CN115937705A

CN115937705A - South single-season and double-season rice identification method based on multi-source data and phenological characteristics

Info

Publication number: CN115937705A
Application number: CN202211589758.4A
Authority: CN
Inventors: 徐保东; 张馨予; 胡琼
Original assignee: Huazhong Agricultural University
Current assignee: Huazhong Agricultural University
Priority date: 2022-12-11
Filing date: 2022-12-11
Publication date: 2023-04-07

Abstract

The invention discloses a method for identifying south single and double cropping rice based on multi-source data and phenological characteristics, which comprises the steps of downloading satellite images of different remote sensing satellites, and processing to obtain a time sequence image data set; obtaining an S-G filtered LSWI image, an S-G filtered NDVI image and a HANTS filtered NDVI image based on the time sequence image dataset; acquiring a cultivated land distribution map layer in a research area; obtaining a potential distribution layer of the rice based on the S-G filtering LSWI image and the S-G filtering NDVI image; and finishing single and double cropping rice identification based on the HANTS filtering NDVI image. The method provided by the invention cooperates with a plurality of medium-resolution remote sensing satellites, fully excavates potential information characteristics of a time sequence image data set through two filtering methods of S-G filtering and HANTS filtering, and improves the classification precision and identification effect of single and double cropping rice.

Description

South single-season and double-season rice identification method based on multi-source data and phenological characteristics

Technical Field

The invention relates to the technical field of large-area single and double cropping rice classification, in particular to a south single and double cropping rice identification method based on multi-source data and phenological characteristics, and belongs to the technical field of agricultural remote sensing.

Background

The rice is one of the important grain crops in the world, and the control of the planting area of the rice is beneficial to guaranteeing the grain safety in the world. Meanwhile, the methane discharged in the rice production process accounts for about 8% of the methane discharge amount related to human beings, so that the spatial distribution information of the rice plays a crucial role in slowing down the climate change. China is the largest rice producing country in the world, and over 60 percent of people use rice as staple food. Single-cropping rice and double-cropping rice are the main rice planting modes in China, and the planting distribution and harvesting area of the single-cropping rice and the double-cropping rice are changed greatly in the last two decades. As the southern China has abundant rain and heat conditions and has a planting pattern with single-cropping rice and double-cropping rice, the single-cropping rice and the double-cropping rice are also main grain crops in the region. Therefore, the method is based on the remote sensing technology means to identify and monitor the planting space distribution of the southern single and double cropping rice in China, is beneficial to making relevant policies of agricultural production and promoting the efficient utilization of water and soil resources, and further provides important data support for realizing the target of sustainable development. However, because southern China is cloudy and rainy and cultivated land is broken, identification of single-season and double-season rice in southern regions is a difficult point in the technical field of agricultural remote sensing.

At present, the existing single and double cropping rice remote sensing identification technology at home and abroad is mainly divided into a data driving method and a phenological method. Data-driven methods such as machine learning and deep learning perform model-driven training by using a large number of on-site sampling points, and generate classification results in combination with remote sensing images. However, this method has the disadvantage of requiring a large amount of manpower and material resources to perform high quality, full-coverage field sample point acquisition in the research area. The phenological method is a classification method for realizing final classification by setting a threshold decision rule based on unique time sequence phenological characteristics of single-season and double-season rice. The phenological method only needs a small number of sample points to adjust the threshold value, and then the model can be constructed. Meanwhile, the method has wide application space range and can be combined with farmland management to carry out corresponding method adjustment. The current single and double cropping rice phenological method identification research is mainly based on two characteristics. The first is the flooding characteristic of the rice transplanting period, and the characteristic is used for distinguishing rice from non-rice. The second is the multiple cropping characteristic of rice, which is used for distinguishing single cropping rice from double cropping rice. However, the current phenological method research is usually based on the identification of two characteristics by a single remote sensing image data set, so that the identification precision needs to be improved, and the identification effect needs to be optimized. How to cooperate with multi-source remote sensing data, encrypt an image time sequence, improve the utilization rate of a remote sensing image time sequence data set, and further improve the classification precision of single and double cropping rice in south China is a hot research topic in the field of current agricultural remote sensing.

Disclosure of Invention

The invention aims to solve the problem of low remote sensing identification precision of single-double cropping rice in the prior art, combine with the unique phenological characteristics of single-double cropping rice in China and cooperatively use multi-source optical data, provide a method for identifying single-double cropping rice in south China based on multi-source data and phenological characteristics, improve the screening process of single-double cropping rice according to different reconstruction effects of different filtering methods on time sequence images, and fully excavate the potential of remote sensing data sets, thereby optimizing the identification effect of single-double cropping rice and realizing high-precision mapping of single-double cropping rice in south China.

