Forest canopy temperature accurate extraction method based on unmanned aerial vehicle thermal infrared remote sensing data

Abstract

The present application belongs to the technical field of forestry remote sensing, unmanned aerial vehicle thermal infrared remote sensing and land surface temperature retrieval, and specifically relates to a method for precisely extracting forest canopy temperature based on unmanned aerial vehicle thermal infrared remote sensing data, which eliminates atmospheric radiation interference through atmospheric correction, optimizes the radiation calibration coefficient in combination with a ground black body calibration source, and improves the radiation calibration precision; the method realizes precise separation of the canopy and the background by using the greenness vegetation index or the normalized vegetation index in combination with morphological filtering, and eliminates background interference; the method constructs a temperature retrieval model based on the Planck radiation law, improves the model universality in combination with the specific radiation rate values of different forest types, and verifies the precision through multiple ground verification points to ensure that the extraction result is reliable. The present application solves the problems of existing methods, such as difficult-to-eliminate interference, low calibration precision and poor universality, realizes rapid and precise extraction of canopy temperature of different types of forests, and can be widely applied to scenes such as forest water stress diagnosis, pest and disease monitoring and forest health condition assessment.

Description

Technical Field

[0001] This invention belongs to the fields of forestry remote sensing, UAV thermal infrared remote sensing, and land surface temperature inversion technology. Specifically, it is a method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data. It is suitable for rapid and accurate acquisition of canopy temperature in different types of forests and can be applied to scenarios such as forest water stress diagnosis, pest and disease monitoring, and ecological environment assessment. Background Technology

[0002] Forest canopy temperature is a core parameter characterizing water stress, forest health, and fire risk in forest ecosystems, and is widely used in forest carbon sequestration estimation, drought monitoring, pest and disease early warning, and forest fire prevention. Traditional canopy temperature acquisition relies on ground-based observations, which are inefficient, have limited coverage, and poor representativeness. Satellite thermal infrared remote sensing has low spatial resolution and is easily affected by clouds and atmospheric interference, making it difficult to meet the needs of refined forestry monitoring. Unmanned aerial vehicle (UAV) thermal infrared remote sensing offers advantages such as high spatiotemporal resolution, maneuverability, and low cost, compensating for the shortcomings of ground observations and satellite remote sensing (low spatial resolution and susceptibility to cloud cover). However, current forest canopy temperature extraction based on UAV thermal infrared remote sensing data still faces several technical bottlenecks, resulting in extraction accuracy that fails to meet practical application requirements. Data interference is a prominent issue: UAV thermal infrared images are easily affected by atmospheric radiation, surface background (soil, rocks, understory vegetation), solar radiation reflection, and other factors. In particular, the forest canopy has gaps and overlaps, and the background information is mixed with the canopy information, making it difficult to separate them effectively, which directly leads to a large error in canopy temperature extraction. Insufficient radiation calibration accuracy: Existing methods mostly use the sensor's factory calibration parameters and do not combine flight scenarios (such as atmospheric humidity, temperature, and altitude) for real-time correction, resulting in low accuracy in converting thermal infrared radiation values ​​to temperature values; Canopy extraction and temperature inversion are disconnected: Most methods simply extract the canopy region by threshold segmentation without combining forest canopy structural characteristics (such as canopy coverage and leaf area index) to optimize the inversion model. This makes it difficult to adapt to the canopy temperature extraction needs of different types of forests and results in poor versatility. Spatial consistency optimization is typically lacking: temperature drift is significant after stitching together individual images, making it impossible to form a continuous and reliable canopy temperature field. Furthermore, the lack of specific purification and correction for forest vertical structure, canopy porosity, and understory background makes it difficult to achieve high-precision, operational-grade canopy temperature extraction. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0004] The technical solution adopted by this invention to solve its technical problem is: the method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data, as described in this invention, includes the following steps: S1. UAV thermal infrared data acquisition and preprocessing: including flight planning, data acquisition and preprocessing, wherein the preprocessing includes image stitching, geometric correction and noise reduction; S2. Thermal infrared radiometric calibration optimization: This includes initial radiometric calibration, atmospheric correction, and radiometric calibration optimization. The atmospheric correction uses an atmospheric radiative transfer model, and the radiometric calibration optimization introduces a ground-based blackbody calibration source to correct the calibration coefficients. S3. Separation of forest canopy from background: including preliminary extraction of canopy area, optimization of canopy area and extraction of canopy radiance. The preliminary extraction of canopy area adopts greenness vegetation index or normalized vegetation index combined with threshold segmentation method. The optimization of canopy area adopts morphological filtering algorithm. S4. Forest canopy temperature inversion: including emissivity determination, temperature inversion model construction, and canopy temperature calculation, wherein the temperature inversion model is constructed based on Planck's radiation law; S5. Accuracy Verification: The measured temperature at the ground temperature verification point is compared with the extracted results to calculate the mean absolute error, root mean square error, and coefficient of determination, and the final result is output.

