Method for Identifying Long-Term Spartina alterniflora Distribution Areas Based on GEE and Multi-Source Remote Sensing Data

Abstract

The present invention discloses a long-term sequence Spartina alterniflora distribution area recognition method based on GEE and multi-source remote sensing data, comprising the following steps: A. Obtain at least two series of satellite remote sensing images of the recognition area, and each series of satellite remote sensing images forms a remote sensing data set; B. Perform modular processing on all remote sensing data sets; C. Perform image fusion processing on the remotely sensed data after modular processing to identify the distribution area of Spartina alterniflora and its spatio-temporal changes. The present invention can improve the deficiencies of the prior art, enhance the recognition accuracy and recognition speed, and is conducive to identifying the distribution area characteristics in the initial stage of Spartina alterniflora colonization.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental remote sensing monitoring, and in particular to a method for identifying the distribution area of Spartina alterniflora with long time series based on GEE and multi-source remote sensing data. Background Art

[0002] Spartina alterniflora is an alien plant originally used to protect tidal flats and river embankments. However, due to its strong reproductive ability and lack of natural enemies, it has grown rapidly in the coastal wetlands of China in recent years, causing serious impacts on the ecosystem. In order to effectively monitor Spartina alterniflora, satellite remote sensing technology has been widely applied. However, the existing remote sensing data processing and identification methods cannot balance speed and accuracy in the process of identifying the distribution areas of plants such as Spartina alterniflora. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method for identifying the distribution area of Spartina alterniflora with long time series based on GEE and multi-source remote sensing data, which can solve the deficiencies of the existing technology, improve the identification accuracy and speed, and is beneficial to identifying the distribution area characteristics in the initial stage of Spartina alterniflora colonization.

[0004] To solve the above technical problems, the technical solutions adopted by the present invention are as follows.

[0005] A method for identifying the distribution area of Spartina alterniflora with long time series based on GEE and multi-source remote sensing data includes the following steps:

[0006] A. Obtain at least two series of satellite remote sensing images of the identification area, and each series of satellite remote sensing images forms a remote sensing data set;

[0007] B. Perform modular processing on all remote sensing data sets;

[0008] C. Perform image fusion processing on the remotely sensed data after modular processing to identify the distribution area of Spartina alterniflora and its spatio-temporal changes.

[0009] Preferably, multi-temporal 10-meter resolution satellite remote sensing images of Sentinal-2 in the identification area are obtained from the GEE platform, and multi-temporal 10-meter resolution satellite remote sensing images of SPOT in the identification area are obtained from the French National Center for Space Studies (CNES).

[0010] Preferably, in step B, first perform cloud removal processing on all remote sensing data sets, including the following steps:

[0011] Respectively generate neural network models for the two remote sensing data sets of SPOT and SENTINAL-2, generate training data sets for each neural network model using the historical data of the corresponding remote sensing data set, and train the corresponding neural network models using the training data sets;

[0012] The images in the remote sensing dataset are segmented using the trained neural network model, corresponding weight values are assigned to each image block, and the weighted image blocks are recombined to obtain the cloud-removed remote sensing data.

[0013] Preferably, a pair of bidirectional transfer functions is provided between the hidden layers of the two neural network models. The pair of bidirectional transfer functions includes a penalty term and is used to share the features extracted by the hidden layers of the two neural network models. After the hidden layer obtains new features, they are combined with the current features.

[0014] Preferably, the modular processing of the SENTINAL-2 satellite remote sensing dataset includes the following steps.

[0015] The suitable growth areas of Spartina alterniflora are divided according to different spectral images in the SENTINAL-2 satellite remote sensing dataset. Feature extraction is performed on other areas in the remote sensing image, and fuzzy processing is performed on the premise of retaining the features. Then, the suitable growth areas of Spartina alterniflora and the other areas after fuzzy processing are recombined into a remote sensing image.

[0016] Preferably, for other areas in the remote sensing image, Gaussian filtering is first performed for denoising processing, and then the image area is traversed to calculate the gray gradient vector of each pixel. When calculating, it is judged whether the magnitude of the gray gradient of the current pixel is the local maximum in its neighborhood in the direction of the gray gradient of the pixel. If so, the pixel is retained for the next round of comparison; otherwise, the pixel is no longer used for comparison. All local maximum pixels are combined into a feature set, and the feature set is linearized to obtain the linear feature set of other areas in the remote sensing image.

