Double-season rice classification method and system based on multi-source remote sensing data and improved index

Abstract

The present invention discloses a double-season rice classification method and system based on multi-source remote sensing data and improved index, and belongs to the technical field of remote sensing image recognition. The method first obtains multi-source remote sensing data in the entire double-season rice planting period, and based on the pre-processed multi-source remote sensing data, respectively extracts the improved double-season rice index characteristics and spectral band characteristics from each optical image, extracts texture features from Sentinel‑2 optical images, extracts terrain features from DEM data, and extracts polarization features from Sentinel‑1 synthetic aperture radar data; finally, all the extracted features are input into a trained random forest model, and a double-season rice spatial distribution map is output. The present invention solves the problem of low observation density and unstable data quality of a single data source by combining multi-source data. The double-season rice phenological index developed by the present invention can accurately capture the phenological characteristics of double-season rice, and can further improve the recognition accuracy by combining other auxiliary features.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing image recognition, and in particular relates to a double-season rice classification method and system based on multi-source remote sensing data and improved indexes. Background Art

[0002] Remote sensing technology plays a key role in the field of crop identification and mapping. Landsat series data can effectively distinguish the spectral characteristics of crops and non-crops and has been widely used in crop identification. Sentinel-2 images are superior to Landsat in crop classification because of their higher spatial resolution. Given that Sentinel-2 and Landsat data provide similar spectral information, joint classification of Landsat and Sentinel-2 data becomes a more reliable solution, which can significantly improve the temporal and spatial resolution of observations. However, the quality of optical images is easily affected by clouds and shadows, especially in years and regions with frequent cloudy and rainy weather, it becomes particularly difficult to obtain high-quality data throughout the planting cycle. Synthetic aperture radar (SAR) data provides a new way for crop identification due to its ability to penetrate clouds. By integrating time series optical images with synthetic aperture radar images, the problem of data availability can be largely solved. Therefore, combining multi-source remote sensing data is an effective way to improve the accuracy of crop extraction in cultivated land.

[0003] Vegetation indices have been widely used in remote sensing phenology extraction. Among them, the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI) were two commonly used indices in the past. In recent years, the kernel NDVI (kNDVI) has become a new vegetation phenology index. Because NDVI has the defect of easy saturation, and compared with EVI, kNDVI is more sensitive to plant physiological and physical parameters, so kNDVI has a better performance in capturing vegetation phenology. The Land Water Index (LSWI) has also been proven to effectively distinguish rice from other crops, but the classification effect often has problems such as low accuracy and poor transferability. In addition, there is a lack of methods for classifying double-season rice. Therefore, it is urgent to explore the classification algorithm of double-season rice with a new index. Summary of the invention

[0004] The purpose of the present invention is to solve the above problems existing in the prior art and provide a double-season rice classification method and system based on multi-source remote sensing data and improved index. Based on multi-source remote sensing data, the present invention develops a new double-season rice index, and combines other auxiliary features to achieve high-precision identification of double-season rice using a robust machine learning algorithm.

[0005] The specific technical solutions adopted by the present invention are as follows:

[0006] In a first aspect, the present invention provides a double-season rice classification method based on multi-source remote sensing data and improved index, comprising:

[0007] S1. Acquire multi-source remote sensing data during the entire double-season rice planting period in the target area, and pre-process them respectively; the multi-source remote sensing data include DEM data, Landsat-8 optical images, Landsat-9 optical images, Sentinel-2 optical images, and Sentinel-1 synthetic aperture radar data;

[0008] S2. Based on the pre-processed multi-source remote sensing data, extract the improved double-season rice index features and spectral band features from each optical image, extract texture features from Sentinel-2 optical images, extract terrain features from DEM data, and extract polarization features from Sentinel-1 synthetic aperture radar data; there are 4 improved double-season rice index features in total, the first index feature is the difference ratio sum between the median synthetic kNDVI index of the double-season early rice at maturity and the median synthetic kNDVI index of the double-season early rice at transplanting period, the second index feature is the difference ratio sum between the median synthetic kNDVI index of the double-season late rice at maturity and the median synthetic kNDVI index of the double-season late rice at transplanting period, the third index feature is the difference ratio sum between the median synthetic LSWI index of the double-season early rice at transplanting period and the median synthetic kNDVI index of the double-season early rice at transplanting period, and the fourth index feature is the difference ratio sum between the median synthetic LSWI index of the double-season late rice at transplanting period and the median synthetic kNDVI index of the double-season late rice at transplanting period;

[0009] S3. Input all the features extracted in S2 into the trained random forest model, and output the spatial distribution map of double-season rice corresponding to the target area.