The above object of the present invention is achieved by the following technical means:

a south single-double cropping rice identification method based on multi-source data and phenological characteristics comprises the following steps:

downloading satellite images of different remote sensing satellites, and respectively performing image preprocessing, image cooperation and time synthesis processing to obtain a time sequence image data set;

step two, calculating vegetation indexes of the sequential image dataset to obtain a vegetation index image dataset, and performing sequential filtering processing on the basis of the vegetation index image dataset to obtain an S-G filtering LSWI image, an S-G filtering NDVI image and an HANTS filtering NDVI image;

step three, acquiring a cultivated land distribution map layer in the research area;

step four, after removing all non-cultivated land pixels in the research area from the S-G filtering LSWI image and the S-G filtering NDVI image by using a cultivated land distribution map layer, distinguishing rice and non-rice by using the S-G filtering LSWI image and the S-G filtering NDVI image and combining the flooding characteristics of the rice in a paddy field steeping period to obtain a potential rice distribution map layer;

and fifthly, removing all non-cultivated land pixels in the research area from the HANTS filtered NDVI image by using a cultivated land distribution layer, and then completing single and double cropping rice identification by using the HANTS filtered NDVI image and a rice potential distribution layer.

The first step comprises the following steps:

respectively downloading a Level-1C-Level Sentinel-2 satellite image, a Landsat-7 satellite image and a Landsat-8 satellite image, carrying out cloud, snow and rolling cloud removal on the downloaded Sentinel-2, landsat-7 and Landsat-8 satellite images, carrying out image cooperative processing, carrying out time synthesis processing on the Sentinel-2, landsat-7 and Landsat-8 satellite images, and carrying out interpolation processing on a synthetic image data set by a time sequence linear interpolation method to finally obtain a time sequence image data set.

The third step as described above includes the following steps: and masking the research area based on the farmland pixel distribution in the GLAD farmland product data set to extract farmland pixels to obtain a farmland distribution map layer in the research area.

In the fourth step, the rice and non-rice are distinguished by combining the flooding characteristics of the rice in the field soaking period, and the obtained potential distribution map layer of the rice is based on the following formula:

wherein i is the scene number serial number, LSWIi is the ith scene S-G filtering LSWI image, NDVIi is the ith scene S-G filtering NDVI image, when the Rice Index (Z) is 1, the pixel Z is the potential Rice, when the Rice Index (Z) is 0, the pixel Z is the non-Rice, and then the Rice potential distribution map layer of the research area is finally obtained.

In the fifth step, the HANTS filtering NDVI images and the potential distribution layers of the rice are utilized to complete the identification of the single and double cropping rice based on the following formula:

wherein i is the scene number serial number, peak _number For pixels Z of HANTS-filtered NDVI image in i scene rangeThe number of peak values; peak _value For the maximal peak value of the pixel Z of the HANTS filtered NDVI image in the i scene range, on the basis of a potential distribution layer of the rice, when the pixel Z meets the SCR (Z) condition, judging the pixel Z to be single cropping rice; when the pixel Z meets the DCR (Z) condition, judging the pixel Z to be double cropping rice; and if the SCR (Z) condition and the DCR (Z) condition are not met, judging that the pixel Z is non-rice, and obtaining the final single-double cropping rice identification result in the research area.

Compared with the prior art, the invention has the following beneficial effects:

in order to solve the problem that the classification precision of the single-season and double-season rice differentiated by the phenological method is low, a plurality of medium-resolution remote sensing satellites are cooperated, and potential information characteristics of a time sequence image data set are fully mined by two filtering methods of S-G filtering and HANTS filtering under the assistance of a small number of on-site sample points. Finally, based on the time sequence characteristics of different filtering data sets, the classification precision and the recognition effect of single and double cropping rice are improved through an improved phenological method, and the drawing requirements of the south China single and double cropping rice are met.

Drawings

FIG. 1 is a flow chart of the method steps of the present invention;

FIG. 2 is a detailed technical process schematic of the method of the present invention;

FIG. 3 is a test study area in an embodiment of the present invention;

FIG. 4 is a time series curve of vegetation indexes after different filtering processes used in embodiments of the present invention, where a is a single cropping rice vegetation index time series curve, b is a single cropping rice S-G filtering vegetation index time series curve, c is a single cropping rice HANT filtering vegetation index time series curve, d is a double cropping rice vegetation index time series curve, e is a double cropping rice S-G filtering vegetation index time series curve, and f is a double cropping rice HANT filtering vegetation index time series curve;

FIG. 5 is a comparison graph of the recognition accuracy of single and double cropping rice in the embodiment of the present invention and the recognition accuracy of single and double cropping rice in the conventional phenology method, wherein A is a producer accuracy comparison graph and B is a user accuracy comparison graph; c is F1-scores comparison graph; d is a comparison graph of overall precision and an F1-scores mean value;

FIG. 6 is a graph showing the results of single and double cropping rice distributions in the test study area in the example of the present invention.

