Combined Multi-Time Series SAR Imagery and InSAR Technology for Rice Identiﬁcation in Cloudy Regions

: The use of remote sensing technology to monitor farmland is currently the mainstream method for crop research. However, in cloudy and misty regions, the use of optical remote sensing image is limited. Synthetic aperture radar (SAR) technology has many advantages, including high resolution, multi-mode, and multi-polarization. Moreover, it can penetrate clouds and mists, can be used for all-weather and all-time Earth observation, and is sensitive to the shape of ground objects. Therefore, it is widely used in agricultural monitoring. In this study, the polarization backscattering coefﬁcient on time-series SAR images during the rice-growing period was analyzed. The rice identiﬁcation results and accuracy of InSAR technology were compared with those of three schemes (single-time-phase SAR, multi-time-phase SAR, and combination of multi-time-phase SAR and InSAR). Results show that VV and VH polarization coherence coefﬁcients can well distinguish artiﬁcial buildings. In particular, VV polarization coherence coefﬁcients can well distinguish rice from water and vegetation in August and September, whereas VH polarization coherence coefﬁcients can well distinguish rice from water and vegetation in August and October. The rice identiﬁcation accuracy of single-time series Sentinel-1 SAR image (78%) is lower than that of multi-time series SAR image combined with InSAR technology (81%). In this study, Guanghan City, a cloudy region, was used as the study site, and a good veriﬁcation result was obtained.


Introduction
Satellite remote sensing has become an important means of crop monitoring. Optical remote sensing can be used to monitor relevant parameters in an all-around manner during the rice-growing period. To classify agricultural crops by remote sensing, a discriminant function is established on the basis of crop characteristics, including brightness, hue, position, texture, and structure. However, optical remote sensing data are often affected by cloud, rain, fog, and other bad weather conditions in practical applications. For example, cloud cover and frequent rainfall in mountainous or basin areas make it difficult to find a suitable image for studying the rice-growing season. In addition, the optical image may contain similar objects with different spectra and different objects with similar spectra. Therefore, the use of optical remote sensing data in the identification and monitoring of rice and other crops will be greatly limited.
Synthetic aperture radar (SAR) can make up for the deficiency of optical remote sensing due to its high resolution; multi-mode; multi-polarization direction; all-weather acquisition; cloud cover penetration; all-time Earth observation; and sensitivity to the shape, material, moisture, and surface roughness of ground objects [1][2][3]. For SAR imagery, the time-phase change in backscattering characteristics of rice is much greater than that of present, InSAR technology has been widely used in the monitoring, identification, and classification of forests, snow, wetlands, soil, and cities, as well as in digital elevation model (DEM) production generation, deformation measurement, earthquake observation and measurement, and 3D reconstruction [21][22][23].
In cloudy regions, the use of optical remote sensing image in the monitoring and identification of rice and other crops is greatly restricted. However, the traditional method is mainly used in single-and multi-time-phase radar images; that is, the difference in the backscattering strength among ground objects is based to distinguish rice from other ground objects while overcoming the problems of the SAR image itself. Polarimetric InSAR technology is used for rice identification. It is very sensitive to the spatial distribution, height, shape, and direction of surface vegetation scatterers. Therefore, it can more efficiently extract the rice-planting areas compared with other technologies.
Although it is not a precedent to monitor rice based on InSAR technology, some studies only calculated and adopted the theoretical coherence diagram, namely, the coherence diagram of the two most different time phases. Based on the time-series polarimetric interference radar, combined with the change of backscattering strength of rice and coherence coefficients, this study created pairs across time, performed rice identification using a time-series Sentinel-1 and a time-series Sentinel-1 scheme incorporating the polarization coherence coefficient, and used high-resolution Sentinel-2 optical images as the result validation. The rice-planting areas were extracted by using the object-oriented classification method, which provides a reference for the monitoring and identification of rice and other crops in cloudy regions.

Study Area
Guanghan is located in the core area of Chengdu Plain in Sichuan Basin, with a total area of 538 km 2 , and is gently inclined from northwest to southeast in the topography. It mainly features plains, combined with low hills, mountains, and flat dams. Among them, the hills are mainly concentrated in the south, and the mountains are mainly in the northeast. The average altitude is around 450 m, with the lowest value of 442 m and the highest value of 734 m in the east. Guanghan (Figure 1) is located in subtropical monsoon climate zone. It is mainly characterized by abundant precipitation, with annual precipitation of more than 800 mm, less rainfall in spring, and more rainstorm in summer. Affected by basin effect and climate, it is cloudy and foggy. Rice has one or two growth seasons annually, and it is transplanted in May, matures in mid-August, and is harvested in mid-September. According to the widely used Biologische Bundesanstalt, Bundessortenamt und CHemische scale (a German scale used to identify the phenological development stages of cereals, BBCH), the phenological stage of rice growth is shown in Table 1. We determined the rice-planting area according to the crop growth terminal, and the whole area in Chengdu Plain is mainly rice.