[0010] As a preferred embodiment of the first aspect, when performing the preprocessing on the multi-source remote sensing data, the three optical images all need to be screened and de-clouded first, wherein the Landsat-8 optical image and the Landsat-9 optical image use the QA band to de-cloud, the Sentinel-2 optical image uses the MSK_CLDPRB band to de-cloud, and the Sentinel-1 synthetic aperture radar data needs to be converted to natural values ​​and have its boundaries removed first, and then the DEM data, the three optical images and the Sentinel-1 synthetic aperture radar data are cropped according to the outline of the target area, and are aligned with the Sentinel-2 optical image as a reference, and finally resampled to the same resolution.

[0011] As a preferred embodiment of the first aspect, the spectral band features extracted from each optical image are the red band, near infrared (Nir) band and shortwave infrared 1 (Swir1) band in the median synthetic image of the entire double-season rice planting cycle.

[0012] As a preferred embodiment of the first aspect, the texture features extracted from the Sentinel-2 optical image are entropy (ENT), angular second moment (ASM) and inverse distance (IDM) extracted from the first principal component of the median synthetic image of the double-season rice transplanting period.

[0013] As a preferred embodiment of the first aspect, the terrain features extracted from the DEM data include slope, aspect and altitude.

[0014] As a preferred embodiment of the first aspect, the polarization feature extracted from the Sentinel-1 synthetic aperture radar data is a ratio of the VV band to the VH band in the median synthetic image during the double-season rice transplanting period.

[0015] As a preferred embodiment of the first aspect, the spatial distribution map of double-season rice needs to be smoothed and denoised by a majority filter to eliminate isolated points to obtain a final spatial distribution map of double-season rice.

[0016] In a second aspect, the present invention provides a double-season rice classification system based on multi-source remote sensing data and improved index, comprising:

[0017] A data acquisition module is used to acquire multi-source remote sensing data in the entire double-season rice planting period in the target area and perform preprocessing respectively; the multi-source remote sensing data includes DEM data, Landsat-8 optical images, Landsat-9 optical images, Sentinel-2 optical images and Sentinel-1 synthetic aperture radar data;

[0018] A feature extraction module is used to extract improved double-season rice index features and spectral band features from each optical image based on the pre-processed multi-source remote sensing data, extract texture features from Sentinel-2 optical images, extract terrain features from DEM data, and extract polarization features from Sentinel-1 synthetic aperture radar data; there are 4 improved double-season rice index features in total, the first index feature is the sum of the difference ratios between the median synthetic kNDVI index of the double-season early rice at maturity and the median synthetic kNDVI index of the double-season early rice at transplanting period, the second index feature is the sum of the difference ratios between the median synthetic kNDVI index of the double-season late rice at maturity and the median synthetic kNDVI index of the double-season late rice at transplanting period, the third index feature is the sum of the difference ratios between the median synthetic LSWI index of the double-season early rice at transplanting period and the median synthetic kNDVI index of the double-season early rice at transplanting period, and the fourth index feature is the sum of the difference ratios between the median synthetic LSWI index of the double-season late rice at transplanting period and the median synthetic kNDVI index of the double-season late rice at transplanting period;

[0019] The classification module is used to input all the features extracted in the feature extraction module into the trained random forest model and output the spatial distribution map of double-season rice corresponding to the target area.

[0020] In a third aspect, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can implement the double-season rice classification method based on multi-source remote sensing data and improved index as described in any one of the schemes of the first aspect above.

[0021] In a fourth aspect, the present invention provides a computer electronic device comprising a memory and a processor;

[0022] The memory is used to store computer programs;

[0023] The processor is used to implement the double-season rice classification method based on multi-source remote sensing data and improved index as described in any one of the solutions of the first aspect when executing the computer program.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1) The present invention combines multi-source data to make up for the problems of low observation density and unstable data quality of a single data source, and develops a double-season rice phenological index to accurately capture the phenological characteristics of double-season rice and combine it with other auxiliary features to further improve recognition accuracy.

[0026] 2) Due to the addition of Sentinel-1 data equipped with C-band synthetic aperture radar (SAR), the present invention is less affected by cloud cover and can handle a wide range of double-season rice classification tasks. Texture variables and end member variables increase the separability of double-season rice to a large extent.

[0027] 3) The present invention performs post-classification processing based on majority filtering, which can further reduce the salt-and-pepper effect of the classification results and improve the reliability of the double-season rice distribution map.

[0028] 4) The data used in the present invention are all publicly available data, so the difficulty of data acquisition is relatively low, and good classification effects can be achieved while taking into account low cost and high efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a schematic diagram of the steps of the double-season rice classification method based on multi-source remote sensing data and improved index;

[0030] Figure 2 The module composition diagram of the double-season rice classification system based on multi-source remote sensing data and improved index;

[0031] Figure 3 A schematic diagram of the components of computer electronic equipment;

[0032] Figure 4 A specific flow chart of the double-season rice classification method in an embodiment of the present invention;

[0033] Figure 5 These are the harmonic fitting results of the kNDVI and LSWI indices in the entire period from January to December in the embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make the above-mentioned purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation mode of the present invention is described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. The technical features in each embodiment of the present invention can be combined accordingly without conflicting with each other.

