Extraction method of wild rice planting area based on multi-temporal remote sensing images

Through multi-time phase remote sensing imaging and decision tree classification methods, the problem of low accuracy of extracting water chestnut planting area in Huangyan District is solved, high-precision water chestnut identification and area estimation are achieved, agricultural decision-making is supported, and accurate data support is provided for vegetable planting.

CN113283281BActive Publication Date: 2024-12-03ZHONGKE HEXIN REMOTE SENSING TECH (SUZHOU) CO LTD +1
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202110220713.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-02-26
Publication Date
2024-12-03
Estimated Expiration
2041-02-26

AI Technical Summary

Technical Problem

The existing technology is difficult to accurately identify and extract the planting area of ​​wild rice, especially in the case of high-fragmentation cultivated land and diverse crop species in Huangyan area, which leads to low statistical accuracy and time-consuming and labor-consuming, and few research on remote sensing of wild rice rice, lack of effective multi-temporal phases. Data application.

Method used

Using the decision tree classification method based on multi-time phase remote sensing images, the decision tree model is constructed to distinguish the water chestnut by obtaining red, green, blue, and near-infrared band images for atmospheric correction, and NDVI and NDWI indexes. Estimate its area.

Benefits of technology

High-precision extraction of water chestnut planting area has been achieved, with user accuracy and mapping accuracy reaching 96.00% and 94.00% respectively, with a Kappa coefficient of 0.93. It effectively identifies water chestnut distribution and supports vegetable planting subsidies and production layout planning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113283281B_ABST
    Figure CN113283281B_ABST
Patent Text Reader

Abstract

The present invention provides a method for extracting the water bamboo planting area based on multi-temporal remote sensing images. The Sentinel-2 time-series images covering the entire growth period of crops are used to process and generate NDVI time-series spectra. According to the growth characteristics of water bamboo, a decision tree classification method is used to effectively distinguish water bamboo from other ground objects. The overall accuracy of crop area extraction is high. The results verify that the decision tree classification results that combine planting characteristics with multi-temporal spectral index characteristics are more consistent with the actual distribution of ground objects.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of agricultural remote sensing, and in particular relates to a method for extracting the planting area of water bamboo based on multi-temporal remote sensing images. Background Technique

[0002] Water bamboo (Zizania altifolia) is a perennial shallow-water herb of the Gramineae family and the genus Zizania, and is one of the "eight aquatic delicacies" in China. It is mainly cultivated in the marsh and paddy fields in the Yangtze River Basin and the south of China. In Huangyan District, Taizhou City, Zhejiang Province, a set of relatively advanced and practical greenhouse water bamboo cultivation techniques suitable for the characteristics of Huangyan District has gradually formed. The "Huangyan greenhouse water bamboo" has tender and white flesh and bright skin color, and has excellent commercial properties. The water bamboo in this area has become a registered product of national agricultural product geographical indication. Double-season water bamboo is a major characteristic industry in Huangyan agriculture. At present, the planting area has exceeded 4,500 hm2, and it has the largest facility water bamboo planting base in the country. Timely and accurate acquisition of the water bamboo planting area can coordinate the structure of the vegetable planting area and the food crop planting area, provide reference for agricultural decision-making, and at the same time provide a basis for vegetable subsidies.

[0003] The water bamboo in Huangyan District is generally planted in two seasons. The water bamboo harvested in summer is called "summer water bamboo", and the water bamboo harvested in autumn is called "autumn water bamboo". The planting time of summer water bamboo is from November of the previous year to early April of the following year. The cultivation techniques of facility agricultural film covering and soil covering for protecting water bamboo are adopted, and it is generally harvested around the Tomb-Sweeping Festival (see "Journal of Zhejiang Agricultural Sciences", 2016.57