Method for monitoring alien invasive species erigeron annuus based on remote sensing data

Abstract

The invention discloses a method for constructing an annual fleabane monitoring model based on remote sensing data, and the method comprises the following steps: 1), carrying out the ground investigation of a target region in the growth period of annual fleabane, and collecting the sample data of annual fleabane; 2) performing remote sensing multispectral image data acquisition on the target area, and preprocessing the acquired remote sensing multispectral image data; 3) obtaining a grassland area; 4) calculating multi-source remote sensing features of the remote sensing multispectral image data of the grassland area preprocessed in the step 2); (5) permutation and combination are carried out on the multi-source remote sensing features obtained in the step (4), the importance of different multi-source remote sensing features in annual fleabane monitoring is evaluated by utilizing an average descending Gini coefficient, and the importance of the multi-source remote sensing features is sequenced; 6) removing multi-source remote sensing features with high correlation by using a Pearson's correlation coefficient, and 7) constructing an erigeron annuus monitoring model.The erigeron annuus monitoring method is suitable for monitoring, management and early prediction and early warning of erigeron annuus.

Description

Technical Field

[0001] The present invention relates to the technical field of plant monitoring, and particularly relates to a plant monitoring based on remote sensing data. Background Art

[0002] The emergence of alien invasive species often poses a serious threat to local biodiversity and ecosystem functions, and also brings huge economic losses and ecological risks. The alien invasive species in grassland ecosystems are diverse and complex, making their monitoring and management a major challenge. Patent application CN202110440472.9 discloses a method for determining the invasion risk of the alien species Alternanthera philoxeroides in Caohai Wetland.

[0003] Erigeron annuus (L.) Pers. belongs to the genus Erigeron of the family Asteraceae, also known as Erigeron acris L., Artemisia annua L., Hieracium umbellatum L., and Artemisia vulgaris L. It is native to Mexico in North America and currently occurs in grassland ecosystems in countries in the temperate and subtropical regions of the Northern Hemisphere of the world. It belongs to the "List of Alien Invasive Species in China (the Third Batch)". Traditional investigation methods mainly rely on ground-based manual observations, which are time-consuming and laborious and difficult to achieve timely monitoring over a large area.

[0004] The rapid development of remote sensing technology provides flexible data support for the accurate identification and large-scale monitoring of species. However, the complex structure of the vegetation canopy in grassland ecosystems and the differences in the performance of remote sensing data at different scales in the monitoring of alien invasive species pose challenges to the monitoring of alien invasive species. Summary of the Invention

[0005] In order to overcome the above problems, the present invention aims to provide a multi-scale monitoring method for the alien invasive species Erigeron annuus using multi-source data fusion technology.

[0006] In the present invention, "scale" refers to the spatial resolution.

[0007] In the present invention, "multi-scale" refers to different spatial resolutions.

[0008] In the present invention, "multi-source" refers to the general term of spectral features, texture features, and temporal features. The present invention uses multi-source data to provide information in the spectral, spatial, and temporal dimensions.

[0009] With the rapid development of remote sensing technology, multi-scale remote sensing has both refined and macroscopic monitoring capabilities, enabling it to adapt to different application scenarios and providing flexible data support for the accurate identification and large-scale monitoring of alien invasive species.

[0010] In the present invention, for unmanned aerial vehicle (UAV) multispectral image data and satellite multispectral image data at different scales, monitoring models for Erigeron annuus are respectively constructed with different combinations of remote sensing features arranged, and the optimal monitoring models for Erigeron annuus at different scales are selected.

[0011] Compared with a single data source, multi-source data can make up for the deficiencies of a single data source in monitoring accuracy and more comprehensively reflect the distribution characteristics and dynamic changes of alien invasive species.

[0012] In the present invention, based on UAV multispectral image data, WorldView-2 satellite multispectral image data, and PlanetScope satellite multispectral image data, monitoring models for Erigeron annuus are independently established and independently verified respectively.

[0013] The present invention provides a method for monitoring the alien invasive species Erigeron annuus based on remote sensing data, comprising the following steps:

[0014] 1) During the growth period of Erigeron annuus, conduct a ground survey of the target area to collect sample data of Erigeron annuus.

[0015] 2) Collect remote sensing multispectral image data of the target area and preprocess the collected remote sensing multispectral image data.

[0016] 3) Obtain grassland areas for the calculation and extraction of multi-source remote sensing features.

[0017] 4) Calculate multi-source remote sensing features for the preprocessed remote sensing multispectral image data of the grassland areas.

[0018] 5) Arrange and combine the calculated multi-source remote sensing features, use the mean decrease Gini coefficient to evaluate the importance of different multi-source remote sensing features in the monitoring of Erigeron annuus, and sort according to the importance of the multi-source remote sensing features.

[0019] 6) Use the Pearson correlation coefficient to remove multi-source remote sensing features with high correlations.

[0020] 7) Use the random forest algorithm to construct a monitoring model for Erigeron annuus.

[0021] Further, in the above step 1), when conducting a ground survey of the target area to collect sample data of Erigeron annuus, the specific location of the invasion of Erigeron annuus is determined according to the morphological characteristics of Erigeron annuus. The morphological characteristics during the growth period of Erigeron annuus are mainly manifested as: the stem is straight and has fine hairs; the leaves on the stem are slender and linear-lanceolate, and the leaves at the base are like a rosette close to the ground.

[0022] Further, for the ground survey in step 1) to collect Erigeron annuus sample data in the target area, if the ground survey sample data is collected during the flowering period of Erigeron annuus, the flowering characteristics of Erigeron annuus can be considered simultaneously. However, subsequent modeling does not depend on the flowering period data.