[0035] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for the purpose of distinction and description, and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features.

[0036] The present invention provides a double-season rice classification method and system based on multi-source remote sensing data and improved index. The technology comprehensively uses multiple remote sensing data sources such as digital elevation model (DEM) data, Landsat-8 / 9 satellite data and Sentinel-1 / 2 satellite data, adopts a new cloud processing method, develops a new double-season rice index, and introduces a new post-processing method to accurately identify double-season rice. The main operation steps are as follows: first, pre-process DEM data, Landsat-8 / 9 data and Sentinel-1 / 2 data, adopt a new cloud probability band declouding algorithm, and perform resampling and image matching, calculate the developed new double-season rice index, extract other auxiliary features, and integrate and stack these feature variables; secondly, use a machine learning algorithm for classification; finally, post-process the classification results through 3×3 window majority filtering and four-pixel connection method and perform accuracy evaluation.

[0037] like Figure 1 As shown, in a preferred embodiment of the present invention, a double-season rice classification method based on multi-source remote sensing data and improved index is provided, and the steps include:

[0038] S1. Acquire multi-source remote sensing data in the entire double-season rice planting period in the target area and perform preprocessing respectively; the multi-source remote sensing data includes digital elevation (DEM) data, Landsat-8 optical images, Landsat-9 optical images, Sentinel-2 optical images and Sentinel-1 synthetic aperture radar data.

[0039] In the embodiment of the present invention, the preprocessing of multi-source remote sensing data needs to be reasonably adjusted according to different types of data. DEM data, Landsat-8 optical images, Landsat-9 optical images, Sentinel-2 optical images and Sentinel-1 synthetic aperture radar data have their own data characteristics, and they need to select an adaptive preprocessing program to ensure that they can be used for subsequent feature extraction.

[0040] In an embodiment of the present invention, the three optical images, Landsat-8 optical images, Landsat-9 optical images, and Sentinel-2 optical images, all need to be first subjected to cloud screening and cloud removal processing. The purpose of cloud screening is to select images with less or no clouds from a large number of remote sensing images for subsequent data analysis and application, while the purpose of cloud removal processing is to remove clouds and cloud shadows in the image, restore the obscured surface information, and improve the clarity and usability of the image. The cloud threshold required for cloud screening can be selected according to actual needs, and the cloud removal processing algorithm can also be optimized according to the actual image type. In an embodiment of the present invention, Landsat-8 optical images and Landsat-9 optical images use the QA band for cloud removal, and Sentinel-2 optical images use the MSK_CLDPRB band for cloud removal.

[0041] In addition, for Sentinel-1 synthetic aperture radar data, its preprocessing operation needs to first perform natural value conversion and boundary removal to improve the data quality and information content.

[0042] In addition, after the above three optical images and Sentinel-1 synthetic aperture radar data are processed accordingly, the DEM data, three optical images and Sentinel-1 synthetic aperture radar data need to be clipped according to the contour of the target area, and aligned with the Sentinel-2 optical image as the benchmark, and finally resampled to the same resolution. The resampling resolution can be adjusted according to actual conditions, for example, it can be resampled to 10m resolution.

[0043] S2. Based on the preprocessed multi-source remote sensing data, the improved double-season rice index characteristics and spectral band characteristics are extracted from each optical image, the texture characteristics are extracted from the Sentinel-2 optical image, the terrain characteristics are extracted from the DEM data, and the polarization characteristics are extracted from the Sentinel-1 synthetic aperture radar data.

[0044] The main difference between rice and other dryland crops is that rice is planted in paddy fields. During the rice transplanting period, the water level in the paddy field is relatively high, and the leaf area of ​​the rice is small and the vegetation coverage is low; while in the maturity period, the vegetation coverage of the rice increases significantly and the paddy field becomes dry. The rice in these two growth stages is highly separable. Since optical images are easily disturbed by cloudy and rainy weather, and the rice-growing areas are mostly humid climates with frequent cloudy and rainy weather, the quality of optical images is difficult to guarantee. In order to better capture the phenological characteristics of double-season rice, the present invention proposes four new improved double-season rice index features, which are specifically aimed at the transplanting period and maturity period of double-season early rice and double-season late rice.

