A Method and Device for Extracting Paddy Field Area Based on High-Resolution Remote Sensing Images

Abstract

The present invention discloses a method and device for extracting paddy field area based on high-resolution remote sensing images. The method includes: collecting remote sensing images during the rice growth period to obtain a first remote sensing image, and collecting remote sensing images after rice harvesting to obtain a second remote sensing image; preprocessing the first remote sensing image and the second remote sensing image to obtain a preprocessed first remote sensing image and a preprocessed second remote sensing image; performing temporal multi-feature difference image analysis on the preprocessed first remote sensing image and the preprocessed second remote sensing image to obtain temporal multi-feature difference image analysis; using the random forest algorithm to interpret the temporal multi-feature difference image, and evaluating the accuracy of the interpretation result, extracting paddy field information, and counting the paddy field area. The method constructs a temporal multi-feature difference image, uses methods such as random forest to extract paddy fields, reduces the data volume, improves the calculation speed, and realizes accurate identification of paddy fields.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of land monitoring, and particularly relates to a method and device for extracting paddy field area based on high-resolution remote sensing images. Background Technique

[0002] Monitoring the changes of paddy fields is a necessary path for estimating rice yields and an important means for evaluating the implementation effects of land improvement projects such as "reclamation of paddy fields". Remote sensing technology has the advantages of being fast, efficient, and objective, and has become the main way for paddy field monitoring;

[0003] For current paddy field monitoring technologies, the monitoring accuracy of medium- and low-resolution remote sensing images is low and the identification ability is poor. When using remote sensing images of Gaofen-2 (GF-2) and Sentinel-2, a single index or a small number of indexes of multi-band high-resolution remote sensing images are not sensitive to the differentiation of ground objects. High-resolution remote sensing images can clearly depict the details of ground objects, but for multi-feature images, they generate high-dimensional data, with a large amount of data and time-consuming calculations. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies of the prior art. The present invention provides a method and device for extracting paddy field area based on high-resolution remote sensing images. The method collects remote sensing images of Gaofen-2 and Sentinel-2, and obtains multiple index images with the largest differences in multiple time phases through difference analysis to distinguish paddy fields from other cultivated lands, improve the sensitivity of paddy field identification, perform principal component analysis on the images combined with multiple features, and then use methods such as random forest to extract paddy fields to reduce the amount of data and improve the calculation speed, so as to achieve accurate identification of paddy fields.

[0005] The present invention provides a method for extracting paddy field area based on high-resolution remote sensing images. The extraction method includes:

[0006] Collect remote sensing images during the rice growth period to obtain a first remote sensing image, and collect remote sensing images after rice harvesting to obtain a second remote sensing image. The first remote sensing image includes a first Sentinel-2 remote sensing image and a first GF-2 remote sensing image, and the second remote sensing image includes a second Sentinel-2 remote sensing image and a second GF-2 remote sensing image;

[0007] Perform preprocessing on the first remote sensing image and the second remote sensing image to obtain a preprocessed first remote sensing image and a preprocessed second remote sensing image;

[0008] Perform time-series multi-feature difference image analysis on the preprocessed first remote sensing image and the preprocessed second remote sensing image to obtain a time-series multi-feature difference image;

[0009] Interpret the time-series multi-feature difference image using the random forest algorithm, evaluate the accuracy of the interpretation results, extract paddy field information, and count the paddy field area.

[0010] Further, the preprocessing of the first remote sensing image and the second remote sensing image to obtain the preprocessed first remote sensing image and the preprocessed second remote sensing image includes:

[0011] Perform preprocessing on the first remote sensing image and the second remote sensing image in sequence, including radiometric correction, atmospheric correction, orthorectification, resampling, geometric registration, and image normalization, to obtain the preprocessed first remote sensing image and the preprocessed second remote sensing image.