Methodology
Based on the time-series SAR images of rice-growing season, after multi-looking processing, registration, polarization filtering, and geographic coding, the polarization postscattering coefficient of typical ground such as rice, artificial building, water, and vegetation was extracted by sample point. After combining with the InSAR technology, the baseline estimation, interference generation, adaptive filtering, and coherence calculations, the VV and VH polarization coherence coefficients of the SAR images were obtained. The polarization coherence coefficient of typical samples such as rice, artificial building, water, and vegetation was extracted from sample data, and then the variation of the time-series coherence coefficient of typical ground objects was analyzed. Selecting the appropriate time-series SAR image polarization information and the coherence coefficient, the contribution of coherence coefficient to time-series SAR image was studied. How to select the appropriate polarization information and the coherence coefficient to extract the riceplanting area information was the key point and difficulty of this study.

Methodology
Based on the time-series SAR images of rice-growing season, after multi-looking processing, registration, polarization filtering, and geographic coding, the polarization post-scattering coefficient of typical ground such as rice, artificial building, water, and vegetation was extracted by sample point. After combining with the InSAR technology, the baseline estimation, interference generation, adaptive filtering, and coherence calculations, the VV and VH polarization coherence coefficients of the SAR images were obtained. The polarization coherence coefficient of typical samples such as rice, artificial building, water, and vegetation was extracted from sample data, and then the variation of the time-series coherence coefficient of typical ground objects was analyzed. Selecting the appropriate time-series SAR image polarization information and the coherence coefficient, the contribution of coherence coefficient to time-series SAR image was studied. How to select the appropriate polarization information and the coherence coefficient to extract the rice-planting area information was the key point and difficulty of this study.
Based on the time-series SAR images and rice improvement identification results of coherence-coefficient SAR images, Sentinel-2 optical image and grain statistical data were verified data with the aim to verify the reliability of rice identification results of SAR images.
The framework of the study was as follows (Figure 2).
Based on the time-series SAR images and rice improvement identification resu herence-coefficient SAR images, Sentinel-2 optical image and grain statistical data w fied data with the aim to verify the reliability of rice identification results of SAR ima The framework of the study was as follows (Figure 2

Data Used in This Study
Six Sentinel-1A single-look complex (SLC) images covering the study area were selected in this study. The parameters are shown in Table 2. The Sentinel-1 [24,25], equipped with C-band SAR sensor, can be used for all-weather Earth obs It has four imaging modes, namely, IW swath mode, stripmap mode, wave m extra-wide swath mode, and it can be used to realize single and dual polarizat products are SAR data with focus. Sufficient signal bandwidth is adopted to re single-look processing in each dimension in these products, and the complex is save the phase information. The SLC in IW mode includes three images correspo three bands in single polarization. The interferometric wide (IW) swath mode is acquisition mode over land and satisfies the majority of service requirements. It data with a 250 km swath at 5 m by 20 m spatial resolution (single look). IW mode three sub-swaths using Terrain Observation with Progressive Scans SAR (TOPSA SAR mode replaces the conventional ScanSAR mode, achieving the same cove resolution as ScanSAR but with a nearly uniform signal-to-noise ratio and distrib

Data Used in This Study
Six Sentinel-1A single-look complex (SLC) images covering the study area in 2017 were selected in this study. The parameters are shown in Table 2. The Sentinel-1 satellite [24,25], equipped with C-band SAR sensor, can be used for all-weather Earth observation. It has four imaging modes, namely, IW swath mode, stripmap mode, wave mode, and extra-wide swath mode, and it can be used to realize single and dual polarization. SLC products are SAR data with focus. Sufficient signal bandwidth is adopted to realize the single-look processing in each dimension in these products, and the complex is used to save the phase information. The SLC in IW mode includes three images corresponding to three bands in single polarization. The interferometric wide (IW) swath mode is the main acquisition mode over land and satisfies the majority of service requirements. It acquires data with a 250 km swath at 5 m by 20 m spatial resolution (single look). IW mode captures three subswaths using Terrain Observation with Progressive Scans SAR (TOPSAR). TOPSAR mode replaces the conventional ScanSAR mode, achieving the same coverage and resolution as ScanSAR but with a nearly uniform signal-to-noise ratio and distributed target ambiguity ratio. IW SLC products contain one image per sub-swath and one per polarization channel, for a total of three (single-polarization) or six (dual-polarization) images in an IW product.