[0017] The fuzzy processing of other areas in the remote sensing image includes the following steps. First, a fuzzy window matrix is defined, and the elements in the fuzzy window matrix are the weight values of the corresponding pixels. The fuzzy window matrix is used to traverse the area to be processed for weighted processing. When the pixels in the linear feature set are included in the fuzzy window matrix, no weighted processing is performed on the pixel, and at the same time, the gray mean value of the pixel and its adjacent pixels is calculated, and then the gray mean value is used to replace the gray value of the pixel adjacent to the pixel in the linear feature set.

[0018] Before the remote sensing image is recombined, feathering processing is performed on the edges of the recombined image blocks. After the remote sensing image is recombined, sharpening processing is performed on the image stitching positions.

[0019] Preferably, the modular processing of the SPOT satellite remote sensing dataset includes the following steps.

[0020] Arrange the remotely sensed images of the SENTINAL-2 satellite after modular processing in chronological order, and match the remotely sensed images of the SPOT satellite with the SENTINAL-2 satellite remotely sensed images that are closest to it in time series; mark the set of linear features in the SPOT satellite remotely sensed images that match the SENTINAL-2 satellite remotely sensed images on the SENTINAL-2 satellite remotely sensed images, and calculate the position transfer function of the SENTINAL-2 satellite remotely sensed images and the SPOT satellite remotely sensed images according to the position deviation of the same set of linear features on the SENTINAL-2 satellite remotely sensed images and the SPOT satellite remotely sensed images, and use the position transfer function to mark the suitable growth area of Spartina alterniflora on the SPOT satellite remotely sensed images in the SENTINAL-2 satellite remotely sensed images that match it.

[0021] Preferably, identifying the distribution area of Spartina alterniflora includes the following steps:

[0022] Calculate the normalized difference vegetation index of the suitable growth area of Spartina alterniflora in the remotely sensed images of the SENTINAL-2 satellite, and then use the normalized difference vegetation index to fit the growth curve of Spartina alterniflora in each suitable growth area of Spartina alterniflora;

[0023] Read the full-band reflectance data of the suitable growth area of Spartina alterniflora in the remotely sensed images of the SPOT satellite, and mark the distribution area of Spartina alterniflora in the remotely sensed images of the SPOT satellite according to the reflectance characteristics of Spartina alterniflora in each band;

[0024] Correct the growth curve of Spartina alterniflora according to the distribution area of Spartina alterniflora in the remotely sensed images of the SPOT satellite in chronological order, use the corrected growth curve of Spartina alterniflora to inversely calculate the distribution area of Spartina alterniflora, and then obtain the spatio-temporal change of the distribution area of Spartina alterniflora according to the change of the distribution area of Spartina alterniflora in time series.

[0025] The beneficial effects brought by adopting the above technical solution are as follows: The present invention comprehensively utilizes two sets of satellite remotely sensed images of SPOT and SENTINAL-2, and by optimizing the processing process of remotely sensed images, reduces the amount of computation while not affecting the recognition accuracy, thereby improving the recognition speed of remotely sensed images. Specific Embodiments

[0026] The existing technologies for the processing and recognition of remotely sensed images usually process remotely sensed images in a unified process, and then perform object recognition according to the processing results. Since the monitoring range of Spartina alterniflora is very wide and the number of remotely sensed images involved is huge, this leads to a large consumption of computing power and time for the processing and recognition of remotely sensed images of Spartina alterniflora at the present stage.

[0027] To address this issue, we abandoned the traditional remote sensing image processing method and redesigned a set of remote sensing image processing and recognition algorithms based on the resources of the GEE platform according to the growth and geographical distribution characteristics of Spartina alterniflora. The specific steps are as follows.

[0028] A. Obtain multi-temporal 10-meter resolution satellite remote sensing images of Sentinal-2 in the identification area from the GEE platform, and obtain multi-temporal 10-meter resolution satellite remote sensing images of SPOT in the identification area from the French National Center for Space Studies (CNES). Each series of satellite remote sensing images forms a remote sensing data set.