[0012] Further, the time-series multi-feature difference image analysis of the preprocessed first remote sensing image and the preprocessed second remote sensing image to obtain the time-series multi-feature difference image includes:

[0013] Extract the first GF-2 remote sensing image of the preprocessed first remote sensing image, and extract the second GF-2 remote sensing image of the preprocessed second remote sensing image;

[0014] Perform difference analysis and multi-scale analysis on the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image to obtain GF-2 analysis data;

[0015] Extract the first Sentinel-2 remote sensing image of the preprocessed first remote sensing image, and extract the second Sentinel-2 remote sensing image of the preprocessed second remote sensing image;

[0016] Perform index analysis and difference analysis on the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image to obtain Sentinel-2 analysis data;

[0017] Based on the GF-2 analysis data and the Sentinel-2 analysis data, construct a time-series multi-feature difference image.

[0018] Further, performing difference analysis and multi-scale analysis on the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image to obtain GF-2 analysis data includes:

[0019] Perform difference analysis on the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image to obtain a GF-2 difference image;

[0020] Perform multi-scale analysis on the GF-2 difference image to obtain a GF-2 multi-scale analysis image.

[0021] Further, the index analysis includes normalized difference vegetation index analysis, normalized difference infrared index analysis, normalized difference moisture index analysis, red edge vegetation index analysis, red edge inflection point index analysis, brightness index analysis, and color index analysis.

[0022] Further, performing index analysis and difference analysis on the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image to obtain Sentinel-2 analysis data, including:

[0023] Performing index analysis on the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image to obtain index analysis data;

[0024] Extracting the maximum value image and the minimum value image of paddy field samples from the index analysis data;

[0025] Performing difference processing and multi-scale analysis on the maximum value image and the minimum value image of paddy field samples to obtain a time series index difference image.

[0026] Further, based on the GF-2 analysis data and the Sentinel-2 analysis data, constructing a time series multi-feature difference image, including:

[0027] Performing overlay analysis on the GF-2 analysis data and the Sentinel-2 analysis data to obtain multi-feature difference analysis data, and constructing a time series multi-feature difference image based on the multi-feature difference analysis data.

[0028] Further, the performing overlay analysis on the GF-2 analysis data and the Sentinel-2 analysis data to obtain multi-feature difference analysis data, and constructing a time series multi-feature difference image based on the multi-feature difference analysis data, including:

[0029] Taking the results of difference analysis of the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image, and performing principal component analysis on the Sentinel-2 analysis data to obtain principal component data;

[0030] Taking the results of multi-scale analysis of the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image, and combining with the principal component data to construct a time series multi-feature difference image.

[0031] Further, perform principal component analysis on the result of the difference analysis of the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image, and the Sentinel-2 analysis data to obtain principal component data, including:

[0032] Take the result of the difference analysis of the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image, perform principal component analysis, and take the first principal component quantity;

[0033] Take the result of the difference analysis of the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image, perform principal component analysis, and take the first two principal component quantities;

[0034] Take the result of the difference analysis of the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image, perform principal component analysis, and take the first three principal component quantities. The present invention also provides a paddy field area extraction device based on high-resolution remote sensing images, and the device includes:

[0035] Data acquisition module: Acquire remote sensing images during the rice growth period to obtain a first remote sensing image, and acquire remote sensing images after rice harvesting to obtain a second remote sensing image. The first remote sensing image includes a first Sentinel-2 remote sensing image and a first GF-2 remote sensing image, and the second remote sensing image includes a second Sentinel-2 remote sensing image and a second GF-2 remote sensing image;

[0036] Preprocessing module: Preprocess the first remote sensing image and the second remote sensing image to obtain a preprocessed first remote sensing image and a preprocessed second remote sensing image;

[0037] Image analysis module: Perform temporal multi-feature difference image analysis on the preprocessed first remote sensing image and the preprocessed second remote sensing image to obtain a temporal multi-feature difference image;

[0038] Statistical analysis module: Use the random forest algorithm to interpret the temporal multi-feature difference image, evaluate the accuracy of the interpretation result, extract paddy field information, and statistically analyze the paddy field area.