[0045] In an embodiment of the present invention, there are a total of four improved double-season rice index characteristics. The first index characteristic is the difference ratio and sum between the median synthetic kNDVI index of the double-season early rice at maturity and the median synthetic kNDVI index of the double-season early rice at transplanting period. The second index characteristic is the difference ratio and sum between the median synthetic kNDVI index of the double-season late rice at maturity and the median synthetic kNDVI index of the double-season late rice at transplanting period. The third index characteristic is the difference ratio and sum between the median synthetic LSWI index of the double-season early rice at transplanting period and the median synthetic kNDVI index of the double-season early rice at transplanting period. The fourth index characteristic is the difference ratio and sum between the median synthetic LSWI index of the double-season late rice at transplanting period and the median synthetic kNDVI index of the double-season late rice at transplanting period.

[0046] It should be noted that the difference ratio of two indices defined in the present invention refers to the difference between the two indices divided by the sum of the two indices. In other words, the difference ratio of index A to index B = (AB) / (A+B).

[0047] Among the four improved double-season rice index features, the first index feature is a double-season early rice index (ERPI), the second index feature is a double-season late rice index (LPRI), and the third index feature is a normalized kNDVI index (NkNDVI EPR ), the fourth index feature is a double-season late rice normalized kNDVI index (NkNDVI LPR ). Therefore, based on the definition of difference ratio and sum, the calculation formulas of the above four improved double-season rice index characteristics are as follows:

[0048]

[0049]

[0050]

[0051]

[0052] Among them, kNDVI1 and kNDVI2 represent the median composite kNDVI of the double-season early rice transplanting period and the double-season early rice maturity period, respectively; kNDVI3 and kNDVI4 represent the median composite kNDVI of the double-season late rice transplanting period and the double-season late rice maturity period, respectively; LSWI1 and LSWI2 represent the median composite LSWI of the double-season early rice transplanting period and the double-season late rice transplanting period, respectively. The median composite kNDVI and median composite LSWI of the above different periods can be calculated based on the median composite image of the corresponding period, and multi-temporal composite can be pre-performed in the data preprocessing stage to obtain the median composite image.

[0053] It should be noted that double-season early rice and double-season late rice refer to the early rice and late rice corresponding to double-season rice in the target area. The transplanting period and maturity period of each of them need to be determined according to the actual rice transplanting time and rice maturity time in the corresponding area, which is a priori knowledge that needs to be determined in advance.

[0054] The above four improved double-season rice index features need to be used in conjunction and simultaneously for double-season rice area identification in remote sensing images. Because these four improved double-season rice index features reflect the differences in remote sensing images between early and late rice, transplanting and maturity periods in double-season rice, respectively, based on these four improved double-season rice index features, accurate feature information can be extracted to identify double-season rice planting areas.

[0055] It should be noted that for the three types of optical images, namely Landsat-8 optical images, Landsat-9 optical images, and Sentinel-2 optical images, each type of optical image needs to extract four improved double-season rice index features as the input of the subsequent classification model.

[0056] In addition, in addition to the above four improved double-season rice index features, in view of the spatial heterogeneity of different double-season rice planting areas, considering more potential feature variables in the case of uncertainty of the optimal index is helpful to more accurately capture the characteristics of double-season rice, thereby achieving high-precision double-season rice classification. The potential feature variables used in the present invention include spectral band features extracted from each optical image, texture features extracted from Sentinel-2 optical images, terrain features extracted from DEM data, and polarization features extracted from Sentinel-1 synthetic aperture radar data.

[0057] In an embodiment of the present invention, the spectral band characteristics extracted from each optical image are the red band, near infrared (Nir) band and shortwave infrared 1 (Swir1) band in the median synthetic image of the entire double-season rice planting cycle (that is, the median synthetic image of the currently extracted optical image from March to November).

[0058] In an embodiment of the present invention, the texture features extracted from the Sentinel-2 optical image are entropy (ENT), angular second moment (ASM) and inverse difference (IDM) extracted from the first principal component of the median synthetic image of the double-season rice transplanting period. Since double-season rice has two transplanting periods, namely the double-season rice early rice transplanting period (April) and the double-season rice late rice transplanting period (August), it is necessary to perform principal component analysis based on the median synthetic image of the Sentinel-2 optical image of each transplanting period, obtain the first principal component and then extract three textures from it. These extracted textures need to be used as inputs for subsequent classification models.

[0059] In an embodiment of the present invention, the terrain features extracted from the DEM data include slope, aspect and altitude.

[0060] In an embodiment of the present invention, the polarization feature extracted from the Sentinel-1 synthetic aperture radar data is the ratio of the VV band and the VH band in the median synthetic image of the double-season rice transplanting period. Since double-season rice has two transplanting periods, namely the double-season rice early rice transplanting period (April) and the double-season rice late rice transplanting period (August), it is necessary to extract the VV band and the VH band from the median synthetic image based on the Sentinel-1 synthetic aperture radar data of each transplanting period, and then calculate the ratio between the bands as the polarization feature of each transplanting period. The polarization features of both transplanting periods need to be used as inputs for subsequent classification models.

