A Rice Identification Method Based on the Fusion of Optical Imagery and SAR Time-Series Data

Abstract

This invention discloses a rice identification method based on the fusion of optical imagery and SAR time-series data. The method is as follows: 1. Extract single-scene optical images and SAR time-series data of the measured area; and perform coarse classification of the measured area based on the single-scene optical images. 2. Perform multi-scale image segmentation based on the coarse classification results, and extract plot boundary information according to the optimal segmentation parameters. 3. Extract SAR time-series feature data through the plot segmentation boundary results to obtain object-oriented SAR time-series feature data. 4. Complete rice area identification and classification through machine learning. This invention relaxes the restrictions on the temporal phase of optical data by using single-temporal optical images as the data source for multi-scale image segmentation methods in areas lacking sufficient clear-sky optical images. Based on object-oriented scale microwave imagery, it can fully overcome the weaknesses of fragmented plots and salt-and-pepper noise at the pixel level, thereby enabling accurate extraction of rice areas.

Description

Technical Field

[0001] This invention belongs to the field of intelligent remote sensing and detection technology for agricultural big data, specifically involving a method for extracting rice areas from SAR time-series data of fused optical image plot boundary information for multi-cloud and regional applications. Background Technology

[0002] As an important global food crop, information on the planting area and distribution of rice is of great significance for rice growth monitoring and planting management, rice disaster monitoring and early warning, rice yield estimation, and macro-management of rice production.

[0003] Currently, optical remote sensing satellite data is the primary data source for rice area monitoring. While multi-temporal optical imagery can extract rice planting area with high accuracy, the passive imaging nature of optical data makes it susceptible to weather conditions such as clouds and rain, hindering the stable acquisition of the required multi-scene optical image datasets. Furthermore, the revisit cycle of medium- and high-resolution optical satellites is often relatively long, making it more difficult to acquire clear-sky effective optical images of regions with complex climates. This results in a lack of image data for key phenological stages of rice, significantly restricting the operational application of rice remote sensing mapping technology. However, using only a single optical image for rice area extraction cannot fully capture information on rice growth changes, making it difficult to distinguish from other green non-rice vegetation, leading to low classification accuracy. Currently used SAR imagery is unaffected by weather and time, and can stably acquire continuous image time series. However, its significant salt-and-pepper noise and lower signal-to-noise ratio compared to optical imagery pose significant challenges for identifying rice paddies in small and fragmented areas (especially identifying field boundaries).

[0004] In cloudy and rainy areas, combining single-phase optical imagery with SAR time-series imagery is a commonly and stably acquired remote sensing image combination. Optical imagery contains grayscale information across multiple bands, facilitating target identification and classification, and is easier to segment statistically due to its high information content. SAR data provides backscattering information of surface objects under different polarization modes and captures their physical characteristics. It offers all-weather, day-night observation capabilities, making it beneficial for obtaining stable and continuous time-series information, and providing a sufficient data source for extracting the temporal variation area of ​​rice paddies. Summary of the Invention

[0005] The purpose of this invention is to provide a method for extracting rice area from SAR time-series data that integrates plot boundary information from optical images for cloudy and rainy areas. This method uses plot boundary information from optical images for SAR data, relaxes the restrictions on temporal data from optical images, reduces the significant salt-and-pepper noise present in SAR data, and better addresses the problem of identifying rice paddies in small and fragmented plots. It efficiently and accurately obtains rice planting area while making full use of existing data sources.

[0006] The specific steps of this invention are as follows:

[0007] Step 1: Extract single-scene optical images and SAR time-series data of the area to be measured; and perform coarse classification of the area to be measured based on the single-scene optical images, dividing different locations in the area to rice paddies, water bodies, non-rice vegetation and other land surfaces.

[0008] Step 2: Extract land parcel boundary information based on coarse classification results.

[0009] Based on the coarse classification results from step 1, areas classified as other land surfaces are removed. Multi-scale image segmentation is then performed on the areas classified as rice, water bodies, and non-rice vegetation to obtain the plot segmentation boundaries. Specifically, multi-scale image segmentation uses Band 3, Band 4, ..., Band 8 images and NDVI (Normalized Difference Vegetation Index) images from a single-scene optical imagery source. The segmentation weights for Band 3, Band 4, ..., Band 8 images and NDVI images are 1:2:2:1:1:2:2, respectively. The segmentation parameters selected are size factor, shape factor, and compactness factor. The size factor is set to 35–40, the shape factor to 0.2–0.3, and the compactness factor to 0.5–0.7.