Detailed Description

The present invention will be described in further detail with reference to examples for the purpose of facilitating the understanding and practice of the invention by those of ordinary skill in the art, and the examples described herein are intended to be illustrative and explanatory only and are not restrictive of the invention.

Example (b):

as shown in FIG. 1, the invention discloses a Chinese south single-double cropping rice phenological identification method based on multi-source remote sensing data, and a test research area is shown in FIG. 3. The detailed technical process schematic diagram is shown in fig. 2, and comprises:

step one, downloading and preprocessing time sequence images

And downloading the Sentinel-2, landsat-7 and 8 satellite images, and respectively carrying out image preprocessing, image cooperation and time synthesis processing to obtain a time sequence image data set. Specifically, the detailed operation steps of the first step are as follows:

data downloading is carried out by using a Google Earth Engine platform, and a Level-1C grade Sentinel-2 satellite image (TOA data set), a Landsat-7 satellite image (Collection 1Tier 1TOA data) and a Landsat-8 satellite image are downloaded respectively. And carrying out cloud, snow and rolling cloud removing treatment on the downloaded Sentinel-2, landsat-7 and Landsat-8 satellite images, and carrying out image cooperative treatment. And after the processing is finished, carrying out time synthesis processing on the Sentinel-2, landsat-7 and Landsat-8 satellite images, setting the size of a time synthesis window to be 10 days, and generating a synthetic image data set by using a median synthesis rule. And carrying out interpolation processing on the synthetic image data set by a time sequence linear interpolation method to remove the influence of a pixel null value on a subsequent filtering effect, and finally obtaining a time sequence image data set.

Step two, calculating different filtering vegetation index data sets

And calculating the vegetation index based on the time sequence image data set generated by calculation in the first step to obtain a ten-day synthetic vegetation index image data set. And performing time sequence processing on the ten-day synthetic vegetation index image data set by using S-G filtering and HANTS filtering to obtain an S-G filtering LSWI image, an S-G filtering NDVI image and an HANTS filtering NDVI image. Specifically, the detailed operation steps of the second step are as follows:

calculating a vegetation index on the basis of the time sequence image data set generated by calculation in the first step, calculating a vegetation index NDVI by using near infrared and red light wave bands, and calculating a vegetation index LSWI by using near infrared and short wave infrared wave bands to obtain a ten-day synthetic vegetation index image data set, which specifically comprises the following steps:

TABLE 1 vegetation index and formula

Where ρ is _NIR In the near infrared band, p _RED In the red wavelength band, p _SWIR Is short wave infrared band.

And performing time sequence filtering processing based on the vegetation index image data set. The specific processing flow is as follows, S-G filtering processing is carried out on the vegetation index image data set by using an sgolayfilt function in Matlab software to obtain an S-G filtering LSWI image and an S-G filtering NDVI image, and HANTS filtering processing is carried out on the vegetation index image data set by using an HANTS function developed by Mohammad Abouli based on Matlab software to obtain an HANTS filtering NDVI image, wherein the parameter setting is shown in Table 2. The filtering effect is shown in fig. 4.

TABLE 2HANTS Filtering and S-G Filtering parameter settings

Step three, non-tillage mask

The method comprises the following steps of performing identification of cultivated land pixels and non-cultivated land pixels in a research area based on a GLAD cultivated land product data set published by Peter Potapov in 2022, wherein the specific flow is as follows: masking the research area based on farmland pixel distribution in the GLAD farmland product data set to extract farmland pixels, removing the influence of non-farmland pixels on single and double cropping rice classification, and obtaining a farmland distribution map layer in the research area.

Step four, S-G filtering data set rice identification

And (3) after eliminating all non-cultivated land pixels in the research area by using cultivated land distribution image layers generated in the third step for calculating and generating S-G filtering LSWI images and S-G filtering NDVI images in the second step, distinguishing paddy rice from non-paddy rice by using the S-G filtering LSWI images and the S-G filtering NDVI images and combining the flooding characteristics of paddy rice in a paddy field soaking period, wherein the judgment standard is shown as a formula 3:

wherein i is a scene number serial number, i is more than 8 and less than 14, the scene number serial number corresponds to the rice flooding period, LSWIi is an ith scene S-G filtering LSWI image, and NDVIi is an ith scene S-G filtering NDVI image. When the Rice Index (Z) is 1, the pixel Z is potential Rice, and when the Rice Index (Z) is 0, the pixel Z is non-Rice. And performing Rice Index calculation on all pixels in the research area range to obtain a final Rice potential distribution map layer of the research area, wherein the Rice potential distribution map layer only contains Rice pixels.