[0023] Further, the Erigeron annuus sample data collected by the ground survey in step 1) above can be augmented with the visible light image data of Erigeron annuus obtained by drones.

[0024] Further, in step 2) above, the remote sensing multispectral image data can be the multispectral image data within the growth period of Erigeron annuus, including the growth period outside the flowering period and / or the flowering period.

[0025] Further, in step 2) above, the remote sensing multispectral image data can be the multispectral image data of the target area within the wavelength range of 400 - 1040 nm.

[0026] Further, in step 2) above, the remote sensing multispectral image data can be the drone multispectral image data and / or the satellite multispectral image data.

[0027] Further, the above satellite multispectral image data can be the WorldView-2 multispectral image data and / or the PlanetScope multispectral image data.

[0028] Further, in step 2) above, for the preprocessing of the collected remote sensing multispectral image data, for the drone multispectral image data, it can include geometric correction, image mosaicking, and radiometric correction; for the WorldView-2 satellite multispectral image data, it can include geometric correction and radiometric correction; for the PlanetScope satellite multispectral image data, it can include cloud removal and median synthesis.

[0029] Further, the image mosaicking of the above drone multispectral images can be the alignment and mosaicking of the collected drone multispectral image data.

[0030] Further, the radiometric correction of the above drone multispectral images can be to construct an empirical relationship between the standard band reflectance provided by a standard diffuse reflectance panel and its pixel values in the drone multispectral images to obtain the reflectance data of the drone multispectral images.

[0031] Further, the radiometric correction of the above WorldView-2 satellite multispectral images can be to convert the original values into surface reflectance data according to the parameters and empirical relationships in the image metadata.

[0032] Furthermore, in the above step 3), the grassland area can be obtained by constructing a model using a random forest algorithm, extracting the grassland area from the satellite multispectral image data, classifying other landforms except grassland in the satellite multispectral image as non-grassland areas, and obtaining a grassland / non-grassland binary classification mask map for extraction and calculation of multi-source remote sensing features of annual ferns; or, obtaining grassland mask data through land use type data in a public database (such as GLC_FCS30, FROM-GLC, etc.) for extraction and calculation of multi-source remote sensing features of annual ferns.

[0033] Furthermore, in the above step 4), the multi-source remote sensing features include spectral features and texture features.

[0034] Furthermore, in the above step 4), the multi-source remote sensing features also include time series features.

[0035] Furthermore, in the above step 5), the multi-source remote sensing features can be arranged and combined, and the average decreasing Gini coefficient in the random forest model can be used to evaluate the importance of different multi-source remote sensing features in annual fleabane monitoring. The larger the average decreasing Gini coefficient of the input feature, the more important the input feature is.

[0036] Furthermore, in the above step 6), the Pearson correlation coefficient can be used to calculate the correlation between different multi-source remote sensing features. The multi-source remote sensing features are ranked according to their importance, and the Pearson correlation coefficient is used to quantify the correlation between the input features. The larger the absolute value of the correlation coefficient, the higher the correlation.

[0037] Furthermore, in the above step 6), multi-source remote sensing features with absolute correlation values ​​greater than 0.7 are removed in sequence according to the importance ranking.

[0038] Furthermore, in the above step 7), the random forest algorithm can be used to construct a monitoring model of annual sedge with different multi-source remote sensing feature combinations, and screen out the best performing models at different scales.

[0039] Furthermore, in the above step 7), the optimal Ipomoea australis monitoring model can be obtained from the constructed Ipomoea australis monitoring model.

[0040] The invention also provides a remote sensing data-based method for constructing a monitoring model for an alien invasive species, i.e., Fleur des Fleurs, and its application in monitoring alien invasive species.

[0041] The present invention also provides a monitoring method for the alien invasive species Juniperus fulvidraco based on remote sensing data, wherein the multi-source remote sensing characteristics of the area to be monitored are input into the monitoring model of the alien invasive species Juniperus fulvidraco at the corresponding scale constructed in the above step 7) to draw a distribution map of Juniperus fulvidraco.

[0042] Furthermore, by using a hotspot analysis tool to determine the hotspot areas of the distribution of Erigeron annuus, the monitoring of the distribution of the alien invasive species Erigeron annuus at different scales can be achieved.

[0043] The present invention also provides an application of the method for monitoring the alien invasive species Erigeron annuus based on remote sensing data in the monitoring of alien invasive species.

[0044] Furthermore, the monitoring model of Erigeron annuus constructed by the present invention can be applied across regions to the monitoring of Erigeron annuus.

[0045] Furthermore, the monitoring model of Erigeron annuus constructed by the present invention can be applied across time to the monitoring of Erigeron annuus. For example, the monitoring model of Erigeron annuus established based on the remote sensing multispectral image data in the non-flowering period of the growth stage can be applied to the monitoring of Erigeron annuus in the non-flowering period of the growth stage, and can also be applied to the monitoring of Erigeron annuus in the flowering period. The monitoring of Erigeron annuus in the flowering period using the monitoring model in the non-flowering period of the growth stage is carried out by using the multi-source remote sensing characteristics of the stem and leaf plants of Erigeron annuus. For another example, the monitoring model of Erigeron annuus established based on the remote sensing multispectral image data in the flowering period can be applied to the monitoring of Erigeron annuus in the flowering period, and can also be applied to the monitoring of Erigeron annuus in the non-flowering period of the growth stage. The monitoring of Erigeron annuus in the non-flowering period of the growth stage using the monitoring model in the flowering period is carried out by using the multi-source remote sensing characteristics of the stem and leaf plants in the multi-source remote sensing characteristics of the flowering period of Erigeron annuus.

