A mangrove classification method based on multi-source images

Abstract

The present invention provides a mangrove classification method based on multi-source images, the method comprising: obtaining multi-source remote sensing image data of a region to be classified, preprocessing the multi-source remote sensing image data; obtaining the spectral characteristics and water body index of the preprocessed optical remote sensing image data and the VH band of the preprocessed radar remote sensing image data, constructing a new mangrove index according to the spectral characteristics, water body index and VH band; using a Gaussian mixture model to train the region to be classified, and obtaining a mangrove classification result. The method uses multi-source remote sensing images as a data source for classification, can suppress weather interference and the influence of tides, dig out more vegetation spatial structure information available in more bands, and combines the Gaussian mixture model to discard sample sets and predetermined parameters, resist the influence of clouds and fog, and improve the accuracy of mangrove information extraction.

Description

Technical Field

[0001] The invention belongs to the field of remote sensing image processing, and in particular relates to a mangrove classification method based on multi-source images. Background Art

[0002] Mangroves mostly grow in shallows and muddy areas. Traditional field operations have great limitations in monitoring mangroves, and there are problems such as discontinuous monitoring data and poor contrast, making it difficult to achieve long-term dynamic monitoring of mangroves. With the vigorous development of remote sensing technology, providing complete and effective monitoring of mangroves in real time and historical tracing has become an important technical means for mangrove monitoring and mapping. In the previous construction of mangrove indices, most of them considered medium and high spatial resolution optical images (Landsat, Sentinel-2 and GF-1, etc.) or time series images. However, traditional optical images are easily affected by strong interference from weather factors such as clouds, fog and precipitation, resulting in a small number of available images. At the same time, the spectral responses of mangroves and coastal vegetation are highly similar, and it is difficult to accurately identify mangroves by relying solely on optical data.

[0003] In the previous mangrove classification research, it is mainly divided into two categories. The first is sample-based supervised classification, which requires a large number of sample sets to train the classifier. Sample training is used to ensure the accuracy of mangrove mapping. However, the quality of the sample set and the choice of the classifier will have a great impact on the classification results. Therefore, mangrove mapping depends largely on the reliability of the sample, and it is often poorly generalized and inefficient when applied to a large area. In addition, due to the diversity of mangrove species, the complexity of the growth environment or other factors, the model trained in each area may not be applicable to another area. The second is unsupervised classification based on index threshold segmentation. The index threshold segmentation is used to identify the target object according to the size of the index. At the same time, each index is composed of different bands, which can better express the properties of the target object than a single band. The index can more objectively reflect the spectral information of the target object and has good applicability under different conditions. Combined with the threshold method, it can achieve fast and large-scale information extraction. However, the traditional threshold segmentation requires a unified threshold through prior knowledge. Due to the heterogeneity of geographic space, the unified threshold is not suitable for large areas. Summary of the invention

[0004] In view of the shortcomings of the above technologies, the present invention provides a mangrove classification method based on multi-source images, which can expand the research scope to land areas and improve the extraction accuracy of mangroves. The mangrove classification method based on multi-source images includes:

[0005] Multi-source remote sensing image data of the area to be classified is obtained, and the multi-source remote sensing image data is preprocessed, wherein the multi-source remote sensing image includes at least one optical remote sensing image from one source and one radar remote sensing image from one source.

[0006] The spectral characteristics of the preprocessed optical remote sensing image data are obtained, wherein the spectral characteristics include bands B1, B2, B3, B4, B5, B6, B7, B8, B8A, B11 and B12, and the water body index is calculated according to the spectral characteristics.

[0007] Obtain the preprocessed radar remote sensing image data and obtain the VH band.

[0008] A new mangrove index is constructed according to the spectral characteristics, water body index and VH band. The calculation formula of the new mangrove index is: SSMI=f(S)×f(B)×f(W), where f(S) is the ratio between the B6 band and the B11 band, and f(B)=1 / (1+e -VH ), f(W) is the water index.

[0009] Calculate the new mangrove index for the area to be classified, use the Gaussian mixture model to train the area to be classified for which the new mangrove index has been calculated, and obtain the mangrove classification results.

