Remote sensing identification method for urban pollen highly-sensitized tree species

Abstract

The invention relates to the field of plant remote sensing recognition, in particular to a remote sensing recognition method for urban pollen highly-sensitized tree species. Comprising the following steps: vectorizing plaque sample data; on a remote sensing big data processing and presenting platform, a remote sensing data set is called, local pollen highly-sensitized tree species plaque samples, arbor, shrub and grass vegetation and boundary vector data are uploaded, and identification, classification and drawing of the pollen highly-sensitized tree species are achieved in a code writing and running mode; analyzing the spatial distribution condition of the remote sensing identification classification result of the highly-sensitized pollen tree species, counting the area of each type of sensitized tree species, obtaining the spatial distribution characteristics of the highly-sensitized tree species in the city, and further evaluating the type, the distribution position and the area of the highly-sensitized tree species in the city; based on a pollen highly-sensitized tree species space identification technology of full-wave band information and six types of vegetation indexes, information such as types, distribution positions, areas, characteristics and the like of urban highly-sensitized tree species is accurately judged, and a new space analysis thought and method from tree species is provided for related research of urban pollen sensitization problems.

Description

Technical Field

[0001] The invention relates to the field of plant remote sensing identification, in particular to a remote sensing identification method for urban pollen highly allergenic tree species. Background Art

[0002] Urban green space plants are an essential part of urban ecological environment construction, and have multiple positive impacts on urban landscape value, ecological functions, residents' lives and public health. However, airborne allergenic plant pollen can easily cause human allergies, ranging from sneezing, coughing, runny nose, and skin rashes to asthma, tracheitis, and cor pulmonale, which can even threaten life. In the process of urban development, the level of green space construction has been continuously improved. For example, the green space area has increased and the plant diversity has increased, resulting in an increasing number of pollen species and quantities. The pollen of allergenic plants has also increased rapidly, and the pollen allergy rate of urban residents has continued to rise. The problem of pollen allergy needs to be solved urgently. As the main body of urban green space, woody plants are an important part of urban vegetation. Among them, the pollen yield and pollen allergenic potential of pollen allergenic woody plants are outstanding. The improvement of urban greening rate and green space construction level also means that the types and number of pollen allergenic tree species are gradually increasing, and the distribution of tree species in urban green space space shows a trend of strong aggregation and high coverage. In addition, residents often come into contact with pollen-allergenic tree species in their activities and suffer from pollen allergies. Therefore, the large-scale distribution of highly pollen-allergenic tree species brings the risk of allergies to urban residents due to high concentrations of allergenic pollen invading the human body.

[0003] The emergence of plant remote sensing identification technology, with its powerful advantages such as wide data coverage, efficient processing flow, and low difficulty in acquisition, has effectively alleviated the negative impact of traditional plant species surveys, such as low efficiency, limited implementation scope in time and space, high cost, and certain difficulties. In the process of the development of plant remote sensing identification and classification technology, the types of satellite remote sensing data with high spectral resolution and high temporal and spatial resolution are constantly increasing, and the application exploration of multi-source remote sensing data with multi-scale and long time series is constantly improving, which makes the support, update and accessibility of remote sensing technology for plant species identification results continue to improve, and promote the sustainable development of plant distribution research. One of the hot development trends in the existing tree species remote sensing classification and identification algorithm research is to rely on the strengthening and updating of big data computing power and algorithms to explore ways to greatly improve the accuracy and speed of remote sensing classification, and to continue to try to apply and refine monitoring and analysis in large-scale spatial and temporal ranges such as regions and the world. This research trend has gradually spread to the research field of pollen allergenic tree species, providing new ideas and methods for the identification and monitoring of allergenic tree species. It is of great significance for judging whether there is a pollen allergy problem and determining its source on a large scale, as well as assessing the risk of pollen allergy. Summary of the invention

[0004] The objective of the present invention is to obtain the specific types, distribution locations and areas of highly pollen - sensitizing tree species from the perspective of remotely sensing and identifying plants, thereby establishing new ideas and methods for spatial analysis of the pollen sensitization risk of highly pollen - sensitizing tree species, so as to better solve the problems such as unknown sources of urban pollen sensitization problems, unknown harm levels, and no basis for prevention and early warning, meet the identification and classification requirements of highly pollen - sensitizing tree species at specific spatio - temporal scales and with high precision, and further become the quantitative analysis basis for scientifically and reasonably reducing the pollen sensitization risk of urban tree species.