. SAR Data Preprocessing
Sentinel-1 data preprocessing [26] involves track correction, thermal denoising, radiometric calibration, multi-look processing, filtering, topographic correction, and backscatter normalization. The ENVI-based SARscape module is used in Sentinel-1A data preprocessing. This module can comprehensively preprocess and analyze polarimetric radar data in various bands. SARscape currently supports the import, geometric correction, multi-look processing, registration, denoising, geocoding, and polarization decomposition of radar data and a series of basic processing functions. Because the study area is mainly in the plain with relatively flat terrain and mountainous area less than 3% of the total area, the slope effect cannot be considered in the processing. Preprocessing is made for VV polarization and VH polarization at the same time. The preprocessing process mainly includes the import, multi-look processing, registration, polarization, filtering, and geocoding of the data ( Figure 3). get ambiguity ratio. IW SLC products contain one image per sub-swath and one per polarization channel, for a total of three (single-polarization) or six (dual-polarization) images in an IW product.

. SAR Data Preprocessing
Sentinel-1 data preprocessing [26] involves track correction, thermal denoising, radiometric calibration, multi-look processing, filtering, topographic correction, and backscatter normalization. The ENVI-based SARscape module is used in Sentinel-1A data preprocessing. This module can comprehensively preprocess and analyze polarimetric radar data in various bands. SARscape currently supports the import, geometric correction, multi-look processing, registration, denoising, geocoding, and polarization decomposition of radar data and a series of basic processing functions. Because the study area is mainly in the plain with relatively flat terrain and mountainous area less than 3% of the total area, the slope effect cannot be considered in the processing. Preprocessing is made for VV polarization and VH polarization at the same time. The preprocessing process mainly includes the import, multi-look processing, registration, polarization, filtering, and geocoding of the data ( Figure 3). Taking VV and VH data on May 19 as the benchmark image, five images from June to October were registered. Time-series filtering [27] was conducted on the registered images to remove the speckle noise of the radar image. This method can be applied to single-band and multi-time-phase radar images. Multi-time-phase radar filtering includes De Grand Taking VV and VH data on May 19 as the benchmark image, five images from June to October were registered. Time-series filtering [27] was conducted on the registered images to remove the speckle noise of the radar image. This method can be applied to single-band and multi-time-phase radar images. Multi-time-phase radar filtering includes De Grand and anisotropic nonlinear diffusion. In this study, De Grand was adopted to process multi-time-phase SAR in the study area. GAMMA-MAP filtering was used in this study. The window was set to 5 × 5. Three RGB channels were used to display VV polarization and VH polarization in the study area after preprocessing.
On the basis of the image after geocoding, RGB channels were used to display the composite graph of VV and VH polarization (R (05/19), G (06/12), B (07/18)) in the study area. The results are shown in Figure 4. and anisotropic nonlinear diffusion. In this study, De Grand was adopted to process multi-time-phase SAR in the study area. GAMMA-MAP filtering was used in this study. The window was set to 5 × 5. Three RGB channels were used to display VV polarization and VH polarization in the study area after preprocessing.
On the basis of the image after geocoding, RGB channels were used to display the composite graph of VV and VH polarization (R (05/19), G (06/12), B (07/18)) in the study area. The results are shown in

Extraction and Analysis of Rice Coherence in the Sentinel-1A Image
The purpose of SAR image interference processing is to obtain the index representing the similarity between two images or among many SAR complex images, called coherence [28]. The similarity of target scatterers is reflected by radar echo. The similarity information and intensity information of ground objects provide the information related to target scatterers. The study on the surface activities in a specific region should be realized through decoherence in the interference analysis. The coherence coefficient is an important index of coherence. It is introduced to measure the similarity of measured ground objects. On the basis of many SAR images at different times in the study area, quantitative analysis was adopted to monitor the change of rice-planting area in different growth periods so as to extract the rice-planting area.
The rice in the study area is cultivated from March to April, including seeding and seedling raising. The rice is transplanted in May, matures in August, and is harvested in September. During the whole period, the rice field exhibits obvious changes. Therefore, this study was based on InSAR technology, and the changes in the rice field were monitored by using Sentinel-1A data in the process of transplanting, heading, ripening, and harvesting of rice so as to provide a reference for the identification of rice-planting areas.
The coherence of VV and VH polarization in Sentinel-1 time series was calculated on the basis of the SARscape module, and then the coherence coefficient is obtained. The data processing for calculating the coherence ( Figure 5) included baseline estimate, interferogram generation, self-adaptive filtering, coherence calculation, registration, geocoding and calibration, and cutting the study areas.