[0039] The present invention provides a paddy field area extraction method and device based on high-resolution remote sensing images. The method uses high-resolution remote sensing imaging technology, obtains multiple index images with the largest differences in multiple time phases through difference analysis to distinguish paddy fields from other cultivated lands, improves the sensitivity of paddy field recognition, performs principal component analysis on the image combined with multiple features, and then uses methods such as random forest to extract paddy fields to reduce the data volume and improve the calculation speed, realizing accurate recognition of paddy fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0041] Figure 1 It is a flowchart of the paddy field extraction method based on multi-feature differences of high-resolution remote sensing time series in the embodiment of the present invention;

[0042] Figure 2 It is the GF-2 remote sensing image after multi-scale analysis during the growth period of late rice in the embodiment of the present invention;

[0043] Figure 3 It is the GF-2 remote sensing image after multi-scale analysis after the late rice is harvested in the embodiment of the present invention;

[0044] Figure 4 It is a schematic diagram of the paddy field area extraction device based on high-resolution remote sensing images in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0046] Embodiment 1:

[0047] Figure 1 It shows a flowchart of the paddy field area extraction method based on high-resolution remote sensing images in the embodiment of the present invention. The paddy field extraction method based on high-resolution remote sensing images provided by the present invention includes:

[0048] S11: Collect remote sensing images during the growth period of rice to obtain the first remote sensing image, and collect remote sensing images after rice harvesting to obtain the second remote sensing image. The first remote sensing image includes the first Sentinel-2 remote sensing image and the first GF-2 remote sensing image, and the second remote sensing image includes the second Sentinel-2 remote sensing image and the second GF-2 remote sensing image. Specifically, extract the paddy field area of late rice growth in Nanxiong City, northern Guangdong Province, obtain remote sensing data of GF-2 and Sentinel-2, preprocess the remote sensing data, identify paddy fields according to the temporal feature differences of rice, and combine the growth period and harvesting period of rice. The differences in paddy fields in remote sensing images are the most obvious. Collect remote sensing images during the growth period of rice to obtain the first remote sensing image and collect remote sensing images after rice harvesting to obtain the second remote sensing image, making the differences in paddy field features obvious in the first remote sensing image and the second remote sensing image, which is conducive to improving the accuracy of paddy field identification.

[0049] Further, the first remote sensing image includes the first Sentinel-2 remote sensing image and the first GF-2 remote sensing image, and the second remote sensing image includes the second Sentinel-2 remote sensing image and the second GF-2 remote sensing image. The swath width of the Sentinel-2 remote sensing image is 290 km, which is suitable for remote sensing applications at the urban and provincial scales. Combining with the high-resolution satellite images of GF-2 can improve the accuracy of paddy field identification.

[0050] Further, in order to improve the accuracy of paddy field identification in remote sensing images, select remote sensing images with few clouds or no clouds during the growth period and harvesting period of rice.

[0051] S12: Preprocess the first remote sensing image and the second remote sensing image to obtain the preprocessed first remote sensing image and the preprocessed second remote sensing image.

[0052] Specifically, the preprocessing includes preprocessing the first remote sensing image and the second remote sensing image by radiometric correction, atmospheric correction, orthorectification, resampling, geometric registration, and image normalization to obtain the preprocessed first remote sensing image and the preprocessed second remote sensing image. The preprocessed first remote sensing image and the preprocessed second remote sensing image can reflect the true reflectance of ground objects, the coordinates of the same location on different remote sensing images are consistent, and the reflectance values of the same ground object are similar. The information is concentrated on 1 to 2 characteristic bands through principal component analysis.

[0053] S13: Perform temporal multi-feature difference image analysis on the preprocessed first remote sensing image and the preprocessed second remote sensing image to obtain a temporal multi-feature difference image.

[0054] Specifically, extract the first GF-2 remote sensing image from the preprocessed first remote sensing image, extract the second GF-2 image from the preprocessed second remote sensing image, perform difference processing on the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image to obtain a first GF-2 difference image, and perform multi-scale analysis on the first GF-2 difference image to obtain a second GF-2 difference image.