[0046] Furthermore, based on the above monitoring of Erigeron annuus using some characteristics of the plant, the monitoring accuracy may be reduced. Establishing a monitoring model based on the remote sensing multispectral image data of the same growth stage for the monitoring of Erigeron annuus in the same growth stage can obtain the maximum monitoring accuracy. However, it may increase the workload of data collection and data processing. The remote sensing multispectral image data of the entire growth stage or multiple growth stages considering the temporal characteristics can be used to construct a more widely used model.

[0047] Specifically, the present invention is as follows:

[0048] 1. A method for constructing a monitoring model of Erigeron annuus based on remote sensing data, comprising the following steps:

[0049] 1) During the growth stage of Erigeron annuus, conduct a ground survey on the target area and collect the sample data of Erigeron annuus;

[0050] 2) Collect remote sensing multispectral image data of the target area, and preprocess the collected remote sensing multispectral image data;

[0051] 3) Obtain the grassland area;

[0052] 4) Calculate the multi-source remote sensing characteristics of the remote sensing multispectral image data of the grassland area preprocessed in step 2);

[0053] 5) Permute the multi-source remote sensing features obtained in step 4), evaluate the importance of different features in the monitoring of Erigeron annuus using the mean decrease Gini coefficient, and rank the importance of the multi-source remote sensing features;

[0054] 6) Use the Pearson correlation coefficient to remove the multi-source remote sensing features with high correlation;

[0055] 7) Construct a monitoring model for Erigeron annuus using the random forest algorithm.

[0056] 2. The method for constructing a monitoring model for Erigeron annuus based on remote sensing data according to item 1, characterized in that: the sample data of Erigeron annuus collected by the ground survey in step 1) is augmented through the visible light image data of the unmanned aerial vehicle of Erigeron annuus.

[0057] 3. The method for constructing a monitoring model for Erigeron annuus based on remote sensing data according to item 1 or 2, characterized in that: in the acquisition of the remote sensing multispectral image data in step 2), the multispectral image data can be multispectral image data in the wavelength range of 400-1040 nm.

[0058] 4. The method for constructing a monitoring model for Erigeron annuus based on remote sensing data according to any one of items 1 to 3, characterized in that: the remote sensing multispectral image data in step 2) is unmanned aerial vehicle multispectral image data and / or satellite multispectral image data.

[0059] 5. The method for constructing a monitoring model for Erigeron annuus based on remote sensing data according to item 4, characterized in that: the satellite multispectral image data is WorldView-2 multispectral image data and / or PlanetScope multispectral image data.

[0060] 6. The method for constructing a monitoring model for Erigeron annuus based on remote sensing data according to any one of items 1 to 5, characterized in that: in step 3), the grassland area is obtained by extracting a binary classification mask map of grassland / non-grassland from the satellite multispectral image, or the grassland mask data is obtained through the land use type data of the public database.

[0061] 7. The method for constructing a monitoring model for Erigeron annuus based on remote sensing data according to any one of items 1 to 6, characterized in that: in step 4), the multi-source remote sensing features include spectral features and texture features.

[0062] 8. The method for constructing a monitoring model for Erigeron annuus based on remote sensing data according to item 7, characterized in that: in step 4), the multi-source remote sensing features further include temporal features.

[0063] 9. The method for constructing a monitoring model of Erigeron annuus based on remote sensing data according to any one of items 1 to 8, characterized in that: in step 6), the Pearson correlation coefficient is used to remove multi-source remote sensing features with an absolute value of correlation greater than 0.7.

[0064] 10. The method for constructing a monitoring model of Erigeron annuus based on remote sensing data according to any one of items 1 to 9, characterized in that: in step 7), the optimal monitoring model of Erigeron annuus is obtained in the constructed monitoring model of Erigeron annuus.

[0065] 11. A monitoring method for Erigeron annuus based on remote sensing data, comprising the following steps:

[0066] 1) Adopt the method for constructing a monitoring model of Erigeron annuus based on remote sensing data according to any one of items 1 to 10 to construct a monitoring model of Erigeron annuus;

[0067] 2) Input the multi-source remote sensing features of the area to be measured into the monitoring model of the corresponding scale constructed in step 1) to draw a distribution map of Erigeron annuus;

[0068] 3) Use a hotspot analysis tool to determine the hotspot areas of the distribution of Erigeron annuus.

[0069] 12. The application of the method for constructing a monitoring model of Erigeron annuus based on remote sensing data according to any one of items 1 to 10 and / or the monitoring method for Erigeron annuus based on remote sensing data according to item 11 in the monitoring of alien invasive species.

[0070] The present invention provides a flexible, efficient and high-precision monitoring method, providing key technical support for the accurate identification and dynamic monitoring of Erigeron annuus. Compared with the prior art, the present invention is based on the fusion of multi-source remote sensing features, including spectral features, texture features, and optionally including temporal features, to establish a monitoring model of Erigeron annuus in the grassland ecosystem. Using the monitoring model of Erigeron annuus of the present invention, Erigeron annuus and its distribution characteristics can be quickly and non-destructively identified.