Extraction and Analysis of Rice Coherence in the Sentinel-1A Image
The purpose of SAR image interference processing is to obtain the index representing the similarity between two images or among many SAR complex images, called coherence [28]. The similarity of target scatterers is reflected by radar echo. The similarity information and intensity information of ground objects provide the information related to target scatterers. The study on the surface activities in a specific region should be realized through decoherence in the interference analysis. The coherence coefficient is an important index of coherence. It is introduced to measure the similarity of measured ground objects. On the basis of many SAR images at different times in the study area, quantitative analysis was adopted to monitor the change of rice-planting area in different growth periods so as to extract the rice-planting area.
The rice in the study area is cultivated from March to April, including seeding and seedling raising. The rice is transplanted in May, matures in August, and is harvested in September. During the whole period, the rice field exhibits obvious changes. Therefore, this study was based on InSAR technology, and the changes in the rice field were monitored by using Sentinel-1A data in the process of transplanting, heading, ripening, and harvesting of rice so as to provide a reference for the identification of rice-planting areas.
The coherence of VV and VH polarization in Sentinel-1 time series was calculated on the basis of the SARscape module, and then the coherence coefficient is obtained. The data processing for calculating the coherence ( Figure 5) included baseline estimate, interferogram generation, self-adaptive filtering, coherence calculation, registration, geocoding and calibration, and cutting the study areas.
The baseline is an important part in coherence generation. The baseline and slope distance of two SAR images form a triangle so as to produce the coherence. The baseline estimate is used to evaluate the quality of interference image pair. Thus, interference can happen. The SAR images generally used for change monitoring must meet the following conditions: the same sensor, the same ascending track (descending track), and incident angle. Taking the SAR image in rice field cultivation monitoring as an example, the sensor, ascending track, and incident angle meet the use conditions [29]. When at least two antennas are overlapped in obtaining the ground reflection, interference can be produced. The baseline-estimated relevant parameters of this study are shown in Figure 6. The accuracy of elevation and deformation increases with the increase of coherence. The baseline is 73.348 m long, which is far less than the critical baseline (486.486 m). The minimum change in the elevation that can be detected is 211.737 m, and the minimum deformation is 0.028 m. The baseline estimate result (Figures 7 and 8) shows that the quality of interference image pair used in the research satisfies the requirement of InSAR technology [30].  The baseline is an important part in coherence generation. The baseline and slope distance of two SAR images form a triangle so as to produce the coherence. The baseline estimate is used to evaluate the quality of interference image pair. Thus, interference can happen. The SAR images generally used for change monitoring must meet the following conditions: the same sensor, the same ascending track (descending track), and incident angle. Taking the SAR image in rice field cultivation monitoring as an example, the sensor, ascending track, and incident angle meet the use conditions [29]. When at least two antennas are overlapped in obtaining the ground reflection, interference can be produced. The baseline-estimated relevant parameters of this study are shown in Figure 6. The accuracy of elevation and deformation increases with the increase of coherence. The baseline is 73.348 m long, which is far less than the critical baseline (486.486 m). The minimum change in the elevation that can be detected is 211.737 m, and the minimum deformation is 0.028 m. The baseline estimate result (Figures 7 and 8) shows that the quality of interference image pair used in the research satisfies the requirement of InSAR technology [30].          On the premise that the baseline estimate meets the interference processing, the interferogram was generated. The interferogram generation results include interferogram ( Figure 9) and flattened interferogram ( Figure 10). On the premise that the baseline estimate meets the interference processing, the interferogram was generated. The interferogram generation results include interferogram ( Figure 9) and flattened interferogram ( Figure 10).  On the basis of the interferogram, self-adaptive filtering and coherence calculation were performed, and the Sentinel-1 coherence graph (Figures 11 and 12) was obtained. The coherence coefficient of the study area was extracted. Preprocessing of the interferometric data included the selection of the suitable SLC pairs for InSAR data analysis, then splitting Sentinel-1 data into sub-swath levels covering the study area (IW2 and IW3), selecting VV polarization for the InSAR coherence analysis, and applying a precise or-bit file, followed by co-registration to the sub-pixel accuracy using the Sentinel-1 TOPS co-registration algorithm.  On the basis of the interferogram, self-adaptive filtering and coherence calculation were performed, and the Sentinel-1 coherence graph (Figures 11 and 12) was obtained. The coherence coefficient of the study area was extracted. Preprocessing of the interferometric data included the selection of the suitable SLC pairs for InSAR data analysis, then splitting Sentinel-1 data into sub-swath levels covering the study area (IW2 and IW3), selecting VV polarization for the InSAR coherence analysis, and applying a precise orbit file, followed by co-registration to the sub-pixel accuracy using the Sentinel-1 TOPS co-registration algorithm.   On the basis of the interferogram, self-adaptive filtering and coherence calculation were performed, and the Sentinel-1 coherence graph (Figures 11 and 12) was obtained. The coherence coefficient of the study area was extracted. Preprocessing of the interferometric data included the selection of the suitable SLC pairs for InSAR data analysis, then splitting Sentinel-1 data into sub-swath levels covering the study area (IW2 and IW3), selecting VV polarization for the InSAR coherence analysis, and applying a precise orbit file, followed by co-registration to the sub-pixel accuracy using the Sentinel-1 TOPS co-registration algorithm.