[0055] Specifically, the ground object boundary features of the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image are obvious. In the image after rice harvesting, the boundary features of the ground objects are clearer. Through difference processing and multi-scale analysis, the influence of the same spectrum of different objects in the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image is further reduced, and the extraction accuracy of the paddy field boundary is improved.

[0056] Specifically, Figure 2 Figure 8 shows the GF-2 remote sensing image after multi-scale analysis during the growth period of late rice in the embodiment of the present invention. Figure 3 Figure 10 shows the GF-2 remote sensing image after multi-scale analysis after late rice harvesting. After the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image are subjected to multi-scale analysis, the influence of the same spectrum of different objects is reduced, and the extraction accuracy of the paddy field boundary is improved.

[0057] Specifically, extract the first Sentinel-2 remote sensing image from the preprocessed first remote sensing image, extract the second Sentinel-2 remote sensing image from the preprocessed second remote sensing image, and perform index analysis on the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image to obtain an index analysis image.

[0058] Furthermore, the index analysis includes normalized difference vegetation index (NDVI) analysis, normalized difference infrared index (NDII) analysis, normalized difference moisture index (NDMI) analysis, red-edge vegetation index (REVI) analysis, red-edge inflection point index (VIP) analysis, brightness index (BI) analysis, and color index (CI) analysis.

[0059] The normalized difference vegetation index is obtained from the formula:

[0060]

[0061] where: NDVI is the normalized difference vegetation index, B nir is the infrared band of the image, and B red is the red band of the image.

[0062] The normalized difference infrared index is obtained from the formula:

[0063]

[0064] Among them: NDII is the Normalized Difference Infrared Index, B nir is the infrared band of the image, B swir1 is the shortwave infrared band of the image;

[0065] The Normalized Difference Moisture Index is obtained from the formula:

[0066]

[0067] Among them: NDMI is the Normalized Difference Moisture Index, B nir is the infrared band of the image, B green is the green band of the image.

[0068] The Red Edge Vegetation Index is obtained from the formula:

[0069]

[0070] Among them: VOG is the Red Edge Vegetation Index, B re1 is the first vegetation red edge band of the image, B re2 is the second vegetation red edge band of the image;

[0071] The Red Edge Inflection Point Index is obtained from the formula:

[0072]

[0073] Among them: REIP is the Red Edge Inflection Point Index, B re1 is the first vegetation red edge band of the image, B re2 is the second vegetation red edge band of the image, B re3 is the third vegetation red edge band of the image.

[0074] The Brightness Index is obtained from the formula:

[0075]

[0076] Among them: BI is the Brightness Index, B red is the red edge band of the image, B green is the green band of the image; The Color Index is obtained from the formula:

[0077]

[0078] Among them: CI is the Color Index, B red is the red band of the image, B green is the green band of the image.

[0079] Further, by comparing the index images during the growth period and after harvesting of rice, the maximum value image and the minimum value image of paddy field samples are extracted from the index analysis images, and the maximum value image and the minimum value image of the paddy field samples are subjected to difference processing and multi-scale analysis to obtain a time series index difference image. Specifically, the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image are subjected to difference processing to obtain a Sentinel-2 difference image.

[0080] Perform principal component analysis on the first GF-2 difference image, the Sentinel-2 difference image, and the time series index difference image, and extract principal component data, including:

[0081] Perform principal component analysis on the first GF-2 difference image and take the first principal component component;

[0082] Perform principal component analysis on the Sentinel-2 difference image and take the first two principal component components;

[0083] Perform principal component analysis on the time series index difference image and take the first three principal component components.

[0084] Combine the principal component data with the second GF-2 difference image to construct a time series multi-feature difference image, so that the images participating in paddy field extraction have the ability to accurately extract the paddy field boundary and high recognition sensitivity, and can be accurately distinguished from other agricultural areas.