[0071] The present invention discloses a monitoring method for the alien invasive species Erigeron annuus based on remote sensing data. The method of the present invention is applicable to technical fields such as the monitoring, management and early prediction and warning of the alien invasive species Erigeron annuus. The present invention can establish a multi-scale monitoring model of the alien invasive species Erigeron annuus according to the multi-scale remote sensing data of the invaded area, combined with spectral features, texture features and optional temporal features, to achieve the rapid, efficient and accurate identification of the alien invasive species Erigeron annuus.

[0072] The present invention effectively solves the limitations of a single data source by using multi-source remote sensing feature fusion technology. By comprehensively utilizing the complementary information of different features, the monitoring accuracy of Erigeron annuus has been significantly improved, providing a scientific basis for the monitoring and control of Erigeron annuus in grassland ecosystems, and having great application potential for promoting ecological management and protection. Brief Description of the Drawings

[0073] Figure 1 It is a technical flow chart of an embodiment of the present invention.

[0074] Figure 2 It is the mapping result of the distribution of the alien invasive species Erigeron annuus in an embodiment of the present invention.

[0075] Figure 3 It is the analysis result of the hot spot area of the distribution of the alien invasive species Erigeron annuus in an embodiment of the present invention. Detailed Embodiments

[0076] To better understand the present invention, the present invention gives the following embodiments in combination with the drawings. It should be understood that the embodiments of the present invention are only used to explain the present invention rather than limit it, and the protection scope of the present invention is only defined by the claims of the present invention. The embodiments provided by the present invention are only some preferred embodiments and do not impose any form of limitation on the present invention. Those skilled in the art can make changes, equivalent substitutions or modifications according to the content of the present invention to form different embodiments. However, any changes and deformations made to the method of the present invention and any equivalent substitutions are within the protection scope of the present invention without departing from the concept of the present invention.

[0077] Figure 1This is a technical flow chart of an implementation mode of the present invention. Taking the alien invasive species Erigeron annuus as the research object, a multi-scale monitoring model of Erigeron annuus was constructed using ground survey data, unmanned aerial vehicle (UAV) and satellite multi-spectral image data. First, the invasion situation of Erigeron annuus was surveyed on the experimental plots, and the centimeter-level real-time kinematic (RTK) positioning technology was used to locate the invasion samples to obtain the sample data of Erigeron annuus. Optionally, the sample data can be augmented by combining the visible light image data of the UAV. Then, the UAV remote sensing multi-spectral image data and / or satellite remote sensing multi-spectral image data of the target area were obtained. The collected multi-spectral image data was preprocessed. For the UAV multi-spectral image data, it can include geometric correction, image mosaicking and radiometric correction; for the WorldView-2 satellite multi-spectral image data, it can include geometric correction and radiometric correction; for the PlanetScope satellite multi-spectral image data, it can include cloud removal and median synthesis. Then, the land use was classified to extract the grassland areas and obtain the grassland mask. Next, the multi-source remote sensing features such as the spectral features and texture features of the grassland areas were calculated. Next, the mean decrease Gini coefficient and Pearson correlation coefficient were used for feature screening, and the screened multi-source remote sensing features were combined. On the basis of optimizing the multi-source remote sensing feature combination, a random forest algorithm was used to construct a monitoring model of Erigeron annuus. Then, the accuracy of the monitoring model of Erigeron annuus was evaluated to determine the optimal monitoring model, and a distribution map of Erigeron annuus was drawn. Finally, a hot spot map of Erigeron annuus was obtained based on the hot spot analysis tool.

[0078] Example 1: Collection of sample data of the alien invasive species Erigeron annuus

[0079] The sample data collection of Erigeron annuus can select the sample data at any growth stage of Erigeron annuus, including but not limited to the florescence data.

[0080] Sufficient sample data can ensure that the established model has good robustness. Generally speaking, the number of samples of each type of Erigeron annuus and non-Erigeron annuus should be no less than 200. In this example, the total number of samples of Erigeron annuus and non-Erigeron annuus should be no less than 400.

[0081] The florescence of Erigeron annuus is around July. In this example, the ground survey was carried out in July, and at this stage, Erigeron annuus was in the florescence.

[0082] For the collection of sample data of the alien invasive species Erigeron annuus based on ground surveys, the specific locations of Erigeron annuus invasion are determined according to the morphological characteristics of Erigeron annuus. The morphological characteristics of Erigeron annuus are mainly manifested as follows: the stems are straight and have fine hairs; the leaves on the stems are slender and linear-lanceolate, and the basal leaves are like a rosette and cling tightly to the ground. If ground survey sample data collection is carried out during the flowering period of Erigeron annuus, the flowering characteristics of Erigeron annuus can be considered simultaneously. When Erigeron annuus is in the flowering period, the flowers are white and generally clustered; the inflorescences are arranged in an umbrella shape, with white or light purple ligulate flowers on the outer layer and yellow tubular flowers in the center.

[0083] The experimental area of this embodiment is located in the Dajiuhu National Wetland Park in Shennongjia Forest Region, Hubei Province (109°58'~110°8'E, 31°24'~31°33'N), as shown in Figure 2 (a) and Figure 3 .

[0084] The initial area of the ground survey can be determined through local historical survey data or on-site exploration. According to the morphological characteristics of Erigeron annuus, the invasion locations of Erigeron annuus are determined in the ground survey. The initial area of the ground survey in this embodiment is Figure 2 (b) the area shown within the box. On July 12, 2022, in Figure 2 (b) the initial area of the ground survey within the box, a ground survey of the invasion status of Erigeron annuus is carried out, and the RTK positioning technology at the centimeter level is used to calibrate the invasion locations of Erigeron annuus. The specific method is as follows: (1) Determine the specific locations of Erigeron annuus invasion according to the morphological characteristics of Erigeron annuus; (2) Select areas where Erigeron annuus invasion is relatively concentrated and set 1m×1m quadrats as samples; (3) Use the RTK positioning technology to record the coordinates of the four corner points of the quadrats.

