude and flight line overlap rate to ensure comprehensive coverage and efficient collection of different environmental characteristics. At the same time, by optimizing the sampling time interval of the thermal infrared sensor and the observation angle of the hyperspectral sensor, the resolution and accuracy of data collection are improved, providing a more detailed and reliable basis for subsequent data processing and analysis.

[0053] In a preferred embodiment of the present invention, it can be further configured as follows: S30 specifically includes the following steps:

[0054] S301. Start the unmanned aerial vehicle and execute the flight plan

[0055] The operator starts the unmanned aerial vehicle through the ground control system and inputs the parameters of the pre-planned flight route, flight altitude, and speed. After receiving the instructions, the unmanned aerial vehicle starts to fly autonomously according to the planned route;

[0056] S302. Synchronously start the sensors to collect data

[0057] During the flight of the unmanned aerial vehicle, the ground control system synchronously starts the thermal infrared sensor and the hyperspectral sensor. The sensors start to collect data according to the preset sampling time interval and observation angle. The thermal infrared sensor is used to capture the surface temperature information, while the hyperspectral sensor is used to obtain detailed information on the surface material composition and structure;

[0058] S303. Data preprocessing and packaging

[0059] The original data collected by the sensor is preprocessed, including steps such as denoising and calibration, to ensure the accuracy and reliability of the data. The preprocessed data is packaged into a transmission format and transmitted to the ground control system in real time;

[0060] S304. Real-time data is transmitted to the ground control system

[0061] The drone transmits the packaged data to the ground control system in real time through wireless communication technology. After receiving the data, the ground control system will perform decompression and parsing processes for subsequent data analysis and processing. By starting the sensor data collection through the ground control system, the synchronous startup and collaborative work of the thermal infrared sensor and the hyperspectral sensor are achieved, ensuring the synchrony and consistency of data collection. The thermal infrared sensor captures the surface temperature information, and the hyperspectral sensor obtains detailed information on the surface material composition and structure. The two complement each other and jointly provide comprehensive and accurate data support for environmental monitoring. In the data preprocessing stage, through steps such as denoising and calibration, the noise and errors in the original data are effectively eliminated, improving the accuracy and reliability of the data. The preprocessed data is packaged into a transmission format and transmitted to the ground control system in real time through wireless communication technology, realizing the real-time transmission and reception of data, enabling the operator to timely understand the flight status of the drone and the data collected, providing timely and accurate information support for subsequent data analysis and processing. In addition, after receiving the data, the ground control system will perform decompression and parsing processes for subsequent data analysis and processing. This step further ensures the integrity and readability of the data, providing a solid foundation for subsequent data mining and applications.

[0062] In a preferred embodiment of the present invention, it can be further configured as follows: In the process of converting the radiation intensity of the thermal infrared image to the surface temperature in S40, the temperature inversion is achieved through the following formula:

[0063] T = f(λ, I)

[0064] Where, T is the underlying surface temperature, λ is the wavelength collected by the thermal infrared sensor, I is the radiation intensity of the thermal infrared image, and f is a non-linear mapping function obtained by training with a CNN model;

[0065] The generation of the land cover classification map includes the following steps:

[0066] After performing band fusion and noise suppression on the hyperspectral image, it is input into the U-Net network for pixel-level classification, and the classification results including buildings, vegetation, water bodies, and bare land are output; The classification accuracy is evaluated through the following formula:

[0067] Accuracy = (TP + TN) / (TP + TN + FP + FN)

[0068] Among them, TP is the true positive, TN is the true negative, FP is the false positive, and FN is the false negative. By introducing specific formulas, the conversion from the radiation intensity of the thermal infrared image to the surface temperature is realized, improving the accuracy and reliability of temperature inversion. Traditional temperature inversion methods often rely on simplified physical models or empirical formulas and are difficult to accurately reflect the temperature distribution under complex surface conditions. However, the non-linear mapping function obtained by training with the CNN model can fully consider the influence of various factors such as surface type, atmospheric conditions, and sensor characteristics, thus realizing the accurate inversion of surface temperature. This not only improves the accuracy of temperature data but also provides more reliable data support for subsequent environmental monitoring and climate change research;

[0069] In the processing of hyperspectral images, by fusing bands and suppressing noise, the signal-to-noise ratio and classification accuracy of the images are effectively improved. Subsequently, the processed images are input into the U-Net network for pixel-level classification. This step fully utilizes the advantages of the U-Net network in the field of image segmentation and can achieve a fine division of surface cover types. The classification results include various types such as buildings, vegetation, water bodies, and bare land, providing rich information for environmental monitoring. At the same time, by introducing indicators such as true positives, true negatives, false positives, and false negatives, the classification accuracy is quantitatively evaluated, which can not only intuitively reflect the quality of the classification results but also provide a basis for the optimization and improvement of the model. In addition, this evaluation method of classification accuracy is also applicable to ground object classification tasks under different surface types, different seasons, and weather conditions, and has strong generality and practicability.