[0061] It should be noted that all the features extracted from S2 above, including the improved double-season rice index feature and spectral band feature extracted from each optical image, the texture feature extracted from the Sentinel-2 optical image, the terrain feature extracted from the DEM data, and the polarization feature extracted from the Sentinel-1 synthetic aperture radar data, are essentially feature maps with the same size as the target area. The eigenvalues ​​corresponding to each pixel in the pixel domain of the target area in the feature maps of all features constitute the feature vector of this pixel position. Based on the feature vector, the random forest model can be used to realize land object classification and identify whether this pixel position is a double-season rice planting area.

[0062] S3. Input all the features extracted in S2 into the trained random forest model, and output the spatial distribution map of double-season rice corresponding to the target area.

[0063] It should be noted that Random Forest is an ensemble learning method in the prior art, which is composed of multiple decision trees. It uses the Bootstrap Aggregating (Bagging) algorithm and decision trees to select the best features, so that it can be better used for classification tasks. Random Forest finally integrates the prediction results of multiple decision trees and uses the majority voting method to determine the final category, thereby improving the accuracy and stability of the model.

[0064] It should be noted that before the random forest model is used for actual reasoning and application, it needs to be optimized and trained using a data set. The specific construction method and training method of the random forest model belong to the prior art and will not be described in detail.

[0065] In addition, the spatial distribution map of double-season rice directly output by the random forest model may have a series of isolated points due to model accuracy or data quality issues, but the planting area of ​​double-season rice often has spatial continuity and presents the characteristics of contiguous distribution. Therefore, for the spatial distribution map of double-season rice directly output by the random forest model, the majority filter can be used for smoothing and denoising, and the final spatial distribution map of double-season rice can be obtained after eliminating isolated points.

[0066] It should be noted that the steps of the double-season rice classification method based on multi-source remote sensing data and improved index shown in S1 to S3 above can essentially be implemented in the form of a computer program.

[0067] In addition, based on the same inventive concept, Figure 2 As shown, the present invention provides a double-season rice classification system based on multi-source remote sensing data and improved index, which includes:

[0068] A data acquisition module is used to acquire multi-source remote sensing data in the entire double-season rice planting period in the target area and perform preprocessing respectively; the multi-source remote sensing data includes DEM data, Landsat-8 optical images, Landsat-9 optical images, Sentinel-2 optical images and Sentinel-1 synthetic aperture radar data;

[0069] A feature extraction module is used to extract improved double-season rice index features and spectral band features from each optical image based on the pre-processed multi-source remote sensing data, extract texture features from Sentinel-2 optical images, extract terrain features from DEM data, and extract polarization features from Sentinel-1 synthetic aperture radar data; there are 4 improved double-season rice index features in total, the first index feature is the sum of the difference ratios between the median synthetic kNDVI index of the double-season early rice at maturity and the median synthetic kNDVI index of the double-season early rice at transplanting period, the second index feature is the sum of the difference ratios between the median synthetic kNDVI index of the double-season late rice at maturity and the median synthetic kNDVI index of the double-season late rice at transplanting period, the third index feature is the sum of the difference ratios between the median synthetic LSWI index of the double-season early rice at transplanting period and the median synthetic kNDVI index of the double-season early rice at transplanting period, and the fourth index feature is the sum of the difference ratios between the median synthetic LSWI index of the double-season late rice at transplanting period and the median synthetic kNDVI index of the double-season late rice at transplanting period;

[0070] The classification module is used to input all the features extracted in the feature extraction module into the trained random forest model and output the spatial distribution map of double-season rice corresponding to the target area.

[0071] Similarly, based on the same inventive concept, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can implement the double-season rice classification method based on multi-source remote sensing data and improved index as described above.

[0072] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0073] Therefore, based on the same inventive concept, the present invention provides a computer-readable storage medium corresponding to a double-season rice classification method based on multi-source remote sensing data and improved index, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it can implement the double-season rice classification method based on multi-source remote sensing data and improved index as described above.

[0074] Therefore, based on the same inventive concept, Figure 3As shown, the present invention also provides a computer electronic device corresponding to the double-season rice classification method based on multi-source remote sensing data and improved index provided in the above embodiment, which includes a memory and a processor;

[0075] The memory is used to store computer programs;

[0076] The processor is used to implement the double-season rice classification method based on multi-source remote sensing data and improved index as described above when executing the computer program;

[0077] Specifically, in the computer-readable storage medium of the above three embodiments, the stored computer program is executed by the processor to perform the above steps S1 to S3.

[0078] It is understandable that the above storage medium may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage. The storage medium may also be a U disk, a mobile hard disk, a magnetic disk or an optical disk, etc., which can store program codes.