[0085] Example 2: Collection of UAV visible light image data

[0086] If the sample data based on ground surveys is insufficient, the sample of Erigeron annuus can be expanded by collecting UAV visible light image data of Erigeron annuus. The expansion of the sample data of Erigeron annuus can select sample data at any growth stage of Erigeron annuus, including but not limited to flowering period data. In other words, the sample expansion is not limited to using the visible light image data of Erigeron annuus during the flowering period, and the visible light image data during the non-flowering period can also be used for sample expansion.

[0087] In Example 1 of the present invention, the sample data is insufficient for modeling and evaluation, so the UAV visible light image data of Erigeron annuus is used to expand the sample.

[0088] The sample of this embodiment is expanded based on the visible light image data of Erigeron annuus during the flowering period.

[0089] Visible light image data was acquired using a vertical takeoff and landing fixed-wing unmanned aerial vehicle (AFD-C model) produced by Beijing Avatar Intelligent Technology Co., Ltd. To obtain higher-quality data, a time period with good weather conditions, clear skies, and no clouds was selected for collection. The UAV data was collected at noon on July 12, 2022 (11:00 - 13:00). The flight parameters were set as follows: the flight altitude was 300 meters, the overlap rates in the forward and cross directions were 80%, and the flight speed was 4 m / s. At the same time, 27 control points with obvious features were calibrated using RTK positioning technology to establish the georeference of the UAV visible light image data. The spatial resolution of the acquired UAV visible light image data was 0.04 m.

[0090] Combined with the acquired UAV visible light image data of Erigeron annuus and the image features of Erigeron annuus in the visible light image data, the sample dataset of Erigeron annuus was further supplemented by the method of manual visual inspection. Specifically, the samples in this embodiment were expanded based on the visible light image data during the flowering period of Erigeron annuus. In the UAV visible light image data during the flowering period of Erigeron annuus, the image features of Erigeron annuus were mainly manifested as white flowers, and they generally clustered together.

[0091] In this embodiment, when the coverage of Erigeron annuus within a pixel (corresponding to a 1 m × 1 m quadrat) is greater than 50%, it is determined as a sample of Erigeron annuus; while when the coverage of Erigeron annuus within a pixel is less than 5%, it is determined as a non-Erigeron annuus sample.

[0092] The sample data expanded in this embodiment was allocated in a ratio of 7:3 for subsequent model construction and accuracy evaluation.

[0093] Example 3: Acquisition of UAV multispectral image data

[0094] The UAV multispectral image data can be image data at any growth stage, including but not limited to image data during the flowering period. The modeling does not depend on the flowering period data of Erigeron annuus.

[0095] Multi-spectral image data was obtained using a vertical takeoff and landing fixed-wing unmanned aerial vehicle (AFD-C type) produced by Beijing Avatar Intelligent Technology Co., Ltd. To obtain higher-quality data, a time period with good weather conditions, clear sky and no clouds was selected for collection. The unmanned aerial vehicle data was collected at noon (11:00 - 13:00) on May 19, 2022 and July 12, 2022. The flight parameters were set as follows: flight altitude was 300 meters, the forward and side overlap rates were 80%, and the flight speed was 4 m / s. Before the flight mission was executed, a MicaSense reflectance calibration panel was placed for subsequent radiometric correction. At the same time, 27 control points with obvious features were calibrated using RTK positioning technology to establish the georeference of the unmanned aerial vehicle multi-spectral image data. The spatial resolution of the obtained unmanned aerial vehicle multi-spectral image data was 0.15 m. The spectral range of the multi-spectral image data was 455 nm - 880 nm, including 5 channels, namely blue light (455 nm - 495 nm), green light (540 nm - 580 nm), red light (658 nm - 678 nm), red edge (707 nm - 727 nm) and near-infrared (800 nm - 880 nm) bands.

[0096] Example 4: Collection of satellite multi-spectral image data

[0097] WorldView-2 multi-spectral image data on August 22, 2022 (cloud cover less than 1%) was obtained through Beijing TopoVision Technology Co., Ltd., and WorldView-2 is commercial data. The spatial resolution of this data was 2 m, and the spectral range was 400 nm - 1040 nm, including 8 channels, namely coastal (400 nm - 450 nm), blue light (450 nm - 510 nm), green light (510 nm - 580 nm), yellow light (585 nm - 625 nm), red light (630 nm - 690 nm), red edge (705 nm - 745 nm), near-infrared 1 (770 nm - 895 nm) and near-infrared 2 (860 nm - 1040 nm) bands.

[0098] PlanetScope multispectral image data from May to September 2022 was obtained through the Planet API. The spatial resolution of this data is 3m, the spectral range is 432nm - 885nm, and it contains 8 channels, namely coastal (432nm - 452nm), blue (465nm - 515nm), green 1 (513nm - 549nm), green 2 (547nm - 583nm), yellow (600nm - 620nm), red (650nm - 680nm), red edge (697nm - 713nm), and near infrared (845nm - 885nm) bands. In order to make the PlanetScope multispectral image data comparable with the drone multispectral image data and the WorldView-2 multispectral image data (i.e., the band ranges tend to be consistent), in the PlanetScope multispectral image data used for subsequent calculations and analyses in this embodiment, the green 2 band was selected for the green band. In fact, in the process of modeling using the PlanetScope multispectral image data, it is not necessary to select the green 2 band for the green band, and the data of the green 1 band and the green 2 band can be used for modeling.

