A method for identifying single and double cropping rice in the south based on multi-source data and phenological characteristics

Abstract

The application discloses a kind of southern single double-cropping rice identification method based on multi-source data and phenological characteristics, download satellite image of different remote sensing satellites, and obtain time-series image dataset by processing;S-G filter LSWI image, S-G filter NDVI image and HANTS filter NDVI image are obtained based on time-series image dataset;Obtain the cultivated land distribution layer in study area;Rice potential distribution layer is obtained based on S-G filter LSWI image and S-G filter NDVI image;Single double-cropping rice identification is completed based on HANTS filter NDVI image.The application cooperates multiple medium-resolution remote sensing satellites, by S-G filtering and HANTS filtering two filtering methods, the potential information characteristics of time-series image dataset are fully tapped, and the classification accuracy and identification effect of single double-cropping rice are improved.

Description

Technical Field

[0001] This invention relates to the field of large-area single and double-cropping rice classification technology, specifically to a method for identifying single and double-cropping rice in southern China based on multi-source data and phenological characteristics, belonging to the field of agricultural remote sensing technology. Background Technology

[0002] Rice is one of the world's most important food crops, and understanding its planting area is crucial for ensuring global food security. Furthermore, methane emissions from rice production account for approximately 8% of all human-related methane emissions, making the spatial distribution of rice a vital tool for mitigating climate change. my country is the world's largest rice producer, with over 60% of its population relying on rice as a staple food. Single-season and double-season rice are the main rice cultivation patterns in my country, and their planting distribution and harvested area have undergone significant changes over the past two decades. Due to abundant rainfall and heat in southern my country, both single-season and double-season rice are cultivated, and both are major food crops in the region. Therefore, identifying and monitoring the spatial distribution of single and double-season rice cultivation in southern my country using remote sensing technology is helpful in formulating agricultural production policies, promoting the efficient use of water and soil resources, and providing important data support for achieving sustainable development goals. However, due to the cloudy and rainy weather and fragmented arable land in southern my country, identifying single and double-season rice in this region remains a challenge in the field of agricultural remote sensing.

[0003] Currently, existing remote sensing identification technologies for single and double-cropping rice both domestically and internationally are mainly divided into data-driven methods and phenological methods. Data-driven methods, such as machine learning and deep learning, use a large number of field sampling points to drive model training and combine remote sensing images to generate classification results. However, this method requires a significant investment of manpower and resources for high-quality, comprehensive field sample collection covering the entire study area. Phenological methods are classification methods based on the unique temporal phenological characteristics of single and double-cropping rice, achieving final classification by setting threshold decision rules. Phenological methods only require a small number of sample points for threshold adjustment to complete model construction. Furthermore, they have a wide range of applicability and can be adjusted according to farmland management. Current research on phenological identification of single and double-cropping rice mainly focuses on two features: first, the flooding characteristic during the rice transplanting period, used to distinguish between rice and non-rice; and second, the multiple cropping characteristic of rice, used to distinguish between single-cropping and double-cropping rice. However, current phenological research often relies on a single remote sensing image dataset for the identification of these two features, resulting in room for improvement in identification accuracy and optimization of identification effects. How to coordinate multi-source remote sensing data, encrypt image time series, improve the utilization rate of remote sensing image time series datasets, and thus improve the classification accuracy of single and double-cropping rice in southern China is a hot research topic in the field of agricultural remote sensing. Summary of the Invention

[0004] The purpose of this invention is to address the problem of low accuracy in remote sensing identification of single and double-season rice in existing technologies. By combining the unique phenological characteristics of single and double-season rice in southern China and using multi-source optical data in a coordinated manner, this invention provides a method for identifying single and double-season rice in southern China based on multi-source data and phenological features. The method improves the single and double-season rice screening process by considering the different reconstruction effects of different filtering methods on time-series images, fully explores the potential of remote sensing datasets, and thus optimizes the identification effect of single and double-season rice, achieving high-precision mapping of single and double-season rice in southern China.

[0005] The above-mentioned objectives of the present invention are achieved through the following technical means:

[0006] A method for identifying single and double-cropping rice in southern China based on multi-source data and phenological characteristics includes the following steps:

[0007] Step 1: Download satellite images from different remote sensing satellites, and perform image preprocessing, image collaboration and temporal synthesis processing respectively to obtain a time-series image dataset;

[0008] Step 2: Calculate vegetation index from the time-series image dataset to obtain a vegetation index image dataset. Perform time-series filtering on the vegetation index image dataset to obtain SG-filtered LSWI images, SG-filtered NDVI images, and HANTS-filtered NDVI images.