The Change of Coherence Coefficient of VV Polarization
On the basis of the analysis of the change in the coherence coefficient of VV polarimetric ground objects, the coherence coefficient of rice first decreases and then increases ( Figure 13). Among them, the minimum value appears in August, when the rice is mature. The radar backscatter coefficient mainly represents the backscatter intensity of rice canopy. In May, the backscatter coefficient of rice mainly represents the backscatter intensity of the water in the rice field. Therefore, the coherence coefficient is small and varies greatly. After harvesting the rice in September, the radar backscatter coefficient mainly

The Change of Coherence Coefficient of VV Polarization
On the basis of the analysis of the change in the coherence coefficient of VV polarimetric ground objects, the coherence coefficient of rice first decreases and then increases ( Figure 13). Among them, the minimum value appears in August, when the rice is mature. The radar backscatter coefficient mainly represents the backscatter intensity of rice canopy.
In May, the backscatter coefficient of rice mainly represents the backscatter intensity of the water in the rice field. Therefore, the coherence coefficient is small and varies greatly. After harvesting the rice in September, the radar backscatter coefficient mainly comes from the surface soil or the surface water in the rice field after harvest. Therefore, the coherence coefficient is similar to that in June and varies slightly. The coherence coefficients of artificial buildings from June to October are more than 0.46, and the difference between the maximum and the minimum is 0.06. The mean value is far higher than that of rice, water area, and vegetation.  Different from the change in VV polarization coherence coefficient, the change in VH polarization coherence coefficient of rice shows a wave shape ( Figure 14). The minimum coherence coefficient is observed in June, August, and October, whereas the maximum coherence coefficient is observed in July and September. Similar to the VV polarization coherence coefficient, the VH polarization coherence coefficient of artificial buildings is greater than 0.37, which is far higher than that of rice, water area, and vegetation. The change in the VH polarization coherence coefficient of water is similar to that of the VV polarization coherence coefficient. Due to time variation and low single-to-noise ratio, the surface shows low interference coherence. The study area is located in the subtropical monsoon climate zone, and thus the growth period of rice overlaps with the rainy season. The seasonal sudden rainfall will cause the rise of water level and the expansion of water area. However, the coherence coefficient of the water changes by not more than 0.01. The change in the VH polarization coherence coefficient of vegetation is also similar to that of VV polarization (Table 4). If the coherence coefficient is greater, the coherence is higher; that is, it varies less. Most of the artificial buildings in the study area are buildings, roads, etc., which change slightly with time. Due to time variation and low single-to-noise ratio, the water surface shows low interference coherence. The study area is located in the subtropical monsoon climate zone. Thus, the growing period of rice overlaps with the rainy season. The seasonal sudden rainfall will cause the rise of water level and the expansion of water area. However, the coherence coefficient of the water changes by not more than 0.01. The climate of Guanghan belongs to the subtropical monsoon climate, but the city is located in the core area of Chengdu Plain. Therefore, there is no large area of vegetation that is common in subtropical monsoon climate zones, such as subtropical evergreen broad-leaved forest. Small areas of vegetation are mostly present in the study area. Therefore, although the rainfall increases, the vegetation grows insignificantly in the growing period of rice. During this period, the coherence coefficient varies slightly (Table 3). Different from the change in VV polarization coherence coefficient, the change in VH polarization coherence coefficient of rice shows a wave shape ( Figure 14). The minimum coherence coefficient is observed in June, August, and October, whereas the maximum coherence coefficient is observed in July and September. Similar to the VV polarization coherence coefficient, the VH polarization coherence coefficient of artificial buildings is greater than 0.37, which is far higher than that of rice, water area, and vegetation. The change in the VH polarization coherence coefficient of water is similar to that of the VV polarization coherence coefficient. Due to time variation and low single-to-noise ratio, the surface shows low interference coherence. The study area is located in the subtropical monsoon climate zone, and thus the growth period of rice overlaps with the rainy season. The seasonal sudden rainfall will cause the rise of water level and the expansion of water area. However, the coherence coefficient of the water changes by not more than 0.01. The change in the VH polarization coherence coefficient of vegetation is also similar to that of VV polarization (Table 4).  On the basis of the VV and VH polarization coherence coefficients of ground objects, the VV polarization coherence coefficient of artificial buildings is greater than the VH polarization coherence coefficient. No remarkable difference can be observed between the VV and VH polarization coherence coefficients of rice, vegetation, and water area. However, the VV and VH polarization coherence coefficients can be used to better distinguish artificial buildings from other ground objects. The VV polarization coherence coefficient can be used to distinguish rice from water area and vegetation in August and September, whereas the VH polarization coherence coefficient can also be used to well distinguish rice from water area and vegetation in August and October. However, after harvesting the rice in October, the SAR images show two intensities of soil and water in the rice field. In addition, the base time in this study is May. The difference between rice planting and harvest is little. Therefore, the coherence coefficient in October is not highly available. It may be more effective to distinguish rice by using the coherence coefficients of two time phases, namely, rice maturity and harvest. On the basis of the comprehensive analysis on the polarization coherence coefficients, this study shows that the polarization interferogram introduced by InSAR technology can support rice identification.