[0079] It is understandable that the above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0080] It should also be noted that those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the various embodiments provided in this application, the division of steps or modules in the system and method is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or steps can be combined or integrated together, and a module or step can also be split.

[0081] In order to better demonstrate the specific implementation and technical effect of the double-season rice classification method based on multi-source remote sensing data and improved index of the present invention, a specific example is given below to demonstrate its test performance on the data set.

[0082] Example

[0083] In this embodiment, the specific process of the double-season rice classification method based on multi-source remote sensing data and improved index is as follows: Figure 4 As shown, the specific steps include:

[0084] Step 1: Obtain multi-source remote sensing data during the entire double-season rice planting period in the target area, including DEM data, three optical image data: Landsat 8, Landsat 9 and Sentinel-2, and dual-polarization C-band synthetic aperture radar (SAR) data: Sentinel-1 SAR GRD.

[0085] In this embodiment, the target area is Zhejiang Province, and the planting period of double-season rice is from March to November. Among them, the transplanting period of double-season early rice is the whole month of April, the maturation period of double-season early rice is the whole month of July, the transplanting period of double-season late rice is the whole month of August, and the maturation period of double-season late rice is the whole month of October.

[0086] First, the multi-source remote sensing data is preprocessed, among which: the three optical images of Landsat 8, Landsat 9 and Sentinel-2 all need to be screened and declouded first. Among them, the QA band is used to remove clouds for Landsat-8 optical images and Landsat-9 optical images. For Sentinel-2 optical images, this embodiment introduces a new cloud probability band (MSK_CLDPRB) for declouding. The value of this band is between 0-100, representing the probability of the existence of clouds in the image. Image declouding is achieved by strictly controlling the cloud probability to 0. The present invention uses the MSK_CLDPRB band for declouding, which can effectively remove the cloud cover in Sentinel-2 optical images.

[0087] In addition, for Sentinel-1 synthetic aperture radar data (Sentinel-1 SAR GRD product), natural value conversion and boundary removal are required first.

[0088] Finally, the DEM data, the three processed optical images, and the Sentinel-1 synthetic aperture radar data were clipped according to the contours of the target area, aligned with the Sentinel-2 optical image, and finally resampled to a resolution of 10 m.

[0089] Step 2: Based on the preprocessed multi-source remote sensing data, extract the improved double-season rice index characteristics, spectral band characteristics, texture characteristics, terrain characteristics, and polarization characteristics. The specific steps are as follows:

[0090] 2.1) Improvement of double-season rice index feature extraction

[0091] Considering that double-season rice can be planted twice in one year, this embodiment develops a new double-season rice index to capture the phenology of double-season rice based on the key phenology of double-season early rice and continuous late rice. Figure 5 As shown in the figure, the harmonic fitting results of the two indices kNDVI and LSWI from January to December are shown. It can be seen that both kNDVI and LSWI can capture the seasonal trend of double-season rice well. The difference between the two indices is large during the transplanting period of double-season early rice and late rice. Based on this difference, this embodiment designs four double-season rice phenological indices based on kNDVI and LSWI: double-season early rice index (ERPI), double-season late rice index (LPRI), normalized kNDVI index (NkNDVI EPR ), double-season late rice normalized kNDVI index (NkNDVI LPR ). These indices need to be calculated based on Landsat-8 / 9 data and Sentinel-2 data respectively.

[0092] The kNDVI and LSWI indices are existing indices. The calculation methods of kNDVI and LSWI are as follows:

[0093]

[0094]

[0095] Among them, Red, Nir and Swir1 represent the reflectance of red (Red), near infrared (Nir) and shortwave infrared 1 (Swir1) respectively.

[0096] Each type of optical image (Landsat 8, Landsat 9 and Sentinel-2) needs to be synthesized based on its own images in different months, and then the four new double-season rice indices are calculated respectively. The calculation method is as follows:

[0097]

[0098]

[0099]

[0100]

[0101] Among them, kNDVI1 and kNDVI2 represent the median composite kNDVI of double-season early rice transplanting period and maturity period, respectively; kNDVI3 and kNDVI4 represent the median composite kNDVI of double-season late rice transplanting period and maturity period, respectively; LSWI1 and LSWI2 represent the median composite LSWI of double-season early rice transplanting period and double-season late rice transplanting period, respectively.

[0102] 2.2) Extraction of spectral bands

[0103] Each optical image (Landsat 8, Landsat 9 and Sentinel-2) selected the red (Red), near infrared (Nir) and short wave infrared 1 (Swir1) of Landsat-8 / 9 and Sentinel-2 respectively. Since the three characteristic values ​​of optical images at different times are different, in this embodiment, for each optical image, the three spectral band characteristics of Red, Nir and Swir1, the three spectral band characteristic values ​​corresponding to the median synthetic image of the entire double-season rice planting cycle (March to November) are used.