[0010] Step 3: Process the SAR time series feature data using the results of the land parcel segmentation boundary, calculate the mean value of the land parcel feature values ​​based on the boundary, and obtain object-oriented SAR time series feature data.

[0011] Step 4: Use machine learning to identify and classify the rice paddy area.

[0012] Extract target features from SAR time-series feature data; target features include several polarization features and / or several texture features. Input the target features into a trained machine learning algorithm to obtain the location, shape, and area of ​​rice planting in the measured area.

[0013] Preferably, the coarse classification in step 1 is as follows: First, extract spectral band features and vegetation index features from the single-scene optical image. Spectral band features include Band 4 and Band 11 images from the single-scene optical image; vegetation index features include Enhanced Vegetation Index (EVI), Soil Adjusted Vegetation Index (SAVI), and Modified Normalized Difference in Water (MNDWI). Then, input the spectral band features and vegetation index features into a support vector machine for coarse classification; the coarse classification initially divides different locations in the measured area into four categories: rice, water bodies, non-rice vegetation, and other surfaces.

[0014] Preferably, the support vector machine described in step 1 is trained and validated; the training and validation samples are from ground survey data and visual interpretation data based on images; the total number of samples is greater than or equal to 1000; all samples are selected as training samples and validation samples in a 2:3 ratio.

[0015] Preferably, in step 2, the segmentation weights for the Band 3, Band 4, ..., Band 8 images and the NDVI images are 1:2:2:1:1:2:2, respectively; the size factor is 35–40, the shape factor is 0.2–0.3, and the compactness factor is 0.5–0.7. Preferably, the size factor is 35, the shape factor is 0.3, and the compactness factor is 0.6.

[0016] As a preferred method, the method for determining the value of the segmentation parameter in step 2 is as follows:

[0017] (1) Set several sets of segmentation parameters. Perform multi-scale image segmentation on the same or different coarse classification results according to each set of segmentation parameters.

[0018] (2) Calculate the area consistency index (ACI) for each group of segmentation parameters. i centroid distance qLoc i Shape error SE i as follows:

[0019]

[0020] qLoc i =dist(centroid(SO) i ),centroid(SO i ))

[0021] SE i =|Asr(RO) i )-Asr(RO i )|

[0022] Among them, SO i The segmentation result obtained from the i-th set of segmentation parameters; RO i This represents the reference result corresponding to the i-th group of segmentation parameters; the reference result is the actual area occupied by the rice part; dist(·) represents distance operation; centroid(·) represents centroid extraction operation; Asr(·) represents aspect ratio extraction operation. i = 1, 2, ..., n. n is the number of segmentation parameter groups set.

[0023] Calculate the total evaluation score (CCS) for each group of segmentation parameters. i =ACI i +qLoc i+SE i The total evaluation score (CCS) is taken. i The highest set of segmentation parameters is used as the final segmentation parameters for multi-scale image segmentation.

[0024] As a preferred option, the process of setting several sets of segmentation parameters is as follows: the value range of the size factor is set to 35–60, and its step size is set to 5; the value range of the shape factor is set to 0.1–0.6, and its step size is set to 0.1; the value range of the compactness factor is set to 0.2–0.7, and its step size is set to 0.1. Six candidate values ​​are obtained for each of the size factor, shape factor, and compactness factor; the candidate values ​​of the size factor, shape factor, and compactness factor are combined according to the orthogonal experimental method to obtain 36 sets of segmentation parameters.

[0025] Preferably, the machine learning algorithm described in step 4 uses a classification regression tree or a support vector machine.

[0026] Preferably, the target features in step 4 include polarization features and texture features. The polarization features include 0616_VH, 0710_VV, 0722_VV, 0815_VH, ​​and 0827_VV. The texture features include 0604_VH_B1, 0604_VV_B1, 0616_VH_B3, 0616_VV_B7, 0628_VH_B6, 0628_VV_B1, 0710_VV_B2, 0710_VV_B5, 0710_VV_B6, 0722_VH_B2, 0722_VH_B3, 0722_VH_B5, 0722_VV_B2, 0722_VV_B5, 0722_VV_B6, 0803_VH_B2, and 0803_VH_B2. _VH_B3, 0803_VH_B4, 0803_VH_B5, 0803_VH_B6, 0803_VH_B8, 0803_VV_B1, 0815_VH_B5, 0827_VV_B7, 0827_VV_B8; where B1 to B8 represent Mean, Variance, Homogeneity, Contrast, Dissimilarity, Entropy, Second Moment, and Correlation, respectively.