[0005] As a further technical solution of the present invention: in S1, the flight planning includes: setting the UAV flight parameters according to the forest type and terrain features of the monitoring area; The flight parameters include flight altitude, flight speed, forward overlap, and lateral overlap; and several ground control points and temperature verification points are set up within the monitoring area.

[0006] As a further technical solution of the present invention: In S1, the data acquisition includes: using a UAV equipped with a thermal infrared sensor to fly according to preset flight parameters, simultaneously collecting thermal infrared image data and UAV position and attitude data, collecting atmospheric parameters during the flight period through a ground meteorological station; and deploying a ground blackbody calibration source and at least two ground temperature correction targets with different emissivity and known temperatures.

[0007] As a further technical solution of the present invention: In S1, the geometric correction adopts ground control points combined with UAV position and attitude data, and realizes image spatial position correction through affine transformation.

[0008] As a further technical solution of the present invention: In step S2, the initial radiometric calibration is: according to the factory calibration parameters of the thermal infrared sensor, the gray value of the thermal infrared image is converted into the radiance value to obtain the initial radiance image. The radiometric calibration optimization involves converting the measured temperature of the blackbody calibration source during flight into a radiance value, comparing it with the radiance value of the corresponding area in the UAV thermal infrared image, correcting the radiometric calibration coefficient, and obtaining an optimized surface radiance image.

[0009] As a further technical solution of the present invention: In S3, the preliminary extraction of the canopy region includes: calculating the greenness vegetation index or normalized vegetation index based on the visible light image or multispectral image synchronously collected by the UAV, and preliminarily segmenting the forest canopy region and the background region by combining the threshold segmentation method to obtain a preliminary canopy mask.

[0010] As a further technical solution of the present invention: In S3, the canopy region optimization includes: combining the structural characteristics of the forest canopy, using the dilation-erosion algorithm in morphological filtering to optimize the preliminary canopy mask, removing small noise in the background area and background interference in the canopy gaps, while filling the voids in the canopy region to obtain an accurate canopy mask.

[0011] As a further technical solution of the present invention: the determination of emissivity in step S4 includes: determining the emissivity of the forest canopy based on the forest type of the monitoring area and in conjunction with the vegetation emissivity database; Among them, coniferous forests and broad-leaved forests are assigned preset specific emissivity values, and the specific emissivity of mixed forests is calculated by weighting the proportions of coniferous forests and broad-leaved forests.

[0012] As a further technical solution of the present invention: In S5, the calculation method of mean absolute error, root mean square error and coefficient of determination in the accuracy verification is as follows: the measured temperature of the ground temperature verification point is compared with the temperature value of the corresponding position in the canopy temperature extraction result point by point, and the mean absolute error, root mean square error and coefficient of determination are calculated respectively.

[0013] As a further technical solution of the present invention: In S5, the final result output includes: cropping and stitching the optimized canopy temperature image, and outputting the forest canopy temperature distribution map and statistical data of the monitoring area, wherein the statistical data includes average temperature, maximum temperature, minimum temperature and temperature standard deviation.