Extraction and Analysis of Backscattering Characteristics of Rice
On the basis of the time-series Sentinel-1A image, the rice-planting areas were extracted according to the backscatter coefficients of various land types in the growth period of rice.  On the basis of the VV and VH polarization coherence coefficients of ground objects, the VV polarization coherence coefficient of artificial buildings is greater than the VH polarization coherence coefficient. No remarkable difference can be observed between the VV and VH polarization coherence coefficients of rice, vegetation, and water area. However, the VV and VH polarization coherence coefficients can be used to better distinguish artificial buildings from other ground objects. The VV polarization coherence coefficient can be used to distinguish rice from water area and vegetation in August and September, whereas the VH polarization coherence coefficient can also be used to well distinguish rice from water area and vegetation in August and October. However, after harvesting the rice in October, the SAR images show two intensities of soil and water in the rice field. In addition, the base time in this study is May. The difference between rice planting and harvest is little. Therefore, the coherence coefficient in October is not highly available. It may be more effective to distinguish rice by using the coherence coefficients of two time phases, namely, rice maturity and harvest. On the basis of the comprehensive analysis on the polarization coherence coefficients, this study shows that the polarization interferogram introduced by InSAR technology can support rice identification.

Extraction and Analysis of Backscattering Characteristics of Rice
On the basis of the time-series Sentinel-1A image, the rice-planting areas were extracted according to the backscatter coefficients of various land types in the growth period of rice.

Change in the VV Polarization Object Backscatter Coefficient
The VV polarization backscatter coefficients of artificial buildings are greater than −3.98 (Table 5), whereas the backscatter coefficients of water areas are less than −19.21. The two ground objects can be well distinguished from rice and vegetation. Except in the rice maturation period on July 18, the polarization backscatter coefficient of rice is slightly higher than that of vegetation and lower than that of vegetation in other periods ( Figure 15).

Change in the VH Polarization Backscatter Coefficient of Ground Objects
Similar to VV polarization, artificial buildings and water area have the maximu minimum polarization backscatter coefficients, respectively ( Figure 16). Both can b tinguished from rice and vegetation obviously. However, the polarization backscat efficients of vegetation and rice are in a spirally interweaved state. The polari backscatter coefficient of rice basically shows an increasing trend in the whole g cycle, except from mid-July to mid-August, which shows a decreasing trend. The p zation backscatter coefficient of vegetation can be better distinguished from rice on 12, but there is little difference in other months (Table 6).

Change in the VH Polarization Backscatter Coefficient of Ground Objects
Similar to VV polarization, artificial buildings and water area have the maximum and minimum polarization backscatter coefficients, respectively ( Figure 16). Both can be distinguished from rice and vegetation obviously. However, the polarization backscatter coefficients of vegetation and rice are in a spirally interweaved state. The polarization backscatter coefficient of rice basically shows an increasing trend in the whole growth cycle, except from mid-July to mid-August, which shows a decreasing trend. The polarization backscatter coefficient of vegetation can be better distinguished from rice on June 12, but there is little difference in other months (Table 6). minimum polarization backscatter coefficients, respectively ( Figure 16). Both can be dis-tinguished from rice and vegetation obviously. However, the polarization backscatter coefficients of vegetation and rice are in a spirally interweaved state. The polarization backscatter coefficient of rice basically shows an increasing trend in the whole growth cycle, except from mid-July to mid-August, which shows a decreasing trend. The polarization backscatter coefficient of vegetation can be better distinguished from rice on June 12, but there is little difference in other months (Table 6).

Change in the VV/VH Polarization Backscatter Coefficients of Ground Objects
VV/VH is the ratio of VV polarization to VH polarization ( Figure 17). The change trend of the VV/VH ratio of artificial buildings and water area is similar to that of VV and VH polarization backscatter coefficients, which can be used to distinguish between rice and vegetation. Except in other months, the VV/VH value of vegetation is difficult to distinguish from rice on 12 June and 11 August (Table 7).