[0104] 2.3) Extraction of texture features

[0105] In order to extract the texture features of double-season rice, monthly median synthetic images corresponding to the transplanting period of double-season rice early rice and double-season rice late rice were generated based on Sentinel-2 optical image data, and principal component analysis was performed respectively, and texture extraction was performed on the first principal component. The extracted textures include: entropy (ENT), angular second moment (ASM), and inverse distance difference (IDM). Therefore, entropy (ENT), angular second moment (ASM), and inverse distance difference (IDM) can be extracted from the first principal component of the monthly median synthetic image of the transplanting period of double-season rice early rice, while entropy (ENT), angular second moment (ASM), and inverse distance difference (IDM) can be extracted from the first principal component of the monthly median synthetic image of the transplanting period of double-season rice late rice. These texture features need to be used as inputs for subsequent classification models.

[0106] 2.4) Extraction of terrain features

[0107] The terrain features were extracted based on DEM, including slope, aspect and elevation.

[0108] 2.5) Extraction of polarization features

[0109] Based on the single co-polarization band of Sentinel-1: vertical transmission / vertical reception (VV), and the dual-band cross-polarization band: horizontal transmission / horizontal reception (VH), the ratio index (BVH) = VV / VH was calculated, and the ratio index was used as the polarization feature of the input classification model.

[0110] Step 3: Based on all the features extracted above, they are stacked and input into the trained random forest model to output the double-season rice planting area mask corresponding to the target area, which is the double-season rice spatial distribution map.

[0111] The random forest model needs to be trained in advance. During the training process, the final features used will be optimized by constructing and adjusting decision trees. In actual reasoning and application, all the constructed decision trees can be combined into a random forest classifier based on the trained random forest model. All the features extracted in the previous step are input into the trained random forest classifier, and new feature inputs can be further formed from the input features according to the trained decision tree combination, thereby generating a classification result mask corresponding to the target area.

[0112] In addition, in this embodiment, for the double-season rice spatial distribution map output by the random forest classifier, it is necessary to further use a 3×3 window majority filter to smooth and denoise it to reduce the salt and pepper effect of the classification result.

[0113] In this embodiment, the rice classification accuracy (RPMA) is used to verify the accuracy of the final classification result on the data set. The calculation method of the rice classification accuracy (RPMA) is as follows:

[0114]

[0115] Where UA represents the user accuracy of double-season rice classification, and PA represents the producer accuracy of double-season rice classification.

[0116] At the same time, in order to better compare the effectiveness of the double-season rice classification method proposed in the present invention, this embodiment conducts an ablation comparison experiment on multiple types of input features to show the impact of each on the final result. The final results are shown in Table 1.

[0117] Table 1. Accuracy evaluation of ablation experiments

[0118]

[0119] From the comparison of the accuracy of different combinations in Table 1, it can be seen that compared with the combination of spectral bands, texture, terrain and polarization features (II), the addition of polarization (III) and improved double-season rice index (IV) have improved the accuracy of double-season rice extraction to a certain extent, which shows that the use of multi-source data can improve the classification accuracy of double-season rice. The highest accuracy is the combination of spectral bands, texture, terrain, polarization and new double-season rice index, with RPMA reaching 94.17%.

[0120] At the same time, this embodiment further conducts the above-mentioned ablation experiment on Google Earth historical images. The classification results of combination I, which only rely on spectral bands and textures, are relatively fragmented and have overestimation. However, with the gradual addition of four multi-source features, namely, terrain, polarization, and improved double-season rice index, the boundaries of the fields are more distinct, especially the addition of improved double-season rice index, which significantly improves the classification results of double-season rice, which are highly consistent with the actual situation in terms of field boundaries and spatial positions, further proving the effectiveness of the double-season rice classification method proposed in this aspect.

[0121] In addition, since the original classification results still have the "salt and pepper effect", there are many fragmented pixels, which are often misclassified results caused by image noise. Therefore, this embodiment performs mode filtering on the results. The present invention introduces the method of removing small pixels by mode filtering, which can effectively remove fragmented pixels.