[0070] In a preferred embodiment of the present invention, it can be further configured as follows: The calculation process of the heat island intensity distribution in step S50 is to accurately identify and extract the highest temperature concentration area from the temperature distribution map, which is defined as the heat island core area. Subsequently, combined with the land cover classification map, especially the distribution information of building density, a quantitative analysis of the heat island intensity is carried out. The calculation of the heat island intensity gradient value UHI adopts the following formula:

[0071]

[0072] Among them, T i and T j respectively represent the temperature values of each pixel point in the heat island core area and the overall monitoring area; A i and A j respectively correspond to the areas represented by these pixel points; represents the weighted average temperature of the heat island core area, that is, T max ; represents the weighted average temperature of the monitoring area, that is, T avg, to quantitatively evaluate the intensity of the heat island effect, the heat island intensity gradient value is further combined with wind speed and humidity data, and the long short-term memory network model is used to predict the heat island diffusion trend in the next 6 hours. By accurately identifying and extracting the area with the highest temperature concentration as the heat island core area and combining the land cover classification map, especially the distribution information of building density, the quantitative analysis of the heat island intensity is realized. This not only improves the accuracy of heat island effect identification but also enables a deeper understanding of the relationship between the heat island effect and human activities such as urban planning and land use. By introducing the calculation formula of the heat island intensity gradient value UHI and comprehensively considering the temperature value and area weight, the evaluation of the heat island intensity becomes more scientific and accurate;

[0073] Predicting the heat island diffusion trend in the next 6 hours through the long short-term memory network model not only improves the timeliness of heat island effect prediction but also enables a more accurate grasp of the dynamic change law of the heat island effect. As an advanced deep learning algorithm, the LSTM model can capture long-term dependencies in time series data and is suitable for predicting environmental problems with complex spatio-temporal variation characteristics such as the heat island effect. By combining the heat island intensity gradient value UHI with meteorological data, the LSTM model can comprehensively consider the influence of various factors and improve the accuracy and reliability of prediction.

[0074] In a preferred embodiment of the present invention, it can be further configured that: in S60, the visualization tool is a geographic information system platform, and the optimization plan includes specific suggestions for increasing green space area, adjusting building layout, and improving material reflectivity. By using the geographic information system platform as the visualization tool, it can intuitively display the distribution, intensity of the heat island effect and its relationship with factors such as urban planning and land use. This not only improves the readability and understandability of data but also helps decision-makers, urban planners, and the public to more clearly recognize the impact of the heat island effect, thereby taking more effective countermeasures. The geographic information system platform can generate an intuitive and easy-to-understand heat island effect distribution map by integrating multiple data sources and visualization technologies, providing strong data support for fields such as urban planning and environmental protection. At the same time, it also combines the optimization plan, including specific suggestions such as increasing green space area, adjusting building layout, and improving material reflectivity. These suggestions directly target the main causes of the heat island effect and help to fundamentally alleviate the heat island effect; [[ID=,7]]

[0075] Based on the geographic information system platform, further in-depth expansion and optimization can be carried out. For example, time series analysis and prediction models can be introduced to mine and analyze the historical data of the heat island effect to predict the future change trend of the heat island effect, which helps urban managers formulate countermeasures in advance and reduce the impact of the heat island effect on the urban ecological environment and residents' lives. In addition, the geographic information system platform can also be integrated with other environmental monitoring systems (such as air quality monitoring systems, water quality monitoring systems, etc.) to achieve the fusion and analysis of multi-source data. At the same time, the specific implementation of the optimization plan can also combine the spatial analysis capabilities of the geographic information system platform to accurately plan and design the increase of green space area, the adjustment of building layout, etc., to ensure the effectiveness and feasibility of the optimization plan.