[0029] B. First, perform cloud removal processing on all remote sensing data sets, including the following steps.

[0030] Generate neural network models for the two remote sensing data sets of SPOT and SENTINAL-2 respectively. A pair of bidirectional transfer functions is set between the hidden layers of the two neural network models. The pair of bidirectional transfer functions includes penalty terms and is used to share the features extracted by the hidden layers of the two neural network models. After the hidden layer obtains new features, they are combined with the current features. Use the historical data of the corresponding remote sensing data set to generate the training data set for each neural network model, and use the training data set to train the corresponding neural network model.

[0031] Use the trained neural network model to divide the images in the remote sensing data set into blocks, assign corresponding weight values to each image block, and recombine the weighted image blocks to obtain the cloud-removed remote sensing data.

[0032] The modular processing of the SENTINAL-2 satellite remote sensing data set includes the following steps.

[0033] Divide the suitable growth areas of Spartina alterniflora according to different spectral images in the SENTINAL-2 satellite remote sensing data set. Extract features from other areas in the remote sensing image, perform fuzzy processing while retaining the features, and then recombine the suitable growth areas of Spartina alterniflora and the other areas after fuzzy processing into a remote sensing image.

[0034] For other areas in the remote sensing image, first perform denoising processing using Gaussian filtering, then traverse the image area, calculate the gray gradient vector of each pixel, and while calculating, judge whether the magnitude of the gray gradient of the current pixel is the local maximum in its neighborhood in the direction of the gray gradient of the pixel. If so, retain the pixel for the next round of comparison; otherwise, no longer use the pixel for comparison. Combine all local maximum pixels into a feature set, and linearize the feature set to obtain the linear feature set of other areas in the remote sensing image.

[0035] Blurring other areas in the remote sensing image includes the following steps. First, define a blurring window matrix, where the elements in the blurring window matrix are the weight values of the corresponding pixels. Use the blurring window matrix to traverse the area to be processed and perform weighted processing. When the pixels in the blurring window matrix include pixels of the linear feature set, do not perform weighted processing on these pixels. At the same time, calculate the gray mean value of this pixel and its adjacent pixels, and then use the gray mean value to replace the gray value of the pixels adjacent to the pixels of the linear feature set.

[0036] Before the remote sensing image is recombined, feathering processing is performed on the edges of the recombined image blocks. After the remote sensing image is recombined, sharpening processing is performed on the image splicing positions.

[0037] The modular processing of the SPOT satellite remote sensing dataset includes the following steps.

[0038] Arrange the processed SENTINAL-2 satellite remote sensing images in time sequence, and match the SPOT satellite remote sensing image with the SENTINAL-2 satellite remote sensing image that is closest to it in time sequence; mark the linear feature set in the SPOT satellite remote sensing image that matches it on the SENTINAL-2 satellite remote sensing image, and calculate the position transfer function of the SENTINAL-2 satellite remote sensing image and the SPOT satellite remote sensing image according to the position deviation of the same linear feature set on the SENTINAL-2 satellite remote sensing image and the SPOT satellite remote sensing image, and use the position transfer function to mark the suitable growth area of Spartina alterniflora on the SPOT satellite remote sensing image in the SENTINAL-2 satellite remote sensing image that matches it.

[0039] C. Perform image fusion processing on the remotely sensed data after modular processing to identify the distribution area of Spartina alterniflora and its spatio-temporal changes; the identification process includes the following steps.

[0040] Calculate the normalized difference vegetation index of the suitable growth area of Spartina alterniflora in the SENTINAL-2 satellite remote sensing image, and then use the normalized difference vegetation index to fit the growth curve of Spartina alterniflora in each suitable growth area of Spartina alterniflora.

[0041] Read the full-band reflectance data of the suitable growth area of Spartina alterniflora in the SPOT satellite remote sensing image, and mark the distribution area of Spartina alterniflora in the SPOT satellite remote sensing image according to the reflectance characteristics of Spartina alterniflora in each band.

[0042] Correct the growth curve of Spartina alterniflora according to the distribution area of Spartina alterniflora in the SPOT satellite remote sensing image in time sequence, use the corrected growth curve of Spartina alterniflora to inversely calculate the distribution area of Spartina alterniflora, and then obtain the spatio-temporal changes of the distribution area of Spartina alterniflora from the changes of the distribution area of Spartina alterniflora in time sequence.