[0014] The beneficial effects of this invention are as follows: This invention constructs a two-stage canopy and background separation process by "coarse extraction of vegetation indices and morphological filtering optimization guided by canopy structure features". It can effectively remove background thermal radiation interference such as soil, rocks, understory vegetation, and litter, while retaining effective temperature information in the canopy gaps. This solves the problem of low accuracy caused by the difficulty in eliminating interference in existing methods, and the extraction accuracy is significantly better than existing methods. This invention, based on the sensor's factory calibration, introduces a ground-based blackbody calibration source and real-time atmospheric parameters (temperature, humidity, and air pressure) for joint radiometric calibration. By converting the measured blackbody temperature into a radiance value and comparing it with the corresponding area in the image, the calibration coefficient is corrected. This effectively eliminates radiometric errors caused by environmental changes during flight (such as changes in atmospheric window transmittance and sensor temperature drift), solves the calibration error problem caused by existing methods that rely solely on factory calibration without incorporating flight scenario correction, and improves the accuracy of converting radiance values ​​to temperature values. This invention fully considers the structural heterogeneity of forest canopies and sets different canopy masking optimization strategies (such as adjusting the size of expansion-corrosion structural elements) and emissivity value strategies (area-weighted calculation for mixed forests) for coniferous forests, broad-leaved forests, and mixed forests. This type-adaptive processing method enables this invention to be applied to canopy temperature extraction in various forest ecosystems such as temperate, subtropical, and tropical regions, overcoming the limitations of existing methods that are "one model, one scenario," and possessing the ability to generalize across regions, seasons, and forest types. This invention allows for a streamlined software process, from image input, radiometric calibration, canopy separation, temperature inversion to accuracy verification, supporting batch processing and automatic parameter adjustment. Ground control and verification points are deployed during the flight planning phase, eliminating the need for manual image-by-image intervention in subsequent processing, significantly reducing the operational threshold and human error. Combined with the large-area coverage capability of the UAV platform (capable of collecting data from several square kilometers of forest per flight), this invention can complete the entire process from data acquisition to canopy temperature distribution map output in a short time, making it suitable for routine monitoring tasks in forestry departments. This invention establishes a closed-loop mechanism of "accuracy verification - parameter feedback - model optimization". Using measured data from multiple ground temperature verification points, statistical indicators such as MAE, RMSE, and R² are calculated. If the preset accuracy is not met, the system automatically adjusts the radiometric calibration coefficient and emissivity, repeating the inversion process until the target is achieved. This mechanism ensures that each monitoring mission outputs canopy temperature products that meet quality requirements, making it particularly suitable for long-term, multi-temporal, and cross-regional forest monitoring networks. This invention employs improved Gaussian filtering and ground control point geometric correction during the image stitching stage, and maintains spatial resolution consistency in subsequent canopy purification and temperature inversion, effectively suppressing the "temperature drift" phenomenon commonly seen after stitching single images. The final output canopy temperature distribution map exhibits continuous, smooth, and seamless spatial characteristics, clearly presenting temperature gradient changes within the canopy and at the forest edge, providing high-quality spatial input for forest fire risk zoning and water stress diagnosis. Attached Figure Description

[0015] The invention will now be further described with reference to the accompanying drawings.

[0016] Figure 1 This is a flowchart of a method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data, according to the present invention. Detailed Implementation

[0017] To make the technical means, main features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0018] like Figure 1 As shown, this embodiment of the invention provides a method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data, including the following steps: S1. UAV thermal infrared data acquisition and preprocessing: S11. Flight Planning: Based on the forest type (coniferous forest, broad-leaved forest, mixed forest) and terrain features of the monitoring area, set the UAV flight parameters: flight altitude of 100 meters, flight speed of 3-7 m / s, forward overlap of 85%, and lateral overlap of 75%. At the same time, set up several ground control points and temperature verification points in the monitoring area. The ground control points are used for image geometric correction, and the temperature verification points are used for subsequent accuracy verification. S12. Data Acquisition: A drone equipped with a thermal infrared sensor is used to fly according to preset flight parameters, and thermal infrared image data and drone position and attitude data (GPS / IMU data) are collected simultaneously. At the same time, atmospheric parameters such as temperature, humidity and air pressure during the flight are collected through ground weather stations. Ground blackbodies are set up in open areas at a height of about 0.20 meters. Ground temperature correction targets are set up in homogeneous grassland and water bodies for subsequent radiometric calibration. S13. Preprocessing: The acquired thermal infrared images are preprocessed, including image stitching, geometric correction, and denoising. Geometric correction uses ground control points combined with UAV position and attitude data to achieve spatial position correction of the images through affine transformation. Denoising uses an improved Gaussian filtering algorithm (filter window 5×5, standard deviation 0.8) to preserve canopy details while eliminating image noise interference.