Step five, HANTS filtering data set single and double cropping rice identification

And (4) eliminating all non-cultivated land pixels in the research area by using the cultivated land distribution layer generated in the third step for the HANTS filtering NDVI images calculated and generated in the second step, and then combining different growth phenological period characteristics of single and double cropping rice on the basis of the rice potential distribution layer of the research area generated in the fourth step to finish the identification of the single and double cropping rice. The judgment criteria are shown in formulas (4) and (5):

wherein i is a scene number serial number, i is more than 20 and less than 27 and corresponds to a single-season rice heading period, i is more than 16 and less than 21 and corresponds to an early rice heading period, i is more than 25 and less than 30 and corresponds to a late rice heading period, and Peaknumber is the peak number of pixel Z when the HANTS filtering NDVI image is in the i scene range; peakvalue is the maximum peak value of the pixel Z of the HANTS filtered NDVI image in the i scene range. On the basis of a potential distribution layer of the rice, when a pixel Z meets an SCR (Z) condition, judging the pixel Z to be single cropping rice; when the pixel Z meets the DCR (Z) condition, judging the pixel Z to be double cropping rice; and if the two conditions are not met, judging the pixel Z to be non-rice. And (4) performing single and double cropping rice identification calculation on all pixels of the potential distribution layer of the rice in the research area to obtain the final single and double cropping rice identification result of the research area.

To evaluate the optimization effect of this example over the general phenological method, the present invention evaluated producer accuracy, user accuracy, overall accuracy and F1-score of single and double cropping rice comparing the three methods, as shown in FIG. 5. Compared with the method of singly using the S-G filtering data lumped body, the precision of the method of the invention is improved by 32.95%, and compared with singly using the HANTS filtering data lumped body, the precision of the method of the invention is improved by 2.01%. The final classification result of the method of the present invention is shown in fig. 6.

It should be noted that the specific embodiments described herein are merely illustrative of the spirit of the invention. Various modifications or additions may be made to the described embodiments or alternatives may be employed by those skilled in the art without departing from the spirit or ambit of the invention as defined in the appended claims.

Claims

1. A south single-double cropping rice identification method based on multi-source data and phenological characteristics is characterized by comprising the following steps:

performing vegetation index calculation on the time sequence image data set to obtain a vegetation index image data set, and performing time sequence filtering processing on the basis of the vegetation index image data set to obtain an S-G filtering LSWI image, an S-G filtering NDVI image and an HANTS filtering NDVI image;

2. The method for identifying the southern single-double cropping rice based on the multi-source data and the phenological characteristics as claimed in claim 1, wherein the first step comprises the following steps:

3. The method for identifying the southern single and double cropping rice based on the multi-source data and the phenological characteristics as claimed in claim 2, wherein the third step comprises the following steps: and masking the research area based on farmland pixel distribution in the GLAD farmland product data set to extract farmland pixels so as to obtain a farmland distribution layer in the research area.

4. The south single-double cropping rice identification method based on multi-source data and phenological characteristics as claimed in claim 3, wherein in the fourth step, the rice and non-rice are distinguished by combining the flooding characteristics of the rice in the field soaking period, and the obtained potential distribution map layer of the rice is based on the following formula:

and when the Rice Index (Z) is 0, the pixel Z is represented as non-Rice, and further a Rice potential distribution map layer of the final research area is obtained.

5. The method for identifying single and double cropping rice in south China based on multi-source data and phenological characteristics as claimed in claim 4, wherein in said fifth step, HANTS filtering NDVI images and potential distribution layers of rice are used to complete the identification of single and double cropping rice based on the following formula:

/>

wherein i is the scene number serial number, peak _number The peak value number of the pixel Z of the HANTS filtered NDVI image in the i scene range; peak _value For the maximum peak value of the pixel Z of the HANTS filtered NDVI image in the i scene range, on the basis of a potential distribution layer of the rice, when the pixel Z meets the SCR (Z) condition, the pixel Z is judged to be single-cropping rice; when the pixel Z meets the DCR (Z) condition, judging the pixel Z to be double cropping rice; and if the SCR (Z) condition and the DCR (Z) condition are not met, judging that the pixel Z is non-rice, and obtaining the final single-double cropping rice identification result in the research area.