[0085] Further, training samples for identifying paddy fields are established on the first remote sensing image and the second remote sensing image, including paddy fields, forest land, other agricultural land, bare land, artificial surfaces, and water areas. Visual discrimination is carried out in combination with experience, unified training samples are established on different images, and the separability analysis of the samples is carried out to ensure that the ground objects expressed by the samples at the same position are the same on different images and the training samples are optimal.

[0086] Analyze the separability of the training samples in combination with the time series multi-feature difference image, the time series index difference image, the first remote sensing image, and the second remote sensing image, and test the paddy field recognition accuracy of the time series multi-feature difference image through the comparison of the separability of the training samples in each image.

[0087] Specifically, the comparison results of the separability of the training samples in each image are shown in Table 1:

[0088] Table 1 Separability of training samples on different images

[0089]

[0090] It can be seen from Table 1 that the separability of the training samples in the time-series multi-feature difference image is high, and the accuracy of paddy field extraction is high.

[0091] S14: Use the random forest algorithm to interpret the time-series multi-feature difference image, evaluate the accuracy of the interpretation result, extract paddy field information, and count the paddy field area.

[0092] Furthermore, the random forest algorithm is an ensemble learning algorithm that can integrate multiple decision trees. It is used to process high-dimensional data and has a fast running speed. It also has a high classification accuracy for a small number of feature samples.

[0093] Specifically, use the random forest algorithm to classify the images in the growth period, the images after harvesting, the time-series index difference images, and the time-series multi-feature difference images respectively, extract the area of the paddy field, and use the on-site classification samples to test the classification accuracy and the accuracy of the area. Compare the accuracies of different images, and the accuracy evaluation results are shown in Table 2:

[0094] Table 2 Accuracy Evaluation of Different Types of Images

[0095]

[0096]

[0097] It can be found from Table 2 that the paddy field extraction accuracy of the time-series multi-feature difference image is the highest, the Kappa coefficient is 0.97, the user accuracy is 98%, and the difference between the paddy field area and the late rice sowing data statistically by the agricultural department is 240.05 hm 2 , that is, the accuracy of the time-series multi-feature difference image is high, meeting the requirements for paddy field area extraction.

[0098] The embodiment of the present invention provides a method for extracting paddy field area based on high-resolution remote sensing images. The high-resolution remote sensing image technology is adopted, and multiple index images with the largest differences in multiple time phases are obtained through difference analysis to distinguish paddy fields from other cultivated lands, improve the sensitivity of paddy field recognition, perform principal component analysis on the image combined with multiple features, and then use the random forest algorithm to extract paddy fields to reduce the data volume and improve the calculation speed, realizing the accurate recognition of paddy fields.

[0099] Embodiment 2:

[0100] Figure 4 The schematic diagram of the device for extracting paddy field area based on high-resolution remote sensing images in the embodiment of the present invention is shown. The device includes:

[0101] Data acquisition module 1: Collect remote sensing images during the growth period of rice to obtain the first remote sensing image, and collect remote sensing images after rice harvesting to obtain the second remote sensing image. The first remote sensing image includes the first Sentinel-2 remote sensing image and the first GF-2 remote sensing image, and the second remote sensing image includes the second Sentinel-2 remote sensing image and the second GF-2 remote sensing image;

[0102] Specifically, identify paddy fields according to the difference in the temporal characteristics of rice. Combining the growth period and the harvesting period of rice, the difference in paddy fields in remote sensing images is the most obvious. Collect remote sensing images during the growth period of rice to obtain the first remote sensing image and collect remote sensing images after rice harvesting to obtain the second remote sensing image, so that the difference in paddy field characteristics in the first remote sensing image and the second remote sensing image is obvious, which is conducive to improving the accuracy of paddy field identification.

[0103] Furthermore, the first remote sensing image includes the first Sentinel-2 remote sensing image and the first GF-2 remote sensing image, and the second remote sensing image includes the second Sentinel-2 remote sensing image and the second GF-2 remote sensing image. The width of the Sentinel-2 remote sensing image is 290 km, which is suitable for remote sensing applications at the city scale and the provincial scale. Combining with the high-resolution satellite image of GF-2 can improve the accuracy of paddy field identification.