Change in the VV/VH Polarization Backscatter Coefficients of Ground Objects
VV/VH is the ratio of VV polarization to VH polarization ( Figure 17). The change trend of the VV/VH ratio of artificial buildings and water area is similar to that of VV and VH polarization backscatter coefficients, which can be used to distinguish between rice and vegetation. Except in other months, the VV/VH value of vegetation is difficult to distinguish from rice on 12 June and 11 August (Table 7). Table 7. VV/VH polarization backscatter coefficient of ground objects.  By comprehensively considering VV, VH, VV/VH, coherence coefficient, and other parameters of Sentinel-1 SAR image, this study aimed to comparatively analyze the rice identification results under two conditions, namely, time series and combination of time series and coherence coefficient so as to verify the contribution of SAR polarization coherence coefficient to rice identification.

Rice Identification on the Basis of the Time-Series Sentinel-1A SAR
On the basis of the comprehensive analysis of VV, VH, VV/VH, coherence coefficient, and other parameters, the features with great difference among different ground objects were considered. In addition, the features were used as the basic features of rice identifi-  By comprehensively considering VV, VH, VV/VH, coherence coefficient, and other parameters of Sentinel-1 SAR image, this study aimed to comparatively analyze the rice identification results under two conditions, namely, time series and combination of time series and coherence coefficient so as to verify the contribution of SAR polarization coherence coefficient to rice identification.

Rice Identification on the Basis of the Time-Series Sentinel-1A SAR
On the basis of the comprehensive analysis of VV, VH, VV/VH, coherence coefficient, and other parameters, the features with great difference among different ground objects were considered. In addition, the features were used as the basic features of rice identification so as to rapidly extract the rice-planting areas. Classification was made based on the VV and VH polarization on 12 June, VV polarization on 11 August, and VV/VH value in two months. In addition, the results are shown in R (12/06VV), G (11/08VV), and B (12/06VH)) in three channels ( Figure 18). In Figure 18, rice is colored red and dark brown, water is colored black, artificial buildings are colored yellow and white, and vegetation is colored dark green. Therefore, the color can be used to easily distinguish rice from other ground objects.
Appl. Sci. 2021, 11, x FOR PEER REVIEW (12/06VH)) in three channels ( Figure 18). In Figure 18, rice is colored red and dark water is colored black, artificial buildings are colored yellow and white, and vege colored dark green. Therefore, the color can be used to easily distinguish rice fro ground objects.

Rice Identification on the Basis of Time-Series SAR + Coherence Coefficient
Time series + coherence coefficient refers to time-series SAR combined with larization coherence coefficients on 12 June and 11 August, and the results are s R (12/06VV), G (11/08VV), and B (12/06VV coherence coefficient) in three channel 19). In Figure 19, rice is colored dark red, water is colored black, artificial build colored blue and white, and vegetation is colored light green. The color can be easily distinguish rice from other ground objects.

Rice Identification on the Basis of Time-Series SAR + Coherence Coefficient
Time series + coherence coefficient refers to time-series SAR combined with the polarization coherence coefficients on 12 June and 11 August, and the results are shown in R (12/06VV), G (11/08VV), and B (12/06VV coherence coefficient) in three channels ( Figure 19). In Figure 19, rice is colored dark red, water is colored black, artificial buildings are colored blue and white, and vegetation is colored light green. The color can be used to easily distinguish rice from other ground objects.

Rice Identification on the Basis of Time-Series SAR + Coherence Coefficient
Time series + coherence coefficient refers to time-series SAR combined with the polarization coherence coefficients on 12 June and 11 August, and the results are shown in R (12/06VV), G (11/08VV), and B (12/06VV coherence coefficient) in three channels ( Figure  19). In Figure 19, rice is colored dark red, water is colored black, artificial buildings are colored blue and white, and vegetation is colored light green. The color can be used to easily distinguish rice from other ground objects. Figure 19. Multi-temporal and coherence coefficient RGB composite map. Figure 19. Multi-temporal and coherence coefficient RGB composite map.

Accuracy Evaluation
To further verify the accuracy of rice extraction on the basis of SAR image, the riceplanting area in the study area was extracted by using object-oriented classification method (segmentation scale 10; the proportion of each band is 1) on the basis of Sentinel-2 optical remote sensing image. The local effect is shown in Figure 20. To verify the reliability of the classification result, 200 validation sample points of four land types were collected from Google Earth, and the classification accuracy of the statistical area of rice in Guanghan City in 2017 was evaluated by using an error matrix based on TTA Mask.