[0027] Preferably, the target feature in step 4 is one polarization feature or texture feature selected from each of the four reproductive period grouping features; at least one polarization feature is selected from the four reproductive period grouping features.

[0028] Preferably, the target features in step 4 are four polarization features, specifically 0616_VH, 0722_VV, 0815_VH, ​​and 0827_VV.

[0029] Preferably, the target features in step 4 use polarization features and texture features. Specifically, the polarization features are 0722_VV and 0827_VV; the texture features are 0615_VH_B3 and 0803_VH_B2; B2 and B3 are Variance and Homogeneity, respectively.

[0030] The beneficial effects of this invention are:

[0031] 1. In areas lacking sufficient clear-sky optical imagery, this invention can acquire rice paddies using single-scene medium-high resolution (10m) optical satellite imagery and extract rich phenological information features contained in stable and continuous time-series SAR images. By using the coarse classification results of single-scene optical imagery and extracting the boundary information of the paddies through multi-scale image segmentation methods, object-oriented SAR time-series feature data is obtained, thereby enabling accurate extraction of rice paddy areas and realizing high-precision operational mapping.

[0032] 2. When using optical imaging and multi-scale segmentation technology to obtain the boundaries of land parcels, this invention can quantitatively determine the optimal segmentation parameters for multi-scale segmentation and evaluate the degree of agreement between the segmentation results and the natural boundaries of land parcels, thereby improving the consistency between the segmentation results and the natural boundaries of paddy fields.

[0033] 3. This invention further examines its adaptability in practical operational scenarios by randomly selecting microwave images from key time phases. The results show that this method has strong potential for operationalization in rice area extraction. Attached Figure Description

[0034] Figure 1 This is a technical roadmap for rice regional extraction according to the present invention;

[0035] Figure 2 This is a schematic diagram of the multi-scale image segmentation process in this invention;

[0036] Figure 3 Image segmentation results based on different size factors;

[0037] Figure 4 The image shows the rice extraction results of this invention (part a is a local result of rice extraction using SAR time-series data with fused optical imagery and plot boundary information; part b is a local result of rice extraction using multi-temporal optical images; part c is a local result of rice extraction based on pixel level using SAR time-series microwave images; parts a, b, and c target the same location). Detailed Implementation

[0038] The present invention will be further described below with reference to the accompanying drawings.

[0039] A rice identification method based on the fusion of optical imagery and SAR time-series data is presented. This method uses single-scene optical imagery combined with SAR time-series data for efficient extraction of rice-growing areas, suitable for environments where complete multi-scene optical imagery is difficult to obtain, such as cloudy and rainy areas. In this embodiment, the experimental area for verifying the effectiveness of this rice identification method is Yangzhou, a typical cloudy and rainy rice-growing region. This method fully combines the advantages of optical and microwave imagery in crop extraction. It extracts plot boundaries based on single-phase Sentinel-2 optical imagery, using natural plot boundary information as the analysis unit for Sentinel-1 microwave imagery, and then uses the microwave imagery for phenological information extraction. The strategy of constructing rice area extraction using SAR time-series data that fuses optical information reduces dependence on specific acquisition phases of optical imagery, allowing for stable acquisition of both optical and microwave imagery. Simultaneously, it leverages the advantages of both types of imagery to achieve high-precision and refined operational extraction of rice.

[0040] like Figure 1 As shown, the specific steps of this rice identification method are as follows:

[0041] Step 1: First, based on the stable acquisition of single-temporal Sentinel-2 optical images, the SVM classification algorithm is used to perform preliminary classification of four types of land features: other land surfaces, non-rice vegetation, water bodies, and rice, to obtain coarse classification results for the four types of land features.

[0042] Step 2: Considering that the speckle noise present in microwave remote sensing images affects the analysis and interpretation of microwave images, this invention, based on the coarse classification results of single-temporal optical images, further uses a multi-scale segmentation method to obtain the natural plot boundaries of the classified water bodies, non-rice vegetation, and rice parts, as the basic analysis unit for object-oriented analysis of microwave images.

[0043] Multiscale segmentation is a method that can generate image polygons (objects) of arbitrary scale and similar attribute information, which is a necessary prerequisite for classification, recognition and information extraction.