[0076] In a preferred embodiment of the present invention, it can be further configured as follows: the input of the U-Net network is the result of dimensionality reduction by band principal component analysis of the hyperspectral image, the output layer uses the Softmax activation function, the classification result is verified by spatial registration with the lidar point cloud data, the convolutional neural network adopts the ResNet-50 architecture, the training data includes historical thermal infrared images and corresponding ground measured temperature data, the loss function is the mean square error. By introducing band principal component analysis (PCA) to perform dimensionality reduction processing on the hyperspectral image, the redundancy and computational amount of the data are effectively reduced, and the training efficiency and classification accuracy of the U-Net network are improved. The hyperspectral image contains rich spectral information, but at the same time, there are also problems such as large data volume and high redundancy. Directly using it for classification tasks will lead to high computational complexity and poor classification effect. Through PCA dimensionality reduction, the main information in the image can be retained, noise and redundancy can be removed, so as to improve the efficiency and accuracy of the classification task. In addition, the output layer uses the Softmax activation function, which can convert the output of the U-Net network into a probability distribution form, facilitating multi-classification tasks. At the same time, verifying the classification result by spatial registration with the lidar point cloud data can further verify the accuracy and reliability of the classification result and improve the stability and credibility of the entire system;

[0077] Optimization has been carried out in terms of the architecture selection and training data of the convolutional neural network (CNN). The ResNet-50 architecture is adopted as the basic model of the CNN. This model has a deep network hierarchy and powerful feature extraction ability, and can capture the subtle features and complex patterns in hyperspectral images, thus improving the performance of classification tasks. At the same time, the training data not only includes historical thermal infrared images, but also corresponding ground measured temperature data. The fusion use of such multi-source data can further improve the generalization ability and accuracy of the model. The mean square error (MSE) is selected as the loss function, which can quantify the difference between the model prediction result and the true value and guide the optimization direction of the model. In addition, a transfer learning strategy can be introduced to fine-tune the model pre-trained in related fields to accelerate the model convergence and improve the classification accuracy. At the same time, the latest progress in the field of deep learning, such as attention mechanism, self-attention network, etc., can be combined to further enhance the feature extraction ability and classification performance of the model. In addition, in the data preprocessing stage, more image enhancement techniques, such as rotation, scaling, flipping, etc., can be adopted to increase the diversity and robustness of the training data.

[0078] In a preferred embodiment of the present invention, it can be further configured as follows: In S30, the data transmission module communicates with the ground control system using a 5G network, and the data compression algorithm is lossy compression based on wavelet transform, and the compression ratio is not less than 10:1. By adopting a 5G network as the communication means between the data transmission module and the ground control system, it can support the high-speed transmission of a large amount of data, which is crucial for environmental monitoring tasks with extremely high real-time requirements. At the same time, using the lossy compression algorithm based on wavelet transform to compress the data can not only effectively reduce the amount of transmitted data and reduce the network burden, but also achieve a high compression ratio on the premise of ensuring data quality, which can improve the overall performance and response speed of the environmental monitoring system and provide a more timely and accurate data basis for subsequent data analysis and decision support;

[0079] There is still room for further expansion and optimization in data compression algorithms. Although lossy compression algorithms based on wavelet transform can already achieve a relatively high compression ratio, in practical applications, the compression algorithms can be refined and adjusted according to the characteristics and requirements of the data. Different compression strategies and parameter settings can be adopted for different types of data (such as temperature data, humidity data, image data, etc.) to achieve a more precise compression effect. In addition, advanced machine learning algorithms can be combined to optimize the compression process intelligently, so as to improve the compression efficiency and data recovery quality. At the same time, in the application of 5G networks, other advantageous features of 5G networks, such as edge computing and network slicing, can also be further explored and utilized to achieve more flexible and efficient data transmission and processing. The implementation of these expansion and optimization measures will further enhance the performance and reliability of the environmental monitoring system, providing more powerful technical support for environmental protection and sustainable development.

[0080] The thermal infrared sensor and the hyperspectral sensor are respectively responsible for capturing surface temperature information and material composition and structure information. The data acquisition module is embedded with a radiation correction unit and a timestamp synchronization unit. The former performs radiation calibration and correction on the image data through preset calibration parameters and correction algorithms, effectively reducing the errors caused by factors such as atmospheric absorption, scattering, and instrument nonlinear response. The latter ensures the temporal consistency of the thermal infrared images and hyperspectral images. The positioning module adopts dual-frequency RTK technology to achieve centimeter-level positioning accuracy, and the sensors are fixed by a rigid bracket to ensure the spatial consistency during data acquisition.