[0005] The technical solution of the present invention is as follows:

[0006] A method for remotely sensing and identifying highly pollen - sensitizing tree species in cities, comprising the following steps:

[0007] Step 1: Vectorize patch sample data. In the ArcGIS software, convert the research samples of highly pollen - sensitizing tree species into.shp vector format, and assign values to the sample category attributes to distinguish tree species, obtaining a sample vector data set.

[0008] Step 2: On the Google Earth Engine (GEE) remote - sensing big - data processing and presentation platform, call the Sentinel - 2 remote - sensing data set of the target time and area, and upload local patch samples of highly pollen - sensitizing tree species, arbors, shrubs and grasses vegetation, and boundary vector data. The identification, classification and mapping of highly pollen - sensitizing tree species are realized by writing and running codes. The main process is as follows:

[0009] Step 2.1: Pre - process remote - sensing image data.

[0010] Step 2.1.1: Screening. Select remote - sensing images according to the target time range, perform atmospheric correction processing to exclude the interference of atmospheric scattering and absorption, and factors such as clouds, aerosols, and illumination on image interpretation, obtaining the true spectral information of ground objects reflected to the sensor.

[0011] Step 2.1.2: Cloud removal. Remove clouds and noise from the images. After obtaining images with cloud cover below a certain percentage, create and synthesize remote - sensing satellite composite images.

[0012] Step 2.1.3: Fusion. Traverse the median values of all images at the same pixel, and calculate to obtain a multi - spectral fusion image presented in the form of pixel medians.

[0013] Step 2.1.4: Mosaic and clipping. Mosaic and splice multiple adjacent images together to obtain a seamless, continuous, large - scale remote - sensing image covering the entire area. According to the vector boundary of the target area, accurately clip the range of the mosaicked image to obtain the corresponding image data. Using the visual interpretation data of arbors, shrubs and grasses vegetation classification as a mask, obtain the vegetation remote - sensing image data of the target area, and uniformly resample all data to the same spatial resolution.

[0014] Step 2.2: Remote sensing feature extraction and combination. Obtain the multi-spectral band reflectance information of the remote sensing image and calculate the vegetation index. Use the band information and the vegetation index as the classification feature values of highly pollen-sensitive tree species, and select the reflectance values of all bands of the remote sensing image data and the vegetation index to set up a remote sensing feature set for classification.

[0015] Step 2.3: Construct a random forest classification model.

[0016] Step 2.3.1: Random sampling. Conduct random sampling with replacement on the initial training data set, obtaining data from the original data that may be repeated or may not be selected at all, to establish a new data training set, and by taking the average of multiple random subsets, reduce the risk of overfitting and enhance the generalization ability of the model to new data.

[0017] Step 2.3.2: Randomly select feature attribute parameters. On the premise that not all features are used for model training, randomly select a feature subset from all feature attributes, and let each decision tree select an optimal feature attribute from the feature subset as the division node. Randomly select the features of each node, only considering the random feature subset and not all feature attributes.

[0018] Step 2.3.3: Majority voting. Output the classification labels of each decision tree respectively, conduct a majority vote on all labels to obtain the final classification result, ensuring that the final prediction result is the integration of the results of all decision trees.

[0019] Step 2.4: Training and verification of the identification and classification of highly pollen-sensitive tree species. Divide the sample data into two categories, training and verification, and input them into the random forest classification model to obtain the classification operation result.

[0020] Step 2.5: Accuracy evaluation and spatial distribution mapping of the remote sensing identification and classification of highly pollen-sensitive tree species. Use a confusion matrix to measure the accuracy of the classification model. First, separately count the number of values of the misclassified and correctly classified results in the classification model, then display them all in a table, and finally obtain the matrix form result table of rows and columns. Determine the correct or incorrect classification situation of a model in the predicted and true categories through the confusion matrix, and whether the classification model confuses two categories.

[0021] Step 3: Analyze the spatial distribution of the remote sensing identification and classification results of highly pollen-sensitive tree species, and count the areas of various highly sensitive tree species in the ArcGIS software to obtain the spatial distribution characteristics of highly pollen-sensitive tree species in the city. The main process is as follows:

[0022] Step 3.1: Spatial distribution analysis of highly pollen-sensitive tree species. Analyze the areas, administrative regions, green space types, specific locations, etc. where highly pollen-sensitive tree species are densely distributed.