[0044] In multi-scale segmentation, the selection of segmentation parameters (including size, shape, and compactness) directly affects the quality of the segmentation results, and further affects the final classification accuracy. Among all segmentation parameters, the most important is the size parameter; an excessively large size will lead to undersegmentation, while a small size will result in oversegmentation. The shape parameter determines the influence of spectral values ​​and shape on object formation, while the compactness parameter determines the compactness of the segmented object boundaries.

[0045] The overall segmentation process is as follows: Figure 2As shown. The quantitative selection of optimal segmentation parameters for multi-scale segmentation has always been a key point in object-oriented classification. In order to make the boundaries of the segmented land parcels match the natural paddy field parcels as closely as possible, this invention proposes a comprehensive evaluation method that integrates area consistency, centroid distance, and shape error index to quantitatively evaluate the segmentation effect, based on the characteristics of each band (Band3, Band4, ..., Band8) of Sentinel-2 optical imagery, the normalized difference vegetation index (NDVI) feature, and image data after removing other land surfaces from the coarse classification results.

[0046] Through a three-factor, six-level orthogonal experiment involving size factor, shape factor, and compactness factor, a multi-scale segmentation method was used to divide farmland plots in the study area. The reference results were compared with the segmentation results, and three evaluation indicators—area consistency, centroid distance, and shape error—were calculated.

[0047] Experiments were conducted on 36 sets of parameters. The final segmentation score was obtained by summing the scores of three evaluation indicators, thereby determining the optimal plot segmentation parameters for the study area (size factor: 35, shape factor: 0.3, compactness factor: 0.6).

[0048] Figure 3 The segmentation results are shown at different sizes (with the other two parameters remaining constant: shape factor: 0.3 and compactness factor: 0.6).

[0049] The size factor has a significant impact on the segmentation results. When the size factor is set small (size factor: 15), the overall segmentation result shows significant oversegmentation, resulting in fragmented segmented objects. Conversely, when the size factor is set large (size factor: 60), the overall segmentation result shows significant undersegmentation, and the resulting boundary may include multiple plots. A size factor of 35 results in segmentation results that closely match the actual plots.

[0050] Step 3: Perform image segmentation based on the multi-scale segmentation parameter results to obtain parcel boundary information. Use the parcel boundary results to process the SAR time-series image, completing the corresponding operations through zonal statistics.

[0051] Based on SAR time-series image data of rice from June to August throughout its entire growth period, the polarization features and gray-scale co-occurrence texture features of microwave images are selected as the preferred feature set for rice area extraction in this invention (as described in the invention description).

[0052] Step 4: In this invention, the land cover classification adopts two representative machine learning algorithms: Classification Regression Tree (CART) and Support Vector Machine (SVM). Both methods are widely used in remote sensing image classification and rice area monitoring.

[0053] In this invention, the accuracy evaluation of rice classification results adopts indicators such as overall classification accuracy (OA), user accuracy (UA), producer accuracy (PA), and Kappa coefficient. All of these indicators can be calculated based on the confusion matrix of the classification results.

[0054] Based on the method of this invention, a multi-temporal optical image method and a rice area extraction scheme using only SAR time-series image method are added, and the rice extraction accuracy under the three classification methods is compared.

[0055] The method proposed in this invention is compared with methods using multi-temporal optical imagery and methods using only SAR time-series imagery. Under the SVM classification algorithm, the method of this invention improves OA, PA, UA, and Kappa by 9.75%, 19.4%, 18.74%, and 0.16%, respectively, compared with the method using only SAR time-series imagery for rice extraction. Under the CART decision tree algorithm, the method of this invention improves OA, PA, UA, and Kappa by 11.69%, 12.41%, 21.61%, and 0.19%, respectively, compared with the method using only SAR time-series imagery for rice extraction.

[0056] Under both classification algorithms, the method of this invention shows a significant improvement in accuracy across all aspects compared to the method using only time-series microwave imagery, which also demonstrates that the addition of plot boundary information from optical imagery plays an important role in improving the accurate identification of rice in time-series SAR imagery.

[0057] Furthermore, the accuracy of the method of the present invention is close to that of rice identification in multi-temporal optical images, which shows that the method provided by the present invention can stably and effectively identify rice even in the absence of multi-scene optical data.

[0058] Figure 4 Part a in the image is a local rice field result map extracted from SAR time-series data based on fused optical imagery of land parcel boundaries. Figure 4 Part b in the image is a local result map of rice based on multi-temporal optical imagery; Figure 4 Part c in the image is a local rice field image based on pixel-level results using SAR time-series microwave imagery.