[0081] According to the underlying surface type of the monitoring area and the preset flight line overlap rate, the flight altitude, speed, and sensor sampling time interval of the UAV are precisely planned to ensure the continuity and integrity of the data. The UAV autonomously flies along the planned route and simultaneously activates the thermal infrared sensor and the hyperspectral sensor to collect data. After the raw data is preprocessed and packed, it is transmitted to the ground control system in real time through the 5G network. The data compression algorithm uses lossy compression based on wavelet transform, and the compression ratio is not less than 10:1, which improves the data transmission efficiency and bandwidth utilization rate. After receiving the data, the ground control system decompresses and analyzes it to prepare for subsequent processing.

[0082] A convolutional neural network is used to process thermal infrared images to generate temperature distribution maps, and a U-Net network is used to process hyperspectral images to generate land cover classification maps. The radiation intensity of the thermal infrared images is converted into surface temperature through a specific formula. The input of the U-Net network is the dimensionality reduction result of the band principal component analysis of the hyperspectral images, and the output layer uses the Softmax activation function for classification. The classification result is spatially registered and verified with lidar point cloud data to ensure the accuracy of the classification result. The convolutional neural network adopts the ResNet-50 architecture, and the training data includes historical thermal infrared images and corresponding ground measured temperature data. The loss function is the mean square error, which improves the processing ability and generalization performance of the model;

[0083] Combining the temperature distribution map and the land cover classification map, the urban heat island intensity distribution, vegetation coverage, and normalized difference vegetation index are calculated. The urban heat island intensity distribution is quantitatively analyzed by accurately identifying the core area of the urban heat island and combining the building density distribution information. The urban heat island intensity gradient value UHI is calculated using a formula. Combining wind speed and humidity data, the long short-term memory network model is used to predict the urban heat island diffusion trend in the next 6 hours;

[0084] The analysis results are visually displayed through a geographic information system platform, and an optimization plan including specific suggestions such as increasing green space area, adjusting building layout, and improving material reflectivity is generated. These plans are aimed at improving the urban thermal environment and enhancing the quality of life of residents. By integrating high-precision sensors, intelligent data processing algorithms, and advanced communication technologies, efficient and accurate monitoring and optimization of the urban thermal environment are achieved, providing strong support for urban planning and management.

[0085] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0086] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made therein without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for collecting environmental monitoring data of low-altitude unmanned aerial vehicles, characterized in that, It includes the following steps: S10. UAV platform construction: Select a multi-rotor UAV as the flight platform, and carry a thermal infrared sensor, a hyperspectral sensor, a data acquisition module and a positioning module; S20. Flight route planning: According to the underlying surface type of the monitoring area and the preset route overlap rate, plan the flight altitude, speed of the UAV and the sampling time interval of the sensor; S30. Synchronous data acquisition: Control the UAV to fly according to the planned route, synchronously acquire thermal infrared image data and hyperspectral image data, and transmit the data to the ground control system in real time; S40. Deep learning data processing: Use a convolutional neural network to process the thermal infrared image to generate a temperature distribution map, and process the hyperspectral image through a U-Net network to generate a land cover classification map; S50. Thermal environment parameter analysis: Combine the temperature distribution map and the land cover classification map to calculate the heat island intensity distribution, vegetation coverage and normalized difference vegetation index; S60. Visualization and optimized output: Display the analysis results through a visualization tool and generate an optimized urban thermal environment plan.

2. The method for collecting low-altitude UAV environmental monitoring data according to claim 1, characterized in that In S10, the wavelength range of the thermal infrared sensor is 7.5 - 13.5 μm, the wavelength range of the hyperspectral sensor is 400 - 1000 nm, and the data acquisition module includes a radiation correction unit and a timestamp synchronization unit. The radiation correction unit is a component for radiometric calibration and correction of the acquired thermal infrared images and hyperspectral images. The radiation correction unit performs pixel-by-pixel radiometric calibration and correction on the acquired image data through preset calibration parameters and correction algorithms. The timestamp synchronization unit records the timestamps of the image data acquired by each sensor and synchronizes them. The positioning module is a dual-frequency RTK positioning module with a positioning accuracy of centimeter level, and the thermal infrared sensor and the hyperspectral sensor are fixed by a rigid bracket to ensure spatial consistency during data acquisition.

