Tidal flat vegetation pattern information extraction method based on unmanned aerial vehicle photogrammetry

Abstract

The invention discloses a tidal flat vegetation pattern information extraction method based on unmanned aerial vehicle photogrammetry. The method roughly comprises the following steps of (1) task planning, (2) data acquisition, (3) data preprocessing, (4) feature extraction, (5) classification and identification, (6) pattern analysis and (7) precision verification. According to the method, a multi-spectrum and RGB fused unmanned aerial vehicle image acquisition scheme is adopted, the acquisition opportunity is optimized in combination with tide forecast data, accurate extraction of multi-level information such as tidal flat vegetation types, coverage and biomass can be achieved through integration of deep learning and a traditional remote sensing classification method, and the method is high in pertinence, easy to operate and suitable for popularization.

Description

Technical Field

[0001] The invention relates to a method for extracting tidal flat vegetation pattern information based on unmanned aerial vehicle photogrammetry, and belongs to the technical field of remote sensing image processing, ecological monitoring and coastal zone resource management. Background Art

[0002] Tidal flat vegetation, as an important component of coastal ecosystems, plays an important role in maintaining biodiversity, preventing coastal erosion and carbon fixation. Traditional tidal flat vegetation survey methods mainly rely on ground field sampling and manual surveys, which are time-consuming and labor-intensive and it is difficult to obtain large-scale, high-precision vegetation spatial distribution information. Although satellite remote sensing can obtain large-scale data, its spatial resolution is limited, making it difficult to accurately identify the subtle patterns of tidal flat vegetation.

[0003] In recent years, UAV photogrammetry has been widely used in the field of environmental monitoring due to its high resolution, low cost, and maneuverability. However, existing UAV-based vegetation survey methods face many challenges when dealing with special ecological environments such as tidal flats: (1) The tidal flat environment is complex and the water-land interface changes frequently; (2) The distribution of tidal flat vegetation is highly spatially heterogeneous; (3) The periodic changes in tides make it difficult to choose the timing of data collection; and (4) Existing algorithms are unable to effectively distinguish different vegetation types with similar spectral characteristics.

[0004] Therefore, it is urgent to develop a vegetation pattern information extraction method specifically for tidal flat environments. Summary of the invention

[0005] The purpose of the present invention is to provide a method for extracting tidal flat vegetation pattern information based on unmanned aerial vehicle photogrammetry to solve the problems existing in the prior art in extracting tidal flat vegetation information.

[0006] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a method for extracting tidal flat vegetation pattern information based on unmanned aerial vehicle photogrammetry, comprising the following steps: (1) Task planning: combining tidal forecast data and the characteristics of the study area, and determining the best time for aerial photography based on the aerial photography suitability index model; (2) Data acquisition: using a multi-rotor unmanned aerial vehicle equipped with an RGB camera and a multispectral camera to obtain tidal flat image data according to a time-segmented collaborative acquisition strategy; (3) Data preprocessing: performing radiation correction, stitching, geometric correction and registration processing on the acquired image data; (4) Feature extraction: extracting a multi-dimensional feature set, including spectral features, texture features, terrain features and spatial context features; (5) Classification and recognition: using an integrated framework of deep learning and traditional classification methods to identify vegetation types; (6) Pattern analysis: calculating multi-scale vegetation pattern indicators and analyzing the spatial distribution characteristics of vegetation; (7) Accuracy verification: verifying the accuracy of the extraction results through ground measured data.

[0007] Preferably, the aerial photography suitability index model is: ASI = α·TC + β·SC + γ·WC, wherein ASI is the aerial photography suitability index, TC is the tidal coefficient, SC is the sunshine condition coefficient, WC is the wind speed condition coefficient, and α, β, and γ are weight coefficients.

[0008] Preferably, the multidimensional feature set includes a tidal flat vegetation optimization index (TVOI), which is calculated as follows: TVOI = (NIR - αRED - βGREEN) / (NIR + αRED - βGREEN) wherein α and β are parameters optimized according to the tidal flat environment.

[0009] Preferably, the integrated classification framework includes: oFirst-level classification: using an improved U-Net deep learning network for vegetation / non-vegetation classification; oSecond level classification: Random forest algorithm is used to classify vegetation types; oEnsemble decision: Use Bayesian Model Averaging method to fuse multiple model results.

[0010] Preferably, the multi-scale vegetation pattern indicators include landscape level indicators, type level indicators and patch level indicators.

[0011] Beneficial effects achieved by the present invention: The present invention adopts a multi-spectral and RGB fusion UAV image acquisition solution, combines it with tidal forecast data to optimize the acquisition timing, and integrates deep learning with traditional remote sensing classification methods to achieve accurate extraction of multi-level information such as tidal flat vegetation type, coverage, biomass, etc. It is highly targeted, easy to operate, and suitable for promotion. DETAILED DESCRIPTION

[0012] The present invention will be further described below in conjunction with specific embodiments. Example