[0023] Step 3.2: Statistical calculation of the spatial distribution area of highly pollen-allergenic tree species. Using the raster area calculation function of ArcGIS software, calculate the distribution area of each tree species, and further calculate the number of species, distribution area, and proportion of highly pollen-allergenic tree species in a specific area.

[0024] The beneficial effects of the present invention are as follows:

[0025] 1. The present invention proposes a spatial identification technology for highly pollen-allergenic tree species based on full-band information and six types of vegetation indices, accurately determining information such as the types, distribution locations, areas, and characteristics of highly allergenic tree species in the city, achieving fine identification of highly pollen-allergenic tree species at a spatial resolution of 10 m, and providing a new spatial analysis idea and method for research related to pollen allergy problems in the city from the perspective of tree species.

[0026] 2. The present invention comprehensively utilizes traditional remote sensing software and emerging remote sensing processing code platforms, and uses the random forest in supervised classification, making the technical solution steps clear, operation efficient, ensuring accuracy, and improving the spatial identification efficiency and accuracy of highly pollen-allergenic tree species. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is the technical flow chart of the present invention;

[0028] Figure 2 is the spatial mapping of the classification and identification results of highly pollen-allergenic tree species in the urban area of Beijing based on full-band and vegetation index eigenvalues of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0029] The following further describes the specific embodiments of the present invention with reference to the drawings:

[0030] Example 1

[0031] As Figure 1 shown, Step 1: In the ArcGIS software, vectorize the field survey sample data of 10 common highly pollen-allergenic tree species and 1 type of other plant patches from the.kml format to the.shp format, and ensure that the number of samples in each category is uniform, with the number of samples for each tree species between 400 and 500, to obtain the vector data set of highly pollen-allergenic tree species patch samples.

[0032] Step 2: Write and run code on the GEE remote sensing processing platform, call the Sentinel-2 remote sensing data set during the high-incidence period of common pollen allergy risks from March to June and from September to October in 2022, and upload the local tree species patch samples, arbors, shrubs, and grasses, and the vector data of the target area boundary in the target area to carry out remote sensing identification. The specific process is as follows:

[0033] Step 2.1: Eliminate the radiation error through the FLAASH atmospheric correction model.

[0034] Step 2.2: Use the QA60 cloud mask band to remove clouds and noise, convert the affected location scene subset to the apparent reflectance at the top of the atmosphere, and obtain the median of the minimum cloud pixels according to the simple cloud fraction to get a composite remote sensing image with all cloud cover less than 20%.

[0035] Step 2.3: Perform image fusion operations on multiple remote sensing images using the pixel median replacement method.

[0036] Step 2.4: First, mosaic and splice multiple adjacent fused images, then read the vector boundary data of the target area and crop to obtain the remote sensing image of the target area.

[0037] Step 2.5: Read the shrub and grass vegetation data as a mask to extract the vegetation remote sensing image data of the target area.

[0038] Step 2.6: Based on the reflectance values of the multi-spectral bands of the remote sensing image, calculate a total of 6 vegetation indices including NDVI, GNDVI, SAVI, OSAVI, TVI, and EVI, which are common, easy to calculate, and representative. Their descriptions and calculation formulas are shown in Table 1. Set all the band information from B1 - B9 and the 6 vegetation indices as the remote sensing feature set for classification.

[0039] Step 2.7: Call the random forest classification model, add a random attribute with a value of a random number from 0 to 1 to the training feature set to generate training samples. Use the samples with the generated random number greater than 0.7 as test data, otherwise as training data, thus dividing the data into two parts. Use 70% of the data for training and 30% for validation.

[0040] Step 2.8: Perform classification operations on the remote sensing image according to the set various data and parameters.

[0041] Step 2.9: Predict the test data and create a confusion matrix for accuracy evaluation. The specific indicators include overall accuracy, user accuracy, producer accuracy, Kappa coefficient, and the harmonic mean accuracy of user accuracy and producer accuracy. Their calculation formulas and descriptions are shown in Table 2. After calculating the classification accuracy of the 5 accuracy evaluation indicators, generate a result data table as shown in Table 3, achieving a high-precision spatial remote sensing identification of high-pollen-sensitizing tree species with a total classification accuracy of 85.93%.