[0099] Example 5. Preprocessing of Drone Multispectral Image Data and Satellite Multispectral Image Data

[0100] For the drone multispectral image data, Agisoft Metashape software was used to perform geometric correction, image stitching, and radiometric correction on the collected drone data. First, images with sufficient clarity and overlap were selected to support subsequent feature matching and spatial positioning. Secondly, the coordinates of the control points were imported to perform geometric correction on the images, and the AlignPhotos tool was used for image stitching. Finally, the Calibrate Reflectance tool and the MicaSense reflectance calibration plate were used for radiometric correction.

[0101] For the WorldView-2 satellite multispectral image data, the obtained images have already been stitched. First, in ENVI 5.3.1, the rational polynomial coefficient orthonormalization method and the NASA digital elevation model were used to perform geometric correction on the WorldView-2 data. Then, the AbsCalFactor and EffectiveBandwidth parameters in the metadata and the FLAASH atmospheric correction method were used for radiometric correction.

[0102] For the PlanetScope satellite multispectral image data, the acquired images have been geometrically corrected, image mosaicked, and radiometrically corrected. First, in Google Earth Engine, the cloud-covered areas are removed using the Q1 band. The pixels with a Q1 band value of 1 indicate that the pixel is not covered by clouds, and other pixels with values not equal to 1 are removed. Second, the median() function is used to obtain the multispectral median composite image data from May to September.

[0103] Example 6: Extract grassland areas

[0104] The grassland areas can be extracted based on satellite multispectral image data or the grassland mask data can be obtained based on the land use type data in the public database, such as GLC_FCS30 or FROM-GLC, etc.

[0105] In this example, based on the PlanetScope multispectral image data and the random forest algorithm, a land use classification model is constructed to extract grassland areas. The land use types include grassland, forest, shrub, water body, building, wetland, and farmland. The grassland is classified as the grassland area, and other land use types are classified as non-grassland areas. Finally, a binary classification mask map of grassland / non-grassland is obtained, and the grassland areas in the UAV and satellite multispectral image data are extracted using the binary classification mask map of grassland / non-grassland.

[0106] The obtained grassland areas are used for the calculation and extraction of subsequent multi-source remote sensing features.

[0107] Example 7: Calculate spectral features, texture features, and temporal features

[0108] For spectral features, the blue, green, red, red edge, and near-infrared bands are selected, and 17 spectral features are calculated using different algebraic combinations, including 7 original bands and 10 common vegetation indices, for the monitoring of the alien invasive species Erigeron annuus. For texture features, 18 features sensitive to spatial information are calculated in Google Earth Engine for the monitoring of the alien invasive species Erigeron annuus. For temporal features, the maximum, minimum, and range of spectral features and texture features are calculated for the monitoring of the alien invasive species Erigeron annuus. Among them, spectral features, texture features, and temporal features are calculated based on UAV multispectral image data; spectral features and texture features are calculated based on WorldView-2 multispectral image data; spectral features, texture features, and temporal features are calculated based on PlanetScope multispectral image data. The calculation formulas for all the above features are shown in Table 1.

[0109] Table 1: Spectral features, texture features, and temporal features for the monitoring of the alien invasive species Erigeron annuus

[0110]

[0111] Note: Red, Green, Blue, RedEdge, and NIR respectively refer to the reflectance of the red light band, green light band, blue light band, red edge band, and near-infrared band.

[0112] Example 8. Feature Screening

[0113] The sample data is divided into training samples and validation samples in a ratio of 7:3, as shown in Table 2.

[0114] Table 2. Number of Samples for Model Training and Validation Remote sensing data Training samples Validation samples Unmanned aerial vehicle 533 229 WorldView-2 531 230 PlanetScope 533 231

[0116] First, based on 70% of the training sample data, a model is constructed using the random forest algorithm, and the input features shown in Table 1 are sorted according to the mean decrease Gini coefficient. Second, the Pearson correlation coefficient is used to calculate the correlation between the input features. The larger the absolute value of the correlation coefficient, the higher the correlation. The features with an absolute correlation value greater than 0.7 are removed in order of importance. Finally, different combinations of remote sensing features at different scales are obtained.

[0117] Table 3 shows the key input features of different combinations of remote sensing features used to construct the monitoring model of the alien invasive species Erigeron annuus at different scales.

[0118] Table 3. Key Input Features of the Monitoring Model of Alien Invasive Species at Different Scales Feature combination Unmanned aerial vehicle WorldView-2 PlanetScope a EAI, TVI GNDVI, EAI, EVI EAI, TVI b EAI, TVI, Blue_SAVG, Blue_DISS GNDVI, EAI, EVI, Blue_SAVG, Blue_CON, NIR1_SAVG,RedEdge_DISS, NIR2_SHADE,Green_CORR, RedEdge_PROM,Blue_IMCORR1, Costal_CORR EAI, TVI, Green_SAVG, Red_DVAR, Blue_PROM, NIR_SAVG, Blue_SHADE, Costal_SHADE, Costal_CORR, RedEdge_DVAR,RedEdge_SVAR, NIR_INTERIA, RedEgde_SHADE, Red_CORR, Blue_IDM, NIR_SHADE c EAI, TVI, Blue_SAVG, Blue_DISS, Green_Diff / EAI, TVI, Green_SAVG, Red_DVAR, Blue_PROM, NIR_SAVG, Blue_SHADE, Costal_SHADE, Costal_CORR, RedEdge_DVAR,RedEdge_SVAR, NIR_INTERIA, RedEgde_SHADE, Red_CORR, Blue_IDM, NIR_SHADE,RedEdge_Diff, GNDVI_Diff, Green_Min,TVI_Diff, GNDVI_Max, EAI_Diff, Red_Diff

[0120] Note: a, b, and c respectively refer to spectral features, spectral features + texture features, and spectral features + texture features + temporal features.