[0104] Furthermore, in order to improve the accuracy of paddy field identification in remote sensing images, select remote sensing images with few clouds or no clouds during the growth period and the harvesting period of rice.

[0105] Preprocessing module 2: Preprocess the first remote sensing image and the second remote sensing image to obtain the preprocessed first remote sensing image and the preprocessed second remote sensing image;

[0106] Specifically, the preprocessing includes sequentially performing radiometric correction, atmospheric correction, orthorectification, resampling, geometric registration, and image normalization on the first remote sensing image and the second remote sensing image to obtain the preprocessed first remote sensing image and the preprocessed second remote sensing image. The preprocessed first remote sensing image and the preprocessed second remote sensing image can reflect the true reflectance of ground objects, the coordinates of the same location on different remote sensing images are consistent, and the reflectance values of the same ground object are similar. The information is concentrated on 1 to 2 characteristic bands through principal component analysis.

[0107] Image analysis module 3: Perform temporal multi-feature difference image analysis on the preprocessed first remote sensing image and the preprocessed second remote sensing image to obtain a temporal multi-feature difference image;

[0108] Specifically, extract the first GF-2 remote sensing image from the preprocessed first remote sensing image, extract the second GF-2 image from the preprocessed second remote sensing image, perform difference processing on the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image to obtain a first GF-2 difference image, and perform multi-scale analysis on the first GF-2 difference image to obtain a second GF-2 difference image.

[0109] Specifically, the ground object boundary features of the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image are obvious. In the image after rice harvesting, the boundary features of the ground objects are clearer. Through difference processing and multi-scale analysis, the influence of the same spectrum of foreign objects in the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image is further reduced, and the extraction accuracy of the paddy field boundary is improved.

[0110] Specifically, extract the first Sentinel-2 remote sensing image from the preprocessed first remote sensing image, extract the second Sentinel-2 remote sensing image from the preprocessed second remote sensing image, and perform index analysis on the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image to obtain an index analysis image.

[0111] Furthermore, by comparing each index image during the rice growth period and after harvesting, extract the maximum value image of the paddy field sample and the minimum value image of the paddy field sample from the index analysis image, and perform difference processing and multi-scale analysis on the maximum value image of the paddy field sample and the minimum value image of the paddy field sample to obtain a time series index difference image.

[0112] Specifically, perform difference processing on the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image to obtain a Sentinel-2 difference image.

[0113] Perform principal component analysis on the first GF-2 difference image, the Sentinel-2 difference image, and the time series index difference image to extract principal component data, including:

[0114] Perform principal component analysis on the first GF-2 difference image and take the first principal component component;

[0115] Perform principal component analysis on the Sentinel-2 difference image and take the first two principal component components;

[0116] Perform principal component analysis on the time series index difference image and take the first three principal component components.

[0117] Combine the principal component data with the second GF-2 difference image to construct a time-series multi-feature difference image, enabling the images involved in paddy field extraction to have the ability to accurately extract paddy field boundaries and high recognition sensitivity, and being able to accurately distinguish from other agricultural areas.

[0118] Further, establish training samples for identifying paddy fields on the first remote sensing image and the second remote sensing image, including paddy fields, forest land, other agricultural land, bare land, artificial surfaces, and water areas. Conduct visual discrimination in combination with experience, establish unified training samples on different images, and perform separability analysis of the samples to ensure that the ground objects expressed by the samples at the same location are the same on different images and the training samples are optimal.

[0119] Analyze the separability of the training samples in combination with the time-series multi-feature difference image, the time-series index difference image, the first remote sensing image, and the second remote sensing image. Through the comparison of the separability of the training samples in each image, test the paddy field recognition accuracy of the time-series multi-feature difference image, and obtain that the separability of the training samples in the time-series multi-feature difference image is high and the paddy field recognition accuracy of the time-series multi-feature difference image is high.