[0043] The present invention selects two representative series of satellite remote sensing images, namely SPOT and SENTINAL-2, for image processing and recognition. The first step in the preprocessing of remote sensing images is to remove clouds in the images. Based on the existing method of using neural network models to remove clouds, the present invention utilizes the corresponding relationship between two sets of remote sensing images in terms of time series and space, and sets a bidirectional transfer function between two neural network models to share the features extracted from the hidden layers of the two neural network models. This can not only accelerate the training convergence speed of the model and improve the F1 score, but also effectively improve the generalization ability of the model during cloud removal, thereby accelerating the image processing speed.

[0044] Traditional remote sensing image recognition methods usually directly use remote sensing images of different bands for direct regional recognition. Due to the large amount of remote sensing image data, this recognition method usually uses the normalized difference vegetation index (NDVI) for indirect recognition, resulting in poor recognition accuracy. Since we use two series of satellite remote sensing images for image recognition, we can use the two series of images to correct each other to improve the recognition accuracy. However, the increase in the number of images will lead to a decrease in the recognition speed. To address this contradiction, we improved the image recognition algorithm. Since the shooting cycle of the SENTINAL-2 satellite is much shorter than that of the SPOT satellite, we select the remote sensing images of the SENTINAL-2 satellite as the main image set, calculate and obtain the suitable growth area of Spartina alterniflora and the linear feature sets of other areas. Then, we use the linear feature sets to match the remote sensing images of the SPOT satellite with those of the SENTINAL-2 satellite, and quickly mark the suitable growth area of Spartina alterniflora on the remote sensing images of the SPOT satellite. This omits the separate processing process of the remote sensing images of the SPOT satellite. When identifying the distribution area of Spartina alterniflora, we only accurately identify the suitable growth area of Spartina alterniflora in the remote sensing images of the SENTINAL-2 satellite to obtain the growth curve of Spartina alterniflora, and the recognition area is greatly reduced, which can improve the recognition speed. Then, we use the distribution area of Spartina alterniflora in the remote sensing images of the SPOT satellite (since the number of remote sensing images of the SPOT satellite is small, the secondary recognition and processing of the suitable growth area of Spartina alterniflora in it will not cause a significant increase in the computational load) to correct the growth curve of Spartina alterniflora to improve the recognition accuracy.

[0045] The present invention comprehensively utilizes two series of satellite remote sensing images, namely SPOT and SENTINAL-2, to conduct long-time series recognition of the distribution area of Spartina alterniflora, with high speed and accuracy, providing a brand-new method for the environmental monitoring of coastal wetlands. Moreover, the recognition method provided by the present invention also has a wide range of application scenarios in other fields. Next, we will conduct in-depth research on this recognition method to expand its applicable fields.

[0046] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and what is described in the above embodiments and the specification is only to illustrate the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will also have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A long-term Spartina alterniflora distribution area identification method based on GEE and multi-source remote sensing data, characterized in that It includes the following steps: A. Obtain at least two series of satellite remote sensing images of the identification area, and each series of satellite remote sensing images forms a remote sensing data set; B. Perform modular processing on all remote sensing data sets; C. Perform image fusion processing on the remotely sensed data after modular processing to identify the Spartina alterniflora distribution area and its spatio-temporal changes; Among them, multi-temporal 10-meter resolution satellite remote sensing images of Sentinal-2 in the identification area are obtained from the GEE platform, and multi-temporal 10-meter resolution satellite remote sensing images of SPOT in the identification area are obtained from the French National Center for Space Studies (CNES); The modular processing of the SPOT satellite remote sensing data set includes the following steps: Arrange the modular processed SENTINAL-2 satellite remote sensing images in time sequence, match the SPOT satellite remote sensing images with the SENTINAL-2 satellite remote sensing images that are closest to them in time sequence; mark the set of linear features in the SPOT satellite remote sensing images that match them on the SENTINAL-2 satellite remote sensing images, calculate the position transfer function of the SENTINAL-2 satellite remote sensing images and the SPOT satellite remote sensing images according to the position deviation of the same set of linear features on the SENTINAL-2 satellite remote sensing images and the SPOT satellite remote sensing images, and use the position transfer function to mark the suitable growth area of Spartina alterniflora on the SPOT satellite remote sensing images in the SENTINAL-2 satellite remote sensing images that match them; Identifying the Spartina alterniflora distribution area includes the following steps: Calculate the normalized difference vegetation index (NDVI) of the suitable growth area of Spartina alterniflora in the SENTINAL-2 satellite remote sensing images, and then use the NDVI to fit the growth curve of Spartina alterniflora in each suitable growth area of Spartina alterniflora; Read the full-band reflectance data of the suitable growth area of Spartina alterniflora in the SPOT satellite remote sensing images, and mark the Spartina alterniflora distribution area in the SPOT satellite remote sensing images according to the reflectance characteristics of Spartina alterniflora in each band; Correct the Spartina alterniflora growth curve according to the Spartina alterniflora distribution area in the SPOT satellite remote sensing images in time sequence, use the corrected Spartina alterniflora growth curve to inversely calculate the Spartina alterniflora distribution area, and then obtain the spatio-temporal changes of the Spartina alterniflora distribution area using the changes in the Spartina alterniflora distribution area in time sequence.