[0019] S2. Thermal infrared radiation calibration optimization: S21. Initial radiometric calibration: Based on the factory calibration parameters of the thermal infrared sensor, the grayscale values ​​of the thermal infrared image are converted into radiance values ​​to obtain the initial radiance image; S22. Atmospheric Correction: Input the atmospheric parameters (temperature, humidity, air pressure) collected by the ground meteorological station during the flight period in step S12 into the atmospheric radiative transfer model to calculate the atmospheric upward radiance, atmospheric downward radiance, and atmospheric transmittance at the flight altitude; based on the calculated atmospheric parameters, perform atmospheric correction on the initial radiance image obtained in step S21 to eliminate the influence of atmospheric path radiation on the thermal infrared signal, and obtain the atmospherically corrected surface radiance image. S23. Radiometric calibration optimization: A ground blackbody calibration source is introduced to convert the measured temperature of the blackbody during flight into a radiance value, which is then compared with the radiance value of the corresponding area in the UAV thermal infrared image. Since the sensor has good linearity of response, only the offset is corrected, the model is improved, the radiometric calibration accuracy is enhanced, and an optimized surface radiance image is obtained.

[0020] S3. Separation of forest canopy from background: S31. Preliminary Canopy Region Extraction: The greenness vegetation index ((2*G - R - B) / (2*G + R + B)) or NDVI vegetation index ((NIR - R) / (NIR + R)) is used, where R is the red band value, G is the green band value, B is the blue band value, and NIR is the near-infrared band value. Combined with threshold segmentation, the visible light or multispectral images simultaneously acquired by the UAV are processed to preliminarily segment the forest canopy region and background region (soil, rocks, and understory vegetation), obtaining a preliminary canopy mask. S32. Canopy Area Optimization: Combining forest canopy structure characteristics (canopy coverage, leaf area index), the morphological filtering (dilation-erosion) algorithm is used to optimize the preliminary canopy mask. By using closing operations (dilation before erosion) and opening operations (erosion before dilation), or by directly using targeted combinations, the fine noise in the background area and background interference in the canopy gaps are removed to obtain an accurate canopy mask. S33. Canopy Radiance Extraction: The optimized canopy mask is superimposed on the optimized surface radiance image obtained in step S23, the radiance values ​​corresponding to the canopy area are extracted, the radiance data of the background area are removed, and the forest canopy radiance image is obtained.

[0021] S4. Forest canopy temperature inversion: S41. Determination of specific emissivity: Based on the forest type of the monitoring area and in conjunction with the vegetation specific emissivity database, the specific emissivity of the forest canopy is determined: evergreen coniferous forest: 0.985, evergreen broad-leaved forest: 0.980; for mixed forests, the specific emissivity is calculated by weighting the coverage ratio of coniferous forest and broad-leaved forest. S42. Temperature Inversion Model Construction: Based on Planck's radiation law, combined with canopy emissivity and optimized canopy radiance values, a forest canopy temperature inversion model is constructed.

[0022] L _emitted _ only = ε*B(λ,T) Where L _emitted _ only The radiance of an actual ground object, ε is the emissivity, B(λ,T) is the radiance of a blackbody at the same temperature, T is the temperature, and λ is the wavelength.

[0023] S43. Canopy Temperature Calculation: Substitute the canopy radiance value obtained in step S33 and the canopy emissivity determined in step S41 into the above inversion model. Based on the inverse operation of Planck's radiation law, the canopy radiance L is calculated. _emitted _ only The blackbody radiance B(λ, T) = L is obtained by calculating the emissivity ε. _emitted _ only / ε, and then obtain the corresponding temperature T (unit: K) through Planck function inversion, and convert it to degrees Celsius (°C) to obtain the preliminary canopy temperature extraction results.

[0024] T_celsius = T - 273.15 Where T_celsius is the canopy temperature (unit: °C), and T is the temperature obtained by inverse operation of Planck's law (unit: K).

[0025] S5. Accuracy Verification: S51. Accuracy Verification: The measured temperature at the ground temperature verification point (synchronously observed by an infrared thermometer) is compared with the temperature value at the corresponding location in the canopy temperature extraction results. The mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) are calculated to verify the extraction accuracy. S52. Final output: The optimized canopy temperature image is cropped and stitched together to output the forest canopy temperature distribution map and statistical data (average temperature, maximum temperature, minimum temperature, temperature standard deviation) of the monitoring area, thus completing the accurate extraction of forest canopy temperature.