[0013] 1. Mission Planning 1.1 Tidal timing selection Based on the tidal forecast data of the study area, low tides during spring tides were selected for aerial photography to maximize exposure of tidal flat vegetation: The aerial photography suitability index model is ASI = α·TC + β·SC + γ·WC Among them, ASI is the aerial photography suitability index, TC is the tidal coefficient, SC is the sunshine condition coefficient, WC is the wind speed condition coefficient, and α, β, and γ are weight coefficients; 1.2 Flight parameter setting According to the area of ​​the study and the required spatial resolution, set the UAV flight altitude, speed, route interval and other parameters; for typical tidal flat environments, the following parameters are recommended: Flight altitude: 80-120 meters Route overlap: 75% forward overlap, 65% lateral overlap Flight speed: 5-8 m / s 2. Data Collection 2.1 Equipment selection Multi-rotor drone is used as the carrying platform, and the following equipment is loaded at the same time: RGB camera: resolution ≥ 20 million pixels Multispectral camera: including near infrared and red edge bands, with a resolution of ≥ 2 million pixels RTK positioning module: positioning accuracy better than 2 cm 2.2 Collaborative acquisition strategy: The time-divided collaborative acquisition strategy of "multispectral first, then RGB" is adopted to ensure that the two types of data are acquired under similar lighting conditions; 3. Data Preprocessing 3.1 Radiation Correction Perform radiation correction on multispectral images and convert them into reflectance data; a secondary correction model based on ground reflectors and atmospheric parameters is used; 3.2 Image stitching and geometric correction Use structured light motion (SfM) technology to stitch images and generate orthophotos and digital surface models (DSM); 3.3 Registration The feature point matching algorithm is used to achieve accurate registration of RGB and multispectral images, with a registration accuracy better than 1 pixel; 4. Feature Extraction 4.1 Spectral feature calculation Common vegetation indices include: Normalized Difference Vegetation Index (NDVI) Enhanced Vegetation Index (EVI) Modified Soil Adjusted Vegetation Index (MSAVI) The tidal flat vegetation optimization index (TVOI) proposed by the present invention is: TVOI = (NIR - αRED - βGREEN) / (NIR + αRED - βGREEN) Among them, α and β are parameters optimized according to the tidal flat environment; 4.2 Texture features Texture features are extracted based on the gray-level co-occurrence matrix (GLCM), including homogeneity, contrast, entropy, etc. 4.3 Terrain characteristics The terrain characteristics such as slope, aspect, curvature, etc. were extracted based on DSM; 4.4 Spatial context features A multi-scale segmentation algorithm is used to extract spatial context information; 5. Classification and Identification 5.1 Sample Construction A stratified random sampling strategy was used to construct the training sample set and the validation sample set; 5.2 Integrated classification framework uses the integrated framework of "deep learning + traditional classifier": First-level classification: using an improved U-Net deep learning network for vegetation / non-vegetation classification Second level classification: Use random forest algorithm to classify vegetation types Integrated decision-making: using the Bayesian Model Averaging method to fuse the results of multiple models 5.3 Post-processing uses object-oriented morphological processing and spatial rule constraints to eliminate "salt and pepper noise" and improve the spatial continuity of classification results; 6. Pattern Analysis 6.1 Calculation of pattern indicators The following pattern indicators are calculated: Landscape level: Shannon diversity index, aggregation index, etc. Type level: plaque density, average plaque size, plaque shape index, etc. Patch level: area, perimeter, shape index, etc. 6.2 Analysis of spatiotemporal changes Establish a time series of pattern indicators and analyze the spatiotemporal changes of vegetation patterns; VII. Accuracy Verification 7.1 Sample Verification The confusion matrix is ​​calculated through ground-measured sample data to evaluate the Overall Accuracy, Kappa coefficient, Producer's Accuracy and User's Accuracy; 7.2 Area Verification The area consistency index was calculated by comparing the vegetation distribution map obtained through field survey with the extracted results.

[0014] It should be understood that the above description of the embodiments is relatively detailed and cannot be regarded as limiting the scope of patent protection of the present invention. For ordinary technicians in the technical field to which the present invention belongs, any modifications and changes made to the present invention without departing from the concept of the present invention should be regarded as belonging to the scope of protection of the present invention.

Claims

1. A method for extracting tidal flat vegetation pattern information based on drone photogrammetry, characterized in that: The following steps are involved: (1) Mission planning: Based on the tide forecast data and the characteristics of the study area, the best time for aerial photography is determined based on the aerial photography suitability index model; (2) Data acquisition: A multi-rotor drone equipped with an RGB camera and a multispectral camera is used to obtain tidal flat image data according to a time-segmented collaborative acquisition strategy; (3) Data preprocessing: Radiometric correction, stitching, geometric correction, and registration are performed on the acquired image data; (4) Feature extraction: Extract multidimensional feature sets, including spectral features, texture features, terrain features, and spatial context features; (5) Classification and recognition: Vegetation type recognition is performed using an integrated framework of deep learning and traditional classification methods; (6) Pattern analysis: Calculate multi-scale vegetation pattern indicators and analyze the spatial distribution characteristics of vegetation; (7) Accuracy verification: Verify the accuracy of the extracted results through ground-based measured data.

2. The method according to claim 1, characterized in that The aerial photography suitability index model is: ASI = α·TC + β·SC + γ·WC, where ASI is the aerial photography suitability index, TC is the tidal coefficient, SC is the sunshine condition coefficient, WC is the wind speed condition coefficient, and α, β, and γ are weight coefficients.

3. The method according to claim 1, characterized in that The multidimensional feature set includes the tidal flat vegetation optimization index TVOI, and the calculation formula is: TVOI = (NIR - αRED - βGREEN) / (NIR + αRED - βGREEN) wherein α and β are parameters optimized according to the tidal flat environment.

4. The method according to claim 1, characterized in that: The integrated classification framework includes: oFirst-level classification: using an improved U-Net deep learning network for vegetation / non-vegetation classification; oSecond level classification: Random forest algorithm is used to classify vegetation types; oEnsemble decision: Use Bayesian Model Averaging method to fuse multiple model results.

5. The method according to claim 1, characterized in that: The multi-scale vegetation pattern indicators include landscape level indicators, type level indicators and patch level indicators.