[0059] from Figure 4 As can be seen from the above, compared with the simple SAR time-series microwave image method, the method of the present invention can better suppress the salt-and-pepper effect of the classification results, and the classification results are more consistent with the land parcels.

[0060] Meanwhile, the method proposed in this invention demonstrates the advantages of microwave image fusion with optical boundaries, showing that the accuracy after adding optical data is higher than that of using time-series microwave data alone, thus verifying the correctness of the invention's approach. On the other hand, it shows that object-oriented classification methods have a wider range of application prospects compared to pixel-oriented methods.

[0061] The results show that the method proposed in this invention has the potential for commercial application and can effectively extract nutrients from rice grown in cloudy and rainy areas.

[0062] To further test the adaptability to operational scenarios, the SAR time-series optimized feature set was grouped according to four key growth periods of rice: transplanting, tillering, jointing, and heading. Each group contains several features, as shown in Table 1.

[0063] Table 1

[0064]

[0065] To compare the roles of polarization features and texture features in rice extraction, three feature sets were selected: full polarization feature set, full texture feature set, and polarization + texture feature set.

[0066] Method for selecting the full polarization feature set: In the four reproductive period group features, one polarization feature is randomly selected from each group to form a set of four polarization features.

[0067] The selection method for the full texture feature set is the same as that for the full polarization feature set.

[0068] Polarization + texture feature set selection method: In the four reproductive stage group features, one feature is randomly selected from each group to form a set of 4 features, which includes at least one polarization feature and one texture feature.

[0069] In this embodiment, the selected full polarization feature sets are 0616_VH, 0722_VV, 0815_VH, ​​and 0827_VV, and the full texture feature sets are 0604_VV_B1, 0710_VV_B5, 0803_VH_B6, and 0827_VV_B7. The polarization + texture feature sets are 0722_VV, 0827_VV, 0615_VH_B3, and 0803_VH_B2; B1, B2, B3, B5, B6, and B7 represent Mean, Variance, Homogeneity, Dissimilarity, Entropy, and SecondMoment, respectively.

[0070] Using the SVM classification algorithm, the rice extraction accuracy under three different feature combinations was obtained. The fully polarized feature set (PA 85.10%, UA 87.43%) and the polarization + texture feature set (PA 85.60%, UA 90.02%) both achieved high accuracy, while the fully texture feature set had lower classification accuracy (PA 72.71%, UA 60.20%). This indicates that polarization features are more beneficial for rice identification than texture features.

[0071] Comparing the average user accuracy and average producer accuracy of rice extraction based on the polarization + texture feature set (PA 85.60%, UA 90.02%) with that using the full feature set (PA 92.83%, UA 90.48%) in a practical operational scenario after simplifying feature information, it can be found that although the accuracy of the polarization + texture feature set does not exceed that of the full feature set, the difference is not significant. This further indicates that in actual operational work, relatively ideal classification accuracy can be achieved even without using all temporal SAR data, relying only on SAR images of rice during a few key growth stages. This simplifies the data and computational resource investment of the method and makes it suitable for operational scenarios.