2. The method for identifying the distribution area of Spartina alterniflora with long time series based on GEE and multi-source remote sensing data according to claim 1, wherein: In step B, first perform cloud removal processing on all remote sensing data sets, including the following steps: Generate neural network models for the two remote sensing data sets of Sentinal-2 and SPOT respectively, generate training data sets for each neural network model using the historical data of the corresponding remote sensing data set, and train the corresponding neural network models using the training data sets; Use the trained neural network models to block the images of the remote sensing data sets, assign corresponding weight values to each image block, and recombine the weighted image blocks to obtain the remotely sensed data after cloud removal processing.

3. The method for identifying the distribution area of Spartina alterniflora with long time series based on GEE and multi-source remote sensing data according to claim 2, characterized in that: A bidirectional transfer function pair is set between the hidden layers of two neural network models. The bidirectional transfer function pair includes a penalty term. The bidirectional transfer function pair is used to share the features extracted by the hidden layers of the two neural network models. After the hidden layer obtains new features, they are combined with the current features.

4. The method for identifying the distribution area of Spartina alterniflora with long time series based on GEE and multi-source remote sensing data according to claim 2, wherein: The modular processing of the Sentinal-2 satellite remote sensing dataset includes the following steps. Based on different spectral images in the Sentinal-2 satellite remote sensing dataset, the suitable growth areas of Spartina alterniflora are demarcated. Feature extraction is performed on other areas in the remote sensing image. After retaining the features, fuzzy processing is carried out. Then, the suitable growth areas of Spartina alterniflora and the other areas after fuzzy processing are recombined into a remote sensing image.

5. The method for identifying the distribution area of Spartina alterniflora with long time series based on GEE and multi-source remote sensing data according to claim 4, wherein: For other areas in the remote sensing image, first, Gaussian filtering is performed for denoising. Then, the image area is traversed, and the gray gradient vector of each pixel is calculated. While calculating, it is judged whether the magnitude of the gray gradient of the current pixel is the local maximum in its neighborhood in the direction of the gray gradient of the pixel. If so, the pixel is retained for the next round of comparison; otherwise, the pixel is no longer used for comparison. All local maximum pixels are combined into a feature set, and the feature set is linearized to obtain the linear feature set of other areas in the remote sensing image. The fuzzy processing of other areas in the remote sensing image includes the following steps. First, a fuzzy window matrix is defined. The elements in the fuzzy window matrix are the weight values of the corresponding pixels. The fuzzy window matrix is used to traverse the area to be processed for weighted processing. When the pixels in the linear feature set are included in the fuzzy window matrix, no weighted processing is performed on this pixel. At the same time, the gray mean value of this pixel and its adjacent pixels is calculated, and then the gray value of the pixel adjacent to the pixel in the linear feature set is replaced with the gray mean value. Before the remote sensing image is recombined, feathering processing is performed on the edges of the recombined image blocks. After the remote sensing image is recombined, sharpening processing is performed on the image splicing positions.