[0120] Statistical analysis module 4: Use the random forest algorithm to interpret the time-series multi-feature difference image, evaluate the accuracy of the interpretation result, extract paddy field information, and count the paddy field area.

[0121] Specifically, use the random forest algorithm to interpret the first remote sensing image, the second remote sensing image, the time-series index difference image, and the time-series multi-feature difference image, and compare the interpretation results. The paddy field boundary recognition accuracy of the time-series multi-feature difference image is high, and it can accurately identify the location of the paddy field.

[0122] The embodiment of the present invention provides a paddy field area extraction device based on high-resolution remote sensing images. The device extracts high-resolution remote sensing images, performs time-series multi-feature difference image processing, obtains time-series multi-feature images, improves paddy field recognition accuracy, and uses the random forest algorithm to extract paddy fields to reduce the data volume and improve the calculation speed, realizing accurate recognition of paddy fields.

[0123] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The storage medium can include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, etc.

[0124] In addition, the embodiments of the present invention have been introduced in detail above. Specific examples have been used in this article to expound the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for extracting paddy field area based on high - resolution remote sensing images, characterized in that, the extraction method includes: Collect remote sensing images during the rice growth period to obtain the first remote sensing image, and collect remote sensing images after rice harvesting to obtain the second remote sensing image. The first remote sensing image includes the first Sentinel - 2 remote sensing image and the first GF - 2 remote sensing image, and the second remote sensing image includes the second Sentinel - 2 remote sensing image and the second GF - 2 remote sensing image; Pre - process the first remote sensing image and the second remote sensing image to obtain the pre - processed first remote sensing image and the pre - processed second remote sensing image; Perform temporal multi - feature difference image analysis on the pre - processed first remote sensing image and the pre - processed second remote sensing image to obtain a temporal multi - feature difference image; Use the random forest algorithm to interpret the temporal multi - feature difference image, evaluate the accuracy of the interpretation result, extract paddy field information, and count the paddy field area; The step of performing temporal multi - feature difference image analysis on the pre - processed first remote sensing image and the pre - processed second remote sensing image to obtain a temporal multi - feature difference image includes: Extract the first GF - 2 remote sensing image of the pre - processed first remote sensing image, and extract the second GF - 2 remote sensing image of the pre - processed second remote sensing image; Perform difference analysis and multi - scale analysis on the pre - processed first GF - 2 remote sensing image and the pre - processed second GF - 2 remote sensing image to obtain GF - 2 analysis data; Extract the first Sentinel - 2 remote sensing image of the pre - processed first remote sensing image, and extract the second Sentinel - 2 remote sensing image of the pre - processed second remote sensing image; Perform index analysis and difference analysis on the pre - processed first Sentinel - 2 remote sensing image and the pre - processed second Sentinel - 2 remote sensing image to obtain Sentinel - 2 analysis data; Based on the GF - 2 analysis data and the Sentinel - 2 analysis data, construct a temporal multi - feature difference image; The step of constructing a temporal multi - feature difference image based on the GF - 2 analysis data and the Sentinel - 2 analysis data includes: Perform overlay analysis on the GF - 2 analysis data and the Sentinel - 2 analysis data to obtain multi - feature difference analysis data, and construct a temporal multi - feature difference image based on the multi - feature difference analysis data; Take the result of the difference analysis of the pre - processed first GF - 2 remote sensing image and the pre - processed second GF - 2 remote sensing image, and perform principal component analysis on the Sentinel - 2 analysis data to obtain principal component data; Take the result of the multi - scale analysis of the pre - processed first GF - 2 remote sensing image and the pre - processed second GF - 2 remote sensing image, and combine it with the principal component data to construct a temporal multi - feature difference image.

2. The paddy field area extraction method according to claim 1, characterized in that, Performing preprocessing on the first remote sensing image and the second remote sensing image to obtain a preprocessed first remote sensing image and a preprocessed second remote sensing image, including: Performing preprocessing on the first remote sensing image and the second remote sensing image in sequence, including radiometric correction, atmospheric correction, orthorectification, resampling, geometric registration, and image normalization, to obtain a preprocessed first remote sensing image and a preprocessed second remote sensing image.