Accuracy Evaluation
To further verify the accuracy of rice extraction on the basis of SAR image, the riceplanting area in the study area was extracted by using object-oriented classification method (segmentation scale 10; the proportion of each band is 1) on the basis of Sentinel-2 optical remote sensing image. The local effect is shown in Figure 20. To verify the reliability of the classification result, 200 validation sample points of four land types were collected from Google Earth, and the classification accuracy of the statistical area of rice in Guanghan City in 2017 was evaluated by using an error matrix based on TTA Mask. The classification accuracies of time-series SAR and time-series SAR combined with coherence coefficient are shown in Tables 8 and 9, respectively. The user classification accuracy of Scheme 1 is 78%, whereas that of Scheme 2 is improved by 3%. On the basis of area verification in Table 10, the accuracy of Scheme 2 is improved by 2.05% compared The classification accuracies of time-series SAR and time-series SAR combined with coherence coefficient are shown in Tables 8 and 9, respectively. The user classification accuracy of Scheme 1 is 78%, whereas that of Scheme 2 is improved by 3%. On the basis of area verification in Table 10, the accuracy of Scheme 2 is improved by 2.05% compared to that of Scheme 1. On the basis of the verification result of the error matrix and statistical area accuracy, the accuracy can be improved significantly after the coherence coefficient is combined with time-series SAR classification. On the basis of the rice area statistical accuracy in Table 10, the accuracy of the rice-planting area extraction of Scheme 2 is significantly higher than that of Scheme 1.   Figure 20 shows that the rice-planting area extracted by this scheme basically corresponds to the spatial position extracted by the optical image (Area A in Figure 20). This scheme proves to be practical and has significance for the large-scale monitoring of rice and other food crops.
The polarization backscatter coefficient of rice cannot be easily distinguished from vegetation in the SAR image, which leads to error in the extraction of rice and vegetation information. On the basis of InSAR, the coherence coefficient was introduced in this study, which helps to distinguish rice and vegetation information and improve the identification accuracy.
Finally, on the basis of the analysis of the overall accuracy and Kappa coefficient of the two schemes, the overall accuracy is improved by 2.74%, and the Kappa coefficient is improved by 0.366 due to the use of coherence coefficient. On the basis of the above analysis, the introduction of coherence coefficient to extract the rice-planting area can significantly improve not only the accuracy of rice identification but also the accuracy of land cover classification. Moreover, it is of great significance for studies on land cover on the basis of SAR images.
In cloudy regions, the application of optical image is restricted. The SAR image can penetrate through clouds and mists and therefore make up for the limitation of optical image. Hence, the use of optical image or SAR image should be determined according to the characteristics of the study area, research purpose, and research methods in rice monitoring and other fields. In this selection, the accuracy difference in the rice extraction result between optical image and SAR image should be analyzed under different cloud covers to obtain a scale. This scale can be used to determine which data source should be selected efficiently and quickly.

Conclusions
In this study, the core area of Chengdu Plain was taken as a case example. The polarization coherence coefficient of rice on the basis of InSAR technology was analyzed, and the polarization backscatter coefficient and polarization coherence coefficient of rice on the basis of time-series SAR image were extracted. The two schemes based on time-series SAR image and time-series SAR combined with the coherence coefficient were analyzed to identify the rice-planting area. The following conclusions can be drawn: (1) The VV and VH polarization coherence coefficients of typical ground objects during the growing period of rice were obtained on the basis of InSAR technology. The analysis showed that the VV and VH polarization coherence coefficients can be used to well distinguish artificial buildings from other ground objects. The VV polarization coherence coefficient can be used to well distinguish rice from water area and vegetation in August and September; the VH polarization coherence coefficient can also be used to well distinguish rice from water area and vegetation in August and October.
(2) The rice identification accuracy on the basis of time-series Sentinel-1 SAR image is lower than that of time-series Sentinel-1 SAR image combined with InSAR technology. The user accuracies of both methods are 78% and 81%, respectively. The rice identification result based on Sentinel-2 optical remote sensing images was compared with three rice identification results based on Sentinel-1 SAR images. Compared with single-time-phase SAR, the time-series SAR image can distinguish rice and vegetation information more significantly, thus improving the rice identification accuracy. The introduction of coherence coefficient not only improves the classification accuracy of artificial buildings but also significantly distinguishes vegetation and rice, thus improving the identification accuracy. Rice identification on the basis of time-series Sentinel-1 SAR image combined with In-SAR technology can achieve a good classification result, which shows that the coherence coefficient improves rice identification.
(3) In monitoring rice and other crops in cloudy regions, optical data are preferred if they are available. If optical data are lacking, SAR data can be used to monitor crops. The Sentinel-1 series images reach 250 km in width, which is significantly conducive to large-scale rice monitoring or mapping in plains or areas with small topographic relief.