[0121] As shown in Table 3, for spectral features, the Erigeron annuus index (EAI) is the key input feature of the monitoring model of the alien invasive species Erigeron annuus at different scales, while the performance of other indices varies at different scales. For texture features, the spatial information of the blue light band (Blue) is relatively important. For temporal features, the range (Diff) of spectral features is relatively more important.

[0122] Example 9. Construction and Evaluation of the Monitoring Model of the Alien Invasive Species Erigeron annuus

[0123] Based on the determined remote sensing feature combinations at different scales, a monitoring model for Erigeron annuus was constructed using the random forest algorithm to evaluate the potential of multi-source remote sensing feature fusion in the monitoring of the alien invasive species Erigeron annuus. The training samples were input into the random forest algorithm to construct the Erigeron annuus monitoring model. Then, the validation data was input into the constructed Erigeron annuus monitoring model for model evaluation. Two evaluation indicators, overall accuracy (OA) and Marco-F1 score, were used to evaluate the accuracy of the model. All modeling processes were carried out in the Rstudio-2023.06.1 software, and the RandomForest() function in the RandomForest package was used for model construction, with the parameters ntree and max_depth being 500 and the default value respectively.

[0124] The calculation methods of overall accuracy and Marco-F1 are as follows:

[0125]

[0126]

[0127]

[0128] In the formula, TP represents true positive, indicating that the prediction result is positive (Positive), the actual is positive, and the prediction is correct (Truth); FP represents false positive, indicating that the prediction result is positive (Positive), the actual is negative, and the prediction is incorrect (False); TN represents true negative, indicating that the prediction result is negative (Negative), the actual is negative, and the prediction is correct (Truth); FN represents false negative, indicating that the prediction result is negative (Negative), the actual is positive, and the prediction is incorrect (False); N represents the number of categories. In this embodiment, it includes two categories, Erigeron annuus and non-Erigeron annuus, and N is 2.

[0129] Example 10. Comparison of model accuracy

[0130] Table 4 shows the accuracy of the monitoring model for the alien invasive species Erigeron annuus at different scales, where OA represents the overall accuracy of the model.

[0131] Table 4. Monitoring model accuracy of the alien invasive species Erigeron annuus at different scales

[0132]

[0133] Note: a, b, and c refer to spectral features, spectral features + texture features, and spectral features + texture features + temporal features respectively.

[0134] As shown in Table 4, at the UAV scale, the model with the highest accuracy is the one that fuses spectral features and texture features (feature combination b), with an overall accuracy of 97.6% and a Marco-F1 score of 0.976. In the WorldView-2 satellite multispectral image data, the model with the highest accuracy is also the one that fuses spectral features and texture features (feature combination b), with an overall accuracy of 77.6% and a Marco-F1 score of 0.776. Although WorldView-2 lacks temporal features, the fusion of its spectral features and texture features can still effectively improve the monitoring accuracy. In the PlanetScope satellite multispectral image data, the model with the highest accuracy is the one that fuses spectral features, texture features, and temporal features (feature combination c), with an overall accuracy of 82.1% and a Marco-F1 score of 0.817. The results show that the fusion of multi-source remote sensing data can effectively improve the accuracy of monitoring the alien invasive species Erigeron annuus, especially for satellite multispectral image data with a low spatial resolution.

[0135] In this embodiment, the spectral features, texture features, and temporal features of the UAV multispectral image data are extracted based on the multispectral image data in May (non-flowering period) and July (flowering stage), and the spectral features, texture features, and temporal features of the PlanetScope satellite multispectral image data are extracted based on the multispectral image data from May to September. The UAV multispectral images and the PlanetScope satellite multispectral images fully cover the multispectral image features of Erigeron annuus in the non-flowering and flowering stages of the growth period, and the model can be directly applied to the monitoring of Erigeron annuus in the non-flowering and flowering periods.

[0136] In this embodiment, the spectral features and texture features of the WorldView-2 satellite multispectral image data are extracted based on the multispectral image data in July (flowering stage), and the model can be directly applied to the monitoring of Erigeron annuus in the non-flowering and flowering stages of the growth period. However, to obtain better monitoring results, when monitoring Erigeron annuus in the non-flowering period, multispectral image data in the non-flowering stage of the growth period can also be used, or multispectral image data in the non-flowering and flowering stages of the growth period can be comprehensively used to extract spectral features, texture features, and temporal features to construct a new model.