[0009] Step 3: Obtain the farmland distribution layer within the study area;

[0010] Step 4: After removing all non-cultivated land pixels in the study area using the cultivated land distribution layer on the SG-filtered LSWI and SG-filtered NDVI images, the SG-filtered LSWI and SG-filtered NDVI images, combined with the flooding characteristics during the paddy field flooding period, are used to distinguish between rice and non-rice, and obtain the potential distribution layer of rice.

[0011] Step 5: After removing all non-cultivated land pixels in the study area using the cultivated land distribution layer on the HANTS-filtered NDVI image, the single and double-season rice identification is completed using the HANTS-filtered NDVI image and the potential rice distribution layer.

[0012] As described above, step one includes the following steps:

[0013] Level-1C Sentinel-2, Landsat-7, and Landsat-8 satellite imagery were downloaded separately. The downloaded Sentinel-2, Landsat-7, and Landsat-8 satellite imagery were then processed to remove clouds, snow, and cirrus clouds, and image co-processing was performed. The Sentinel-2, Landsat-7, and Landsat-8 satellite imagery were then temporally composited, and the composite image dataset was interpolated using a temporal linear interpolation method to finally obtain a temporal image dataset.

[0014] As described above, step three includes the following steps: masking the study area based on the arable land pixel distribution in the GLAD arable land product dataset to extract arable land pixels, thereby obtaining the arable land distribution layer in the study area.

[0015] As mentioned above, in step four, the flooding characteristics during the paddy field flooding period are combined to distinguish between rice and non-rice, and the potential distribution layer of rice is obtained based on the following formula:

[0016]

[0017] Where i is the scene number, LSWii is the i-th SG-filtered LSWI image, and NDVIi is the i-th SG-filtered NDVI image. When Rice Index(Z) is 1, it means that pixel Z is a potential rice field, and when Rice Index(Z) is 0, it means that pixel Z is a non-rice field. This forms the final potential rice distribution layer for the study area.

[0018] As described above, in step five, the identification of single and double-cropping rice is completed using HANTS-filtered NDVI imagery and a potential rice distribution layer based on the following formula:

[0019]

[0020]

[0021] Where i is the scene number, Peak number The number of peak values ​​for pixel Z in a HANTS-filtered NDVI image within the i-scene range; Peak value The maximum peak value of pixel Z in the i-scene range of the HANTS-filtered NDVI image is used. Based on the potential distribution layer of rice, when pixel Z meets the SCR(Z) condition, pixel Z is judged to be single-season rice; when pixel Z meets the DCR(Z) condition, pixel Z is judged to be double-season rice; when neither the SCR(Z) condition nor the DCR(Z) condition is met, pixel Z is judged to be non-rice, thus obtaining the final single-season and double-season rice identification results for the study area.

[0022] Compared with the prior art, the present invention has the following advantages:

[0023] This invention addresses the issue of low accuracy in classifying single- and double-cropping rice using phenological methods. It collaborates with multiple medium-resolution remote sensing satellites and, with the assistance of a limited number of field sample points, employs both SG and HANTS filtering methods to fully extract the latent information features from time-series image datasets. Ultimately, based on the temporal characteristics of different filtered datasets, an improved phenological method enhances the classification accuracy and recognition performance of single- and double-cropping rice, fulfilling the mapping requirements for single- and double-cropping rice in southern China. Attached Figure Description

[0024] Figure 1 This is a flowchart of the method steps of the present invention;

[0025] Figure 2 This is a detailed technical flow diagram of the method of the present invention;

[0026] Figure 3 This is the test and research area in the embodiments of the present invention;

[0027] Figure 4 These are time-series vegetation index curves after different filtering processes used in the embodiments of the present invention, wherein a is the time-series vegetation index curve of single-season rice, b is the time-series vegetation index curve of single-season rice with SG filtering, c is the time-series vegetation index curve of single-season rice with HANT filtering, d is the time-series vegetation index curve of double-season rice, e is the time-series vegetation index curve of double-season rice with SG filtering, and f is the time-series vegetation index curve of double-season rice with HANT filtering.

[0028] Figure 5 This is a comparison chart of the identification accuracy of single and double-season rice in this embodiment of the invention and the identification accuracy of single and double-season rice using the traditional phenological method. In this chart, A is the comparison chart of producer accuracy, B is the comparison chart of user accuracy, C is the comparison chart of F1-scores, and D is the comparison chart of overall accuracy and the mean of F1-scores.

[0029] Figure 6 This is a distribution map of single and double-season rice in the test study area of ​​this invention embodiment. Detailed Implementation

[0030] To facilitate understanding and implementation of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to examples. The implementation examples described herein are only for illustration and explanation and are not intended to limit the present invention.