3. The paddy field area extraction method according to claim 1, characterized in that Performing difference analysis and multi-scale analysis on the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image to obtain GF-2 analysis data, including: Performing difference analysis on the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image to obtain a GF-2 difference image; Performing multi-scale analysis on the GF-2 difference image to obtain a GF-2 multi-scale analysis image.

4. The paddy field area extraction method according to claim 1, characterized in that The index analysis includes normalized difference vegetation index analysis, normalized difference infrared index analysis, normalized difference moisture index analysis, red-edge vegetation index analysis, red-edge inflection point index analysis, brightness index analysis, and color index analysis.

5. The paddy field area extraction method according to claim 1, characterized in that Performing index analysis and difference analysis on the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image to obtain Sentinel-2 analysis data, including: Performing index analysis on the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image to obtain index analysis data; Extracting the maximum value image of paddy field samples and the minimum value image of paddy field samples of the index analysis data; Performing difference processing and multi-scale analysis on the maximum value image of paddy field samples and the minimum value image of paddy field samples to obtain a time-series index difference image.

6. The paddy field area extraction method according to claim 1, characterized in that Taking the results of difference analysis of the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image, and performing principal component analysis on the Sentinel-2 analysis data to obtain principal component data, including: Taking the results of difference analysis of the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image, performing principal component analysis, and taking the first principal component; Taking the results of difference analysis of the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image, performing principal component analysis, and taking the first two principal components; Taking the results of difference analysis of the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image, performing principal component analysis, and taking the first three principal components.

7. A paddy field area extraction device based on high-resolution remote sensing images, It is characterized in that the device includes: a data acquisition module: acquiring remote sensing images during the growth period of rice to obtain a first remote sensing image, and acquiring remote sensing images after rice harvesting to obtain a second remote sensing image, where the first remote sensing image includes a first Sentinel-2 remote sensing image and a first GF-2 remote sensing image, and the second remote sensing image includes a second Sentinel-2 remote sensing image and a second GF-2 remote sensing image; a preprocessing module: preprocessing the first remote sensing image and the second remote sensing image to obtain a preprocessed first remote sensing image and a preprocessed second remote sensing image; an image analysis module: performing time-series multi-feature difference image analysis on the preprocessed first remote sensing image and the preprocessed second remote sensing image to obtain a time-series multi-feature difference image; extracting the first GF-2 remote sensing image of the preprocessed first remote sensing image, and extracting the second GF-2 remote sensing image of the preprocessed second remote sensing image; performing difference analysis and multi-scale analysis on the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image to obtain GF-2 analysis data; extracting the first Sentinel-2 remote sensing image of the preprocessed first remote sensing image, and extracting the second Sentinel-2 remote sensing image of the preprocessed second remote sensing image; performing index analysis and difference analysis on the preprocessed first Sentinel-2 remote sensing image and the preprocessed second Sentinel-2 remote sensing image to obtain Sentinel-2 analysis data; constructing a time-series multi-feature difference image based on the GF-2 analysis data and the Sentinel-2 analysis data; The constructing a time-series multi-feature difference image based on the GF-2 analysis data and the Sentinel-2 analysis data includes: performing overlay analysis on the GF-2 analysis data and the Sentinel-2 analysis data to obtain multi-feature difference analysis data, and constructing a time-series multi-feature difference image based on the multi-feature difference analysis data; taking the result of difference analysis of the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image, and performing principal component analysis on the Sentinel-2 analysis data to obtain principal component data; taking the result of multi-scale analysis of the preprocessed first GF-2 remote sensing image and the preprocessed second GF-2 remote sensing image, and combining the principal component data to construct a time-series multi-feature difference image; a statistical analysis module: using the random forest algorithm to interpret the time-series multi-feature difference image, evaluating the accuracy of the interpretation result, extracting paddy field information, and counting the paddy field area.

Images