[0010] Furthermore, the calculation formula of the water body index LSWI is: LSWI=(B8-B11) / (B8+B11).

[0011] Furthermore, the Gaussian mixture model uses the expectation maximization algorithm to estimate model parameters.

[0012] Furthermore, the Gaussian mixture model predicts the result through multiple Gaussian probability density functions, and the number of clusters of the Gaussian probability density function is 4.

[0013] Furthermore, a confusion matrix is ​​used to evaluate the mangrove classification results, and the evaluation indicators include overall accuracy, F1 score, producer accuracy and user accuracy.

[0014] Furthermore, mangrove products, field survey data and Google Earth data are combined to assist in selecting verification samples and verify the accuracy of the mangrove classification results.

[0015] Furthermore, the optical remote sensing image is taken from Sentinel-2, and the radar remote sensing image is taken from Sentinel-1.

[0016] The present invention provides a mangrove classification method based on multi-source images, which uses multi-source remote sensing images as data sources for classification, obtains a band combination that can accurately express the properties of mangroves, determines the band that can best express the change in water content of mangroves, and constructs a new mangrove index for classification. This method can suppress weather interference and the influence of tides, and mine more available vegetation spatial structure information in bands. In combination with a Gaussian mixture model, it discards sample sets and predetermined parameters, resists the influence of clouds and fog, and improves the accuracy of mangrove information extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of the mangrove classification method based on multi-source images.

[0018] Figure 2 This is the principle structure diagram of the Gaussian mixture model.

[0019] Figure 3 Schematic diagram of the mangrove classification method based on multi-source images.

[0020] Figure 4 This is a diagram showing the classification results of mangroves in a certain area in the embodiment. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0022] The embodiment of the present invention provides a mangrove classification method based on multi-source images, as shown in the attached figure. Figure 1 As shown, the method includes:

[0023] The first step is to obtain multi-source remote sensing image data of the area to be classified and pre-process the multi-source remote sensing image data, wherein the multi-source remote sensing image includes at least one optical remote sensing image from one source and one radar remote sensing image from one source.

[0024] In this step, the optical remote sensing images are taken from Sentinel-2 and the radar remote sensing images are taken from Sentinel-1.

[0025] The second step is to obtain the spectral characteristics of the pre-processed optical remote sensing image data, which include bands B1, B2, B3, B4, B5, B6, B7, B8, B8A, B11 and B12, and calculate the water index according to the spectral characteristics. The pre-processed radar remote sensing image data is obtained to obtain the VH band.

[0026] The calculation formula of the water body index LSWI is: LSWI=(B8-B11) / (B8+B11).

[0027] Based on drawing spectral characteristics and backscattering coefficient curves and analyzing the differences between mangroves and other land types, the description of the spectral bands, water body index and radar remote sensing image data is shown in Table 1.

[0028] Table 1

[0029]

[0030] The third step is to construct a new mangrove index based on the spectral characteristics, water body index and VH band. The calculation formula of the new mangrove index is: SSMI=f(S)×f(B)×f(W), where f(S) is the ratio between the B6 band and the B11 band, and f(B)=1 / (1+e -VH ), f(W) is the water index.

[0031] The fourth step is to calculate the new mangrove index for the area to be classified, and use the Gaussian mixture model to train the area to be classified for which the new mangrove index has been calculated to obtain the mangrove classification results.

[0032] In this step, the Gaussian mixture model predicts the results through multiple Gaussian probability density functions (normal distribution curves), and the number of clusters of the Gaussian probability density function is 4. The principle structure of the Gaussian mixture model is shown in the attached figure. Figure 2 As shown, the first step is to construct a Gaussian mixture model and initialize the number of models K. In this embodiment, K is 4; the second step is the estimation step (E-Step) to calculate the posterior probability; the third step is the maximization step (M-Step) to update the parameters; the fourth step is to calculate the log-likelihood function of the model and determine whether the log-likelihood function converges. If it converges, the final result is displayed. If not, return to the second step.