[0026] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data, characterized in that, Includes the following steps: S1. UAV thermal infrared data acquisition and preprocessing: including flight planning, data acquisition and preprocessing, wherein the preprocessing includes image stitching, geometric correction and noise reduction; S2. Thermal infrared radiometric calibration optimization: This includes initial radiometric calibration, atmospheric correction, and radiometric calibration optimization. The atmospheric correction uses an atmospheric radiative transfer model, and the radiometric calibration optimization introduces a ground-based blackbody calibration source to correct the calibration coefficients. S3. Separation of forest canopy from background: including preliminary extraction of canopy area, optimization of canopy area and extraction of canopy radiance. The preliminary extraction of canopy area adopts greenness vegetation index or normalized vegetation index combined with threshold segmentation method. The optimization of canopy area adopts morphological filtering algorithm. S4. Forest canopy temperature inversion: including emissivity determination, temperature inversion model construction, and canopy temperature calculation, wherein the temperature inversion model is constructed based on Planck's radiation law; S5. Accuracy Verification: The measured temperature at the ground temperature verification point is compared with the extracted results to calculate the mean absolute error, root mean square error, and coefficient of determination, and the final result is output.

2. The method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data according to claim 1, characterized in that, In S1, the flight planning includes: setting UAV flight parameters based on the forest type and terrain features of the monitoring area; The flight parameters include flight altitude, flight speed, forward overlap, and lateral overlap; and several ground control points and temperature verification points are set up within the monitoring area.

3. The method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data according to claim 1, characterized in that, In S1, the data acquisition includes: using a UAV equipped with a thermal infrared sensor to fly according to preset flight parameters, simultaneously collecting thermal infrared image data and UAV position and attitude data, collecting atmospheric parameters during the flight period through a ground meteorological station, and deploying a ground blackbody calibration source and at least two ground temperature correction targets with different emissivity and known temperatures.

4. The method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data according to claim 1, characterized in that, In S1, the geometric correction uses ground control points combined with UAV position and attitude data to achieve spatial position correction of the image through affine transformation.

5. The method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data according to claim 1, characterized in that, In S2, the initial radiometric calibration is as follows: based on the factory calibration parameters of the thermal infrared sensor, the grayscale value of the thermal infrared image is converted into a radiance value to obtain an initial radiance image. The radiometric calibration optimization involves converting the measured temperature of the blackbody calibration source during flight into a radiance value, comparing it with the radiance value of the corresponding area in the UAV thermal infrared image, correcting the radiometric calibration coefficient, and obtaining an optimized surface radiance image.

6. The method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data according to claim 1, characterized in that, In S3, the preliminary extraction of the canopy region includes: calculating the greenness vegetation index or normalized vegetation index based on visible light images or multispectral images synchronously collected by UAVs, and preliminarily segmenting the forest canopy region and background region using a threshold segmentation method to obtain a preliminary canopy mask.

7. The method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data according to claim 1, characterized in that, In S3, the canopy region optimization includes: combining the structural characteristics of the forest canopy, using the dilation-erosion algorithm in morphological filtering to optimize the preliminary canopy mask, removing small noise in the background area and background interference in the canopy gaps, while filling the voids in the canopy region to obtain an accurate canopy mask.

8. The method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data according to claim 1, characterized in that, The determination of emissivity described in S4 includes: determining the emissivity of the forest canopy based on the forest type of the monitoring area and in conjunction with a vegetation emissivity database; Among them, coniferous forests and broad-leaved forests are assigned preset specific emissivity values, and the specific emissivity of mixed forests is calculated by weighting the coverage ratio of coniferous forests and broad-leaved forests.

9. The method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data according to claim 1, characterized in that, In S5, the calculation method for the mean absolute error, root mean square error, and coefficient of determination in the accuracy verification is as follows: the measured temperature of the ground temperature verification point is compared with the temperature value of the corresponding position in the canopy temperature extraction result point by point, and the mean absolute error, root mean square error, and coefficient of determination are calculated respectively.

10. The method for accurately extracting forest canopy temperature based on UAV thermal infrared remote sensing data according to claim 1, characterized in that, In S5, the final output includes: cropping and stitching the optimized canopy temperature image to output a forest canopy temperature distribution map and statistical data of the monitoring area, including average temperature, maximum temperature, minimum temperature and temperature standard deviation.

Images