3. A method for collecting environmental monitoring data of a low-altitude unmanned aerial vehicle according to claim 1, characterized in that, In S20, the underlying surface types of the monitoring area include but are not limited to urban built-up areas, vegetation-covered areas, water bodies, bare areas, and agricultural land. Among them, the planned range of the flight altitude is 50 - 150 meters, the route overlap rate is not less than 70%, the sampling time interval of the thermal infrared sensor is 1 - 5 seconds, and the observation angle of the hyperspectral sensor is vertically downward ±15°.

4. A method for collecting low-altitude UAV environmental monitoring data according to claim 1, characterized in that, S30 specifically includes the following steps: S301. Start the UAV and execute the flight plan The operator starts the UAV through the ground control system and inputs the parameters of the pre-planned flight route, flight altitude, and speed. After receiving the command, the UAV starts to fly autonomously according to the planned route; S302. Synchronously start the sensor to collect data During the flight of the UAV, the ground control system synchronously starts the thermal infrared sensor and the hyperspectral sensor. The sensors start to collect data according to the preset sampling time interval and observation angle. The thermal infrared sensor is used to capture the surface temperature information, while the hyperspectral sensor is used to obtain detailed information on the surface material composition and structure; S303. Data preprocessing and packaging The original data collected by the sensor is preprocessed, including steps such as denoising and calibration, to ensure the accuracy and reliability of the data. The preprocessed data is packaged into a transmission format and transmitted to the ground control system in real time; S304. Real-time data transmission to the ground control system The drone transmits the packaged data to the ground control system in real time through wireless communication technology. After receiving the data, the ground control system will perform decompression and parsing processes for subsequent data analysis and processing.

5. A method for collecting environmental monitoring data of a low-altitude unmanned aerial vehicle according to claim 1, characterized in that, In the conversion process of the radiation intensity of the thermal infrared image to the surface temperature in S40, the temperature inversion is achieved through the following formula: T = f(λ, I) where T is the underlying surface temperature, λ is the wavelength collected by the thermal infrared sensor, I is the radiation intensity of the thermal infrared image, and f is a non-linear mapping function obtained by training with a CNN model; The generation of the land cover classification map includes the following steps: After performing band fusion and noise suppression on the hyperspectral image, it is input into the U-Net network for pixel-level classification, and the classification results including buildings, vegetation, water bodies, and bare land are output. The classification accuracy is evaluated by the following formula: Accuracy = (TP + TN) / (TP + TN + FP + FN) where TP is the true positive, TN is the true negative, FP is the false positive, and FN is the false negative.

6. The method for collecting low-altitude UAV environmental monitoring data according to claim 1, characterized in that In the calculation process of the heat island intensity distribution in S50, the highest temperature concentration area is accurately identified and extracted from the temperature distribution map and defined as the heat island core area. Subsequently, combined with the land cover classification map, especially the distribution information of the building density, a quantitative analysis of the heat island intensity is carried out. The calculation of the heat island intensity gradient value UHI adopts the following formula: Among them, T i and T j represent the temperature values of each pixel in the heat island core area and the whole monitoring area respectively; A i and A j correspond to the areas represented by these pixels respectively; represents the weighted average temperature of the heat island core area, that is, T max ; represents the weighted average temperature of the monitoring area, that is, T avg , to quantitatively evaluate the intensity of the heat island effect. The heat island intensity gradient value is further combined with wind speed and humidity data to predict the future 6-hour heat island diffusion trend through a long short-term memory network model.

7. A method for collecting environmental monitoring data of low-altitude unmanned aerial vehicles according to claim 1, characterized in that, In S60, the visualization tool is a geographic information system platform, and the optimization plan includes specific suggestions for increasing the green area, adjusting the building layout, and improving the material reflectivity.

8. A method for collecting low-altitude UAV environmental monitoring data according to claim 5, characterized in that, The input of the U-Net network is the result of dimensionality reduction by principal component analysis of the hyperspectral image bands. The output layer uses the Softmax activation function. The classification results are spatially registered and verified with the lidar point cloud data. The convolutional neural network adopts the ResNet-50 architecture. The training data includes historical thermal infrared images and corresponding ground measured temperature data, and the loss function is the mean square error.

9. A method for collecting low-altitude UAV environmental monitoring data according to claim 4, characterized in that, In S30, the data transmission module communicates with the ground control system using a 5G network. The data compression algorithm is lossy compression based on wavelet transform, and the compression ratio is not less than 10:1.

Images