[0031] Example:

[0032] like Figure 1 As shown, this invention discloses a method for identifying the phenology of single and double-cropping rice in southern China based on multi-source remote sensing data. The test study area is as follows: Figure 3As shown. A detailed technical process diagram is shown below. Figure 2 As shown, it includes:

[0033] Step 1: Time-series image download and preprocessing

[0034] Download Sentinel-2 and Landsat-7 / 8 satellite imagery, and perform image preprocessing, image co-processing, and temporal composite processing to obtain a time-series image dataset. Specifically, the detailed steps in step one are as follows:

[0035] Data was downloaded using the Google Earth Engine platform, including Level-1C Sentinel-2 satellite imagery (TOA dataset), Landsat-7 satellite imagery (Collection 1 Tier 1 TOA data), and Landsat-8 satellite imagery. The downloaded Sentinel-2, Landsat-7, and Landsat-8 satellite imagery underwent cloud removal, snow removal, and cirrus cloud removal, and image co-processing was also performed. After processing, the Sentinel-2, Landsat-7, and Landsat-8 satellite imagery were temporally composited, with a composite window size of 10 days, using the median composite rule to generate a composite image dataset. Temporal linear interpolation was then used to interpolate the composite image dataset to remove the influence of null pixels on subsequent filtering effects, ultimately yielding a time-series image dataset.

[0036] Step 2: Calculation of different filtered vegetation index datasets

[0037] Vegetation indices are calculated based on the time-series image dataset generated in Step 1, resulting in a 10-day composite vegetation index image dataset. SG filtering and HANTS filtering are then used to perform time-series processing on the 10-day composite vegetation index image dataset, yielding SG-filtered LSWI images, SG-filtered NDVI images, and HANTS-filtered NDVI images. Specifically, the detailed steps in Step 2 are as follows:

[0038] Based on the time-series image dataset generated in step one, vegetation indices are calculated. The NDVI (Near Infrared and Red bands) and LSWI (Near Infrared and Shortwave Infrared bands) are used to calculate the vegetation index, resulting in a ten-day composite vegetation index image dataset, as detailed below:

[0039] Table 1. Vegetation Index and Calculation Formula

[0040]

[0041] Where, ρ NIR For the near-infrared band, ρ RED For the red light band, ρSWIR It is in the shortwave infrared band.

[0042] Temporal filtering was performed on the vegetation index image dataset. The specific processing flow is as follows: the sgolayfilt function in Matlab was used to perform SG filtering on the vegetation index image dataset, obtaining SG-filtered LSWI and SG-filtered NDVI images. Simultaneously, the HANTS function developed by Mohammad Abouali in Matlab was used to perform HANTS filtering on the vegetation index image dataset, obtaining HANTS-filtered NDVI images. The parameter settings are shown in Table 2. The filtering results are as follows. Figure 4 As shown.

[0043] Table 2 HANTS Filter and SG Filter Parameter Settings

[0044]

[0045]

[0046] Step 3: Covering non-arable land

[0047] Based on the GLAD farmland product dataset released by Peter Potapov in 2022, the farmland and non-farmland pixels in the study area were identified. The specific process is as follows: The farmland pixel distribution in the GLAD farmland product dataset was used to mask the study area to extract farmland pixels, and the influence of non-farmland pixels on the classification of single and double-cropping rice was removed to obtain the farmland distribution layer in the study area.

[0048] Step 4: Rice Identification Using the SG Filter Dataset

[0049] After removing all non-cultivated land pixels in the study area from the cultivated land distribution layer generated in step three, the SG-filtered LSWI and SG-filtered NDVI images generated in step two are used in conjunction with the flooding characteristics of rice paddies during the paddy field flooding period to distinguish between rice and non-rice. The judgment criteria are shown in Formula 3:

[0050]

[0051] Where i represents the scene number, 8 < i < 14 corresponding to the scene number during the rice flooding period, LSWii is the i-th SG-filtered LSWI image, and NDVIi is the i-th SG-filtered NDVI image. When Rice Index(Z) is 1, it indicates that pixel Z is a potential rice field; when Rice Index(Z) is 0, it indicates that pixel Z is not a rice field. Rice Index calculations are performed on all pixels within the study area to obtain the final potential rice distribution layer for the study area, which contains only rice pixels.