[0042] Step 2.10: Set the output parameters such as a spatial resolution of 10m, a range of the target area, a projection of EPSG:4326 (WGS84), and a maximum number of pixels of 1e13 to obtain the raster data of the classification results of high-pollen-sensitizing tree species, and its spatial distribution is as Figure 2 shown.

[0043] Step 3: Count the spatial distribution area of highly pollen-sensitizing tree species to obtain the aggregated spatial distribution characteristics and distribution area of highly pollen-sensitizing tree species in the city. The sensitizing tree species are relatively dispersed in the central part of Beijing urban area and relatively concentrated in the surrounding areas, accounting for 55.80% of the total green space vegetation area. Among them, Sophora japonica has the widest distribution, and the park green spaces between the Third Ring Road and the Fifth Ring Road are the aggregation areas of pollen-sensitizing tree species. The specific process is as follows:

[0044] Step 3.1: In the ArcGIS software, use the raster area calculation function. Take the raster area of each tree species as its spatial distribution area, and take the sum of the raster areas of all highly pollen-sensitizing tree species and other plants as the green space area, as shown in Table 4.

[0045] Step 3.2: Calculate the number of species, distribution area, and proportion in the green space of highly pollen-sensitizing tree species in a specific area, as shown in Table 5.

[0046] Table 1 Calculation formulas and explanations of common vegetation indices of the present invention

[0047]

[0048]

[0049] Table 2 Calculation formulas and explanations of accuracy evaluation indicators of the present invention

[0050]

[0051]

[0052] Table 3 Classification accuracy evaluation results of highly pollen-sensitizing tree species in Beijing urban area based on full-band and vegetation index eigenvalues of the present invention

[0053]

[0054] Table 5 Spatial recognition and classification area of highly pollen-sensitizing tree species in Beijing urban area of the present invention

[0055]

[0056] Example 2

[0057] In the computer system, select the raster data of other tree species obtained in Example 1 (that is, Figure 2 ) with NDVI values between 0.6 - 0.7 as the area to be detected. Determine the Sabina chinensis supplementary area by measuring the backscattering coefficient in the VV polarization mode of the area to be detected and selecting the area with coefficients between 10 - 12, and merge the Sabina chinensis supplementary area into the raster data of the Sabina chinensis tree species.

[0058] Significance of Example 2

[0059] It is well-known that two-leaf plants such as Chinese juniper will produce leaves that are completely different in appearance (with huge differences in shape, color, and size) during different growth periods. In traditional satellite remote sensing identification methods for Chinese juniper, the dominant tree species method is usually adopted for identification. However, it is easy to understand that urban tree species are usually planted in artificial patches with a single type of plant and almost the same growth period. If a certain artificial Chinese juniper forest just enters the leaf-changing period, then the remote sensing data of its messy leaves will have a huge difference from the preset parameters in the database and will then be identified as other tree species. In Example 2, the areas that have been identified as other tree species are screened by the NDVI value and the backscattering coefficient of the VV polarization method, and the areas that may be misidentified are quickly selected in a simple way, so as to avoid the omission of some Chinese juniper planting areas. Of course, this example is only for showing the identification of two-leaf plants, and it does not mean that this technology can only be used for Chinese juniper. In fact, it is also applicable to the identification and classification of plants with obvious two-leaf characteristics such as Syngonium podophyllum.

[0060] Working principle of Example 2

[0061] In practice, the NDVI value of Chinese juniper is generally between 0.5 and 0.8. For areas with an NDVI value greater than 0.7, due to their strong growth, the possibility of being misidentified is relatively small; while for areas with an NDVI value less than 0.6, even if they are Chinese juniper, their growth is too slow and the possibility of spreading pollen is relatively low. Therefore, the area to be detected is set to 0.6 - 0.7. Through this simple screening method, the tree species areas suspected to be Chinese juniper among other tree species can be quickly obtained. Finally, the suspected tree species areas are verified by the backscattering coefficient of the VV polarization method, greatly improving the accuracy rate.

[0062] What is described in the above examples and the specification only illustrates the principle and the best embodiments of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed.