[0122] The above-described embodiments are only some preferred implementations of the present invention, but are not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A double-season rice classification method based on multi-source remote sensing data and improved index, characterized in that: include: S1. Acquire multi-source remote sensing data during the entire double-season rice planting period in the target area, and pre-process them respectively; the multi-source remote sensing data include DEM data, Landsat-8 optical images, Landsat-9 optical images, Sentinel-2 optical images, and Sentinel-1 synthetic aperture radar data; S2. Based on the pre-processed multi-source remote sensing data, extract the improved double-season rice index characteristics and spectral band characteristics from each optical image, extract texture characteristics from the Sentinel-2 optical image, extract terrain characteristics from the DEM data, and extract polarization characteristics from the Sentinel-1 synthetic aperture radar data; There are 4 improved double-season rice index characteristics in total, the first index characteristic is the sum of the difference ratios between the median synthetic kNDVI index of the double-season early rice at maturity and the median synthetic kNDVI index of the double-season early rice at transplanting period, the second index characteristic is the sum of the difference ratios between the median synthetic kNDVI index of the double-season late rice at maturity and the median synthetic kNDVI index of the double-season late rice at transplanting period, the third index characteristic is the sum of the difference ratios between the median synthetic LSWI index of the double-season early rice at transplanting period and the median synthetic kNDVI index of the double-season early rice at transplanting period, and the fourth index characteristic is the sum of the difference ratios between the median synthetic LSWI index of the double-season late rice at transplanting period and the median synthetic kNDVI index of the double-season late rice at transplanting period; S3. Input all the features extracted in S2 into the trained random forest model, and output the spatial distribution map of double-season rice corresponding to the target area.

2. The double-season rice classification method based on multi-source remote sensing data and improved index according to claim 1, characterized in that: When performing the above preprocessing on multi-source remote sensing data, all three optical images need to be screened and de-clouded first. Landsat-8 and Landsat-9 optical images use the QA band to de-cloud, Sentinel-2 optical images use the MSK_CLDPRB band to de-cloud, and Sentinel-1 synthetic aperture radar data need to be converted to natural values ​​and have their boundaries removed first. Then, the DEM data, the three optical images, and the Sentinel-1 synthetic aperture radar data are cropped according to the outline of the target area, and are aligned based on the Sentinel-2 optical image, and finally resampled to the same resolution.

3. The double-season rice classification method based on multi-source remote sensing data and improved index according to claim 1, characterized in that: The spectral band features extracted from each optical image are the red band, near infrared band and shortwave infrared 1 band in the median synthetic image of the entire double-season rice planting cycle.

4. The double-season rice classification method based on multi-source remote sensing data and improved index according to claim 1, characterized in that: The texture features extracted from the Sentinel-2 optical image were entropy, angular second moment, and inverse disparity extracted from the first principal component of the median synthetic image during the double-season rice transplanting period.

5. The double-season rice classification method based on multi-source remote sensing data and improved index according to claim 1, characterized in that: The terrain features extracted from DEM data include slope, aspect and elevation.

6. The double-season rice classification method based on multi-source remote sensing data and improved index according to claim 1, characterized in that: The polarimetric features extracted from the Sentinel-1 synthetic aperture radar data are the ratios of the VV band and the VH band in the median synthetic image during the double-season rice transplanting period.

7. The double-season rice classification method based on multi-source remote sensing data and improved index according to claim 1, characterized in that: The double-season rice spatial distribution map needs to be smoothed and denoised by a majority filter, and the final double-season rice spatial distribution map is obtained after eliminating isolated points.

8. A double-season rice classification system based on multi-source remote sensing data and improved index, characterized in that: include: A data acquisition module is used to acquire multi-source remote sensing data in the entire double-season rice planting period in the target area and perform preprocessing respectively, wherein the multi-source remote sensing data includes DEM data, Landsat-8 optical images, Landsat-9 optical images, Sentinel-2 optical images and Sentinel-1 synthetic aperture radar data; A feature extraction module is used to extract improved double-season rice index features and spectral band features from each optical image based on the preprocessed multi-source remote sensing data, extract texture features from Sentinel-2 optical images, extract terrain features from DEM data, and extract polarization features from Sentinel-1 synthetic aperture radar data; There are 4 improved double-season rice index characteristics in total, the first index characteristic is the sum of the difference ratios between the median synthetic kNDVI index of the double-season early rice at maturity and the median synthetic kNDVI index of the double-season early rice at transplanting period, the second index characteristic is the sum of the difference ratios between the median synthetic kNDVI index of the double-season late rice at maturity and the median synthetic kNDVI index of the double-season late rice at transplanting period, the third index characteristic is the sum of the difference ratios between the median synthetic LSWI index of the double-season early rice at transplanting period and the median synthetic kNDVI index of the double-season early rice at transplanting period, and the fourth index characteristic is the sum of the difference ratios between the median synthetic LSWI index of the double-season late rice at transplanting period and the median synthetic kNDVI index of the double-season late rice at transplanting period; The classification module is used to input all the features extracted in the feature extraction module into the trained random forest model and output the spatial distribution map of double-season rice corresponding to the target area.

9. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the double-season rice classification method based on multi-source remote sensing data and improved index as described in any one of claims 1 to 7 can be implemented.

10. A computer electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to implement the double-season rice classification method based on multi-source remote sensing data and improved index as described in any one of claims 1 to 7 when executing the computer program.

Images