[0137] Example 11. Monitoring mapping for identifying the alien invasive species Erigeron annuus

[0138] Figure 2 The results are for the monitoring mapping of the invasive species Erigeron annuus. The regional results show a comparison at a broader and macroscopic level, while the plot results show a comparison at a smaller and microscopic level. Through the comparison of these two methods, the credibility of the monitoring results can be better illustrated. By comparing the model accuracies in Table 4, the UAV model that fuses spectral features and texture features is determined as the optimal monitoring model for Erigeron annuus. Figure 2(a) shows the overall distribution of Erigeron annuus drawn based on the optimal monitoring model of Erigeron annuus. Figure 2 (b) and Figure 2 (c) are respectively the comparison charts of visible light images and optimal prediction results in the invaded area. Figure 2 (e) to Figure 2 (p) are respectively the comparison charts of visible light images and optimal prediction results of the sample plots. Specifically, Figure 2 (e) is the visible light image of sample plot Z-1. Figure 2 (f) to Figure 2 (h) are respectively the prediction results of the drone, WorldView-2, and PlanetScope multispectral image data of sample plot Z-1. Figure 2 (i) to Figure 2 (l) and Figure 2 (m) to Figure 2 (p) are respectively the comparison results of sample plot Z-2 and sample plot Z-3. It can be seen that through visual judgment, the optimal prediction results and the distribution of Erigeron annuus in the visible light image (i.e., the white flowering area) have a high degree of consistency. All three types of data can better identify the spatial distribution of Erigeron annuus, but in terms of spatial details, the drone data is better than the satellite data. As Figure 2 (k) shows, there are certain misclassifications in the WorldView-2 data, but the newly added temporal features in the PlanetScope data can overcome its limitations to a certain extent. Generally speaking, the drone multispectral image data can provide more refined prediction results, while the satellite multispectral can provide prediction results for a larger spatial range. Appropriate data can be selected according to actual needs for the monitoring of Erigeron annuus.

[0139] Example 12: Identifying the hotspots of the alien invasive species Erigeron annuus

[0140] In ArcGIS 10.8, the Getis-Ord GI* hotspot analysis tool was used to calculate the aggregation degree of Erigeron annuus using the inverse distance weighting method and Euclidean distance, and the hotspot areas of Erigeron annuus invasion were generated. Figure 3 is the analysis result of the hotspot areas of Erigeron annuus distribution. As Figure 3 shown, the higher the Z value, the higher the aggregation degree of Erigeron annuus, indicating the more serious invasion degree of this area.

Claims

1. A method for constructing a monitoring model of annual sedge based on remote sensing data, comprising the following steps: 1) During the growing period of annual sedge, conduct ground surveys in the target area and collect annual sedge sample data; 2) collecting remote sensing multispectral image data of the target area, and preprocessing the collected remote sensing multispectral image data; 3) Get the grassland area; 4) calculating multi-source remote sensing features for the remote sensing multispectral image data of the grassland area preprocessed in step 2); 5) Arranging and combining the multi-source remote sensing features obtained in step 4), using the average decreasing Gini coefficient to evaluate the importance of different multi-source remote sensing features in the monitoring of annual sedge, and ranking the importance of the multi-source remote sensing features; 6) Use the Pearson correlation coefficient to remove multi-source remote sensing features with high correlation; 7) A random forest algorithm was used to construct a monitoring model for Ipomoea australis.

2. The method for constructing a monitoring model of annual ferns based on remote sensing data according to claim 1, characterized in that: The sample data of the annual sedge collected by the ground survey described in step 1) are expanded by using the visible light image data of the annual sedge from unmanned aerial vehicles.

3. The method for constructing a monitoring model of annual ferns based on remote sensing data according to claim 1 or 2, characterized in that: In the remote sensing multispectral image data acquisition described in step 2), the multispectral image data is multispectral image data within the wavelength range of 400 to 1040 nm.

4. The method for constructing a monitoring model of annual ferns based on remote sensing data according to any one of claims 1 to 3, characterized in that: The remote sensing multispectral image data described in step 2) is unmanned aerial vehicle multispectral image data and / or satellite multispectral image data.

5. The method for constructing a monitoring model of annual ferns based on remote sensing data according to claim 4, characterized in that: The satellite multispectral image data is WorldView-2 multispectral image data and / or PlanetScope multispectral image data.

6. The method for constructing a monitoring model of annual ferns based on remote sensing data according to any one of claims 1 to 5, characterized in that: In step 3), the grassland area is obtained by extracting a binary classification mask map of grassland / non-grassland from satellite multispectral images, or obtaining grassland mask data through land use type data in a public database.

7. The method for constructing a monitoring model of annual ferns based on remote sensing data according to any one of claims 1 to 6, characterized in that: In step 4), the multi-source remote sensing features include spectral features and texture features.

8. The method for constructing a monitoring model of annual ferns based on remote sensing data according to claim 7, characterized in that: In step 4), the multi-source remote sensing features also include time series features.

9. The method for constructing a monitoring model of annual ferns based on remote sensing data according to any one of claims 1 to 8, characterized in that: In step 6), the Pearson correlation coefficient is used to remove multi-source remote sensing features with absolute correlation values ​​greater than 0.

7.

10. The method for constructing a monitoring model of annual ferns based on remote sensing data according to any one of claims 1 to 9, characterized in that: In step 7), the optimal monitoring model of the annual sedge is obtained from the constructed monitoring models of the annual sedge.

11. A monitoring method for annual fleabane based on remote sensing data, comprising the following steps: 1) constructing a monitoring model for Ipomoea australis based on remote sensing data using the method for constructing an Ipomoea australis monitoring model based on remote sensing data according to any one of claims 1 to 10; 2) inputting the multi-source remote sensing characteristics of the area to be measured into the monitoring model of the corresponding scale constructed in step 1) to draw a distribution map of the annual sedge; 3) Use hotspot analysis tools to determine the hotspot areas of annual sedge distribution.

12. Use of the method for constructing a monitoring model of Fleur des Fleurs based on remote sensing data according to any one of claims 1 to 10 and / or the monitoring method of Fleur des Fleurs based on remote sensing data according to claim 11 in monitoring invasive alien species.

Images