[0052] Step 5: Identification of Single and Double Season Rice in the HANTS Filtered Dataset

[0053] After removing all non-cultivated land pixels in the study area from the HANTS-filtered NDVI image generated in step two using the cultivated land distribution layer generated in step three, and based on the potential rice distribution layer of the study area generated in step four, the identification of single and double-cropping rice is completed by combining the different growth phenological characteristics of single and double-cropping rice. The judgment criteria are shown in formulas (4) and (5):

[0054]

[0055]

[0056] Where i represents the scene number, 20 < i < 27 corresponds to the scene number during the heading stage of single-season rice, 16 < i < 21 corresponds to the scene number during the heading stage of early rice, and 25 < i < 30 corresponds to the scene number during the heading stage of late rice. Peaknumber is the number of peak values ​​of pixel Z in the HANTS-filtered NDVI image within scene i; Peakvalue is the maximum peak value of pixel Z in the HANTS-filtered NDVI image within scene i. Based on the potential distribution layer of rice, when pixel Z meets the SCR(Z) condition, pixel Z is determined to be single-season rice; when pixel Z meets the DCR(Z) condition, pixel Z is determined to be double-season rice; if neither condition is met, pixel Z is determined to be non-rice. Single and double-season rice identification calculations were performed on all pixels in the potential distribution layer of rice in the study area to obtain the final single and double-season rice identification results for the study area.

[0057] To evaluate the optimization effect of this embodiment compared with the ordinary phenological method, this invention evaluated and compared the producer accuracy, user accuracy, overall accuracy, and F1-score of the three methods for single and double-cropping rice, such as... Figure 5 As shown. The method of this invention improves the overall accuracy by 32.95% compared to using the SG filter dataset alone, and by 2.01% compared to using the HANTS filter dataset alone. The final classification results of the method of this invention are as follows. Figure 6 As shown.

[0058] It should be noted that the specific embodiments described in this invention are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains can make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A method for identifying single and double-cropping rice in southern China based on multi-source data and phenological characteristics, characterized in that, Includes the following steps: Step 1: Download satellite images from different remote sensing satellites, and perform image preprocessing, image collaboration and temporal synthesis processing respectively to obtain a time-series image dataset; Step 2: Calculate vegetation index from the time-series image dataset to obtain a vegetation index image dataset. Perform time-series filtering on the vegetation index image dataset to obtain SG-filtered LSWI images, SG-filtered NDVI images, and HANTS-filtered NDVI images. Step 3: Obtain the farmland distribution layer within the study area; Step 4: After removing all non-cultivated land pixels in the study area using the cultivated land distribution layer on the SG-filtered LSWI and SG-filtered NDVI images, the SG-filtered LSWI and SG-filtered NDVI images, combined with the flooding characteristics during the paddy field flooding period, are used to distinguish between rice and non-rice, and obtain the potential distribution layer of rice. Step 5: After removing all non-cultivated land pixels in the study area using the cultivated land distribution layer on the HANTS-filtered NDVI image, the single- and double-cropping rice identification is completed using the HANTS-filtered NDVI image and the potential rice distribution layer. In step four, the flooding characteristics during the paddy field flooding period are combined to distinguish between paddy fields and non-paddy fields, and the potential distribution layer of paddy fields is obtained based on the following formula: Where i is the scene number, LSWii is the i-th SG-filtered LSWI image, and NDVIi is the i-th SG-filtered NDVI image. When Rice Index (Z) is 1, it indicates that pixel Z is a potential rice field, and when Rice Index (Z) is 0, it indicates that pixel Z is not a rice field. This yields the final potential rice distribution layer for the study area. In step five, the identification of single- and double-cropping rice is completed using HANTS-filtered NDVI images and a potential rice distribution layer based on the following formula: Where i is the scene number, Peak number The number of peak values ​​for pixel Z in a HANTS-filtered NDVI image within the i-scene range; Peak value The maximum peak value of pixel Z in the i-scene range of the HANTS-filtered NDVI image is used. Based on the potential distribution layer of rice, when pixel Z meets the SCR(Z) condition, pixel Z is judged to be single-season rice; when pixel Z meets the DCR(Z) condition, pixel Z is judged to be double-season rice; when neither the SCR(Z) condition nor the DCR(Z) condition is met, pixel Z is judged to be non-rice, thus obtaining the final single-season and double-season rice identification results for the study area.

2. The method for identifying single and double-cropping rice in southern China based on multi-source data and phenological characteristics according to claim 1, characterized in that, Step one includes the following steps: Level-1C Sentinel-2, Landsat-7, and Landsat-8 satellite imagery were downloaded separately. The downloaded Sentinel-2, Landsat-7, and Landsat-8 satellite imagery were then processed to remove clouds, snow, and cirrus clouds, and image co-processing was performed. The Sentinel-2, Landsat-7, and Landsat-8 satellite imagery were then temporally composited, and the composite image dataset was interpolated using a temporal linear interpolation method to finally obtain a temporal image dataset.

3. The method for identifying single and double-cropping rice in southern China based on multi-source data and phenological characteristics according to claim 2, characterized in that, Step three includes the following steps: masking the study area based on the arable land pixel distribution in the GLAD arable land product dataset to extract arable land pixels, thereby obtaining the arable land distribution layer in the study area.

Images