Claims

1. A rice identification method based on the fusion of optical imagery and SAR time-series data, characterized in that: Step 1: Extract single-scene optical images and SAR time-series data of the area to be measured; and perform coarse classification of the area to be measured based on the single-scene optical images, dividing different locations in the area to rice paddies, water bodies, non-rice vegetation and other land surfaces. Step 2: Extract land parcel boundary information based on the coarse classification results; Based on the coarse classification results in step 1, areas classified as other land surfaces are removed; multi-scale image segmentation is performed on areas classified as rice, water bodies and non-rice vegetation in the coarse classification results to obtain the plot segmentation boundary results; the multi-scale image segmentation specifically uses Band3, Band4, ..., Band8 images and normalized vegetation index images from single-scene optical images for segmentation; the segmentation parameters selected are size factor, shape factor and compactness factor; The method for determining the values ​​of the segmentation parameters is as follows: (1) Set several sets of segmentation parameters; perform multi-scale image segmentation on the same or different coarse classification results according to each set of segmentation parameters; The process of setting several sets of segmentation parameters is as follows: the value range of the size factor is set to 35-60, and its step size is set to 5; the value range of the shape factor is set to 0.1-0.6, and its step size is set to 0.1; the value range of the compactness factor is set to 0.2-0.7, and its step size is set to 0.1; six candidate values ​​are obtained for each of the size factor, shape factor, and compactness factor; the candidate values ​​of the size factor, shape factor, and compactness factor are combined according to the orthogonal experimental method to obtain 36 sets of segmentation parameters. (2) Calculate the area consistency index (ACI) for each group of segmentation parameters. i centroid distance qLoc i Shape error SE i as follows: qLoc i =dist(centroid(SO i ),centroid(RO i )) SE i =|Asr(SO i )-Asr(RO i , Among them, SO i The segmentation result obtained from the i-th set of segmentation parameters; RO i This is the reference result corresponding to the i-th group of segmentation parameters; the reference result is the actual area occupied by the rice part; dist(·) represents distance operation; centroid(·) represents centroid extraction operation; Asr(·) represents aspect ratio extraction operation; i = 1, 2, ..., n; n is the number of segmentation parameter groups set; Calculate the total evaluation score (CCS) for each group of segmentation parameters. i =ACI i +qLoc i +SE i The total evaluation score (CCS) is taken. i The highest set of segmentation parameters is used as the final segmentation parameters for multi-scale image segmentation. Step 3: Process the SAR time series feature data using the land parcel segmentation boundary results, calculate the mean value of the land parcel feature values ​​based on the boundary, and obtain object-oriented SAR time series feature data; Step 4: Use machine learning to identify and classify rice in the tested area; Extract target features from SAR time-series feature data; target features include several polarization features and several texture features; target features are selected from each of the four growth period group features, one polarization feature or one texture feature; at least one polarization feature is among the four growth period group features; input the target features into a trained machine learning algorithm to obtain the location, shape and area of ​​rice planting in the measured area.

2. The rice identification method based on the fusion of optical imagery and SAR time-series data according to claim 1, characterized in that: The coarse classification in step 1 is as follows: First, extract the spectral band features and vegetation index features from the single-scene optical image; the spectral band features include the Band 4 and Band 11 band images of the single-scene optical image; the vegetation index features include the enhanced vegetation index, the soil-adjusted vegetation index, and the corrected normalized water body difference index; then, input the spectral band features and vegetation index features into the support vector machine for coarse classification; the coarse classification initially divides different locations in the measured area into four categories, namely rice, water bodies, non-rice vegetation, and other land surfaces.

3. The rice identification method based on the fusion of optical imagery and SAR time-series data according to claim 1, characterized in that: In step 2, the segmentation weights of the Band3, Band4, ..., Band8 images and the NDVI images are 1:2:2:1:1:2:2, respectively; the size factor is 35-40, the shape factor is 0.2-0.3, and the compactness factor is 0.5-0.

7.

4. The rice identification method based on the fusion of optical imagery and SAR time-series data according to claim 1, characterized in that: The machine learning algorithm described in step 4 uses either a classification regression tree or a support vector machine.

5. The rice identification method based on the fusion of optical imagery and SAR time-series data according to claim 1, characterized in that: The polarization features include 0616_VH, 0710_VV, 0722_VV, 0815_VH, ​​and 0827_VV; the texture features include 0604_VH_B1, 0604_VV_B1, 0616_VH_B3, 0616_VV_B7, 0628_VH_B6, 0628_VV_B1, 0710_VV_B2, 0710_VV_B5, 0710_VV_B6, 0722_VH_B2, 0722_VH_B3, 0722_VH_B5, 0722_VV_B2, and 072... 2_VV_B5, 0722_VV_B6, 0803_VH_B2, 0803_VH_B3, 0803_VH_B4, 0803_VH_B5, 0803_VH_B6, 0803_VH_B8, 0803_VV_B1, 0815_VH_B5, 0827_VV_B7, 0827_VV_B8; where B1 to B8 represent Mean, Variance, Homogeneity, Contrast, Dissimilarity, Entropy, Second Moment, and Correlation, respectively.

6. The rice identification method based on the fusion of optical imagery and SAR time-series data according to claim 1, characterized in that: The target features in step 4 are four polarization features, specifically 0616_VH, 0722_VV, 0815_VH, ​​and 0827_VV.

7. The rice identification method based on the fusion of optical imagery and SAR time-series data according to claim 1, characterized in that: The target features in step 4 use polarization features and texture features; the polarization features are specifically 0722_VV and 0827_VV. The specific texture features are 0615_VH_B3 and 0803_VH_B2; B2 and B3 are Variance and Homogeneity, respectively.

Images