Claims

1. Remote sensing identification method for highly pollen-allergenic tree species in cities, characterized in that, It includes the following steps: Step 1: Vectorize the patch sample data; Step 2: On the Google Earth Engine remote sensing big data processing and presentation platform, call the Sentinel-2 remote sensing dataset of the target time and area, and upload the local pollen highly sensitizing tree species patch samples, arbors, shrubs and grasses vegetation, and boundary vector data. Realize the identification classification and mapping of pollen highly sensitizing tree species by writing and running codes; Step 3: Analyze the spatial distribution of the remote sensing identification classification results of pollen highly sensitizing tree species, and count the areas of various sensitizing tree species in the ArcGIS software to obtain the spatial distribution characteristics of highly sensitizing tree species in the city.

2. The evaluation method according to claim 1, wherein: The method for vectorizing the patch sample data in Step 1: In the ArcGIS software, convert the pollen highly sensitizing tree species research samples into.shp vector format, and assign sample category attributes to distinguish tree species types to obtain a sample vector dataset.

3. The evaluation method according to claim 1, wherein: The process of Step 2 is as follows: Step 2.1: Preprocess the remote sensing image data; Step 2.2: Extract and combine remote sensing features, obtain the multi-spectral band reflectance value information of the remote sensing image, and calculate the vegetation index for calculation. Use the band information and vegetation index as the classification feature values of pollen highly sensitizing tree species, and select the reflectance values and vegetation index of all bands of the remote sensing image data to set up a remote sensing feature set for classification; Step 2.3: Construct a random forest classification model; Step 2.4: Training and verification of the identification classification of pollen highly sensitizing tree species. Divide the sample data into two categories, training and verification, and input them into the random forest classification model to obtain the classification operation results; Step 2.5: Accuracy evaluation and spatial distribution mapping of the remote sensing identification classification of pollen highly sensitizing tree species. Use the confusion matrix to measure the accuracy of the classification model. First, separately count the number of result values of incorrect and correct classification in the classification model, then display them all in a table, and finally obtain the matrix form result table of rows and columns. Determine the correct or incorrect classification of a model in the predicted and true categories through the confusion matrix, and whether the classification model confuses two categories.

4. The evaluation method according to claim 3, characterized in that: The preprocessing of the remote sensing image data includes: Screening: Select remote sensing images according to the target time range, perform atmospheric correction processing, exclude the interference of atmospheric scattering and absorption, and factors such as clouds, aerosols, and light on image interpretation to obtain the true spectral information of ground objects reflected to the sensor; Cloud removal: Perform cloud and noise removal on the image. After obtaining all images with cloud cover below a certain percentage, create and synthesize remote sensing satellite composite images; Fusion: Traverse the median at the same pixel of all images and calculate to obtain a multi-spectral fusion image presented in the form of pixel median; Mosaicking and cropping: Mosaic and splice multiple adjacent images together to obtain a large-scale remote sensing image that is seamless, continuous, and covers the entire region. Accurately crop the range of the mosaicked image according to the target area vector boundary, obtain the corresponding image data, use the arbors, shrubs and grasses vegetation classification visual interpretation data as a mask to obtain the vegetation remote sensing image data of the target area, and uniformly resample all data to the same spatial resolution.

5. The evaluation method according to claim 3, characterized in that: The construction of the random forest classification model includes: Random sampling is performed on the initial training dataset by randomly sampling with replacement, obtaining data from the original data that may be repeated or not selected at all to establish a new data training set. By taking the average of multiple random subsets, the risk of overfitting is reduced and the generalization ability of the model to new data is enhanced; Randomly select feature attribute parameters. On the premise that not all features are used for model training, randomly select a feature subset from all feature attributes, and let each decision tree select an optimal feature attribute as the splitting node from the feature subset. Randomly select the features of each node, only considering the random feature subset without considering all feature attributes; Majority voting. Each decision tree outputs a classification label respectively, and a majority vote is taken on all labels to obtain the final classification result, ensuring that the final prediction result is the integration of the results of all decision trees.

6. The evaluation method according to claim 1, wherein: The process of step 3 is as follows: Analysis of the spatial distribution of highly pollen-allergenic tree species, analyzing the areas, administrative regions, green space types, and specific locations where highly pollen-allergenic tree species are densely distributed; Statistics of the spatial distribution area of highly pollen-allergenic tree species. Using the raster area calculation function of ArcGIS software, calculate the distribution area of each tree species, and further calculate the number of species, distribution area, and proportion of highly pollen-allergenic tree species in a specific area.

7. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the recognition method described in any one of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the recognition method described in any one of claims 1-6.

Images