[0033] As attached Figure 3 After obtaining the mangrove classification results, the mangrove products, field survey data and Google Earth data can be combined to assist in selecting verification samples, and the mangrove classification results are verified for accuracy. The mangrove classification results are evaluated using a confusion matrix, and the evaluation indicators include overall accuracy (OverallAccuracy, OA), F1 score, producer accuracy (ProducerAccuracy, PA) and user accuracy (UserAccuracy, UA).

[0034] In another embodiment, Sentinel-2 and Sentinel-1 remote sensing images of a certain area are obtained and preprocessed. The study selected VV polarization and VH polarization Sentinel-1SAR data in 2020, and the processing flow included noise and radiation calibration processing. Sentinel-2MSI data with cloud cover less than 5% from March to October 2020 were selected, and the processing flow included declouding and median synthesis. In order to keep the resolution of Sentinel-1 and Sentinel-2 images consistent, the spatial resolution (nearest neighbor method) of the two data sets was resampled to 10 meters.

[0035] Based on Sentinel-2 data, the water index is obtained using the index formula, and the spectral characteristics and backscattering coefficient curves are plotted and analyzed; the SSMI index of each category in a certain area is calculated; the features after the SSMI index calculation are put into the Gaussian mixture model for training and the mangrove classification results of six regions are obtained, as shown in the attached figure. Figure 4 In order to quantitatively evaluate the stability and applicability of this method, the classification accuracy obtained by using the confusion matrix is ​​shown in Table 2.

[0036] Table 2

[0037]

[0038] It can be seen from Table 2 that the method of the present application can suppress weather interference and the influence of tides, mine the available vegetation spatial structure information in more bands, combine with the Gaussian mixture model, discard the sample set and predetermined parameters, resist the influence of clouds and fog, and improve the accuracy of mangrove information extraction.

[0039] A person skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0040] One or more embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present application should be included in the protection scope of the present application.

Claims

1. A mangrove classification method based on multi-source images, characterized in that: The method comprises: Acquire multi-source remote sensing image data of the area to be classified, and pre-process the multi-source remote sensing image data, wherein the multi-source remote sensing image includes at least one optical remote sensing image from one source and one radar remote sensing image from one source; Acquire spectral features of the preprocessed optical remote sensing image data, wherein the spectral features include bands B1, B2, B3, B4, B5, B6, B7, B8, B8A, B11 and B12, and calculate the water body index according to the spectral features; Obtain pre-processed radar remote sensing image data to obtain the VH band; A new mangrove index is constructed according to the spectral characteristics, water body index and VH band. The calculation formula of the new mangrove index is: SSMI=f(S)×f(B)×f(W), where f(S) is the ratio between the B6 band and the B11 band, and f(B)=1 / (1+e -VH ), f(W) is the water index; Calculate the new mangrove index for the area to be classified, use the Gaussian mixture model to train the area to be classified for which the new mangrove index has been calculated, and obtain the mangrove classification results.

2. The mangrove classification method according to claim 1, characterized in that: The calculation formula of the water body index LSWI is: LSWI=(B8-B11) / (B8+B11).

3. The mangrove classification method according to claim 1, characterized in that: The expectation-maximization algorithm is used in the Gaussian mixture model to estimate model parameters.

4. The mangrove classification method according to claim 1, characterized in that: The Gaussian mixture model predicts the result through multiple Gaussian probability density functions, and the number of clusters of the Gaussian probability density functions is 4.

5. The mangrove classification method according to claim 1, characterized in that: The confusion matrix was used to evaluate the mangrove classification results, and the evaluation indicators included overall accuracy, F1 score, producer accuracy and user accuracy.

6. The mangrove classification method according to claim 1, characterized in that: Mangrove products, field survey data and Google Earth data were combined to assist in selecting verification samples, and the mangrove classification results were verified for accuracy.

7. The mangrove classification method according to claim 1, characterized in that: The optical remote sensing image is taken from Sentinel-2, and the radar remote sensing image is taken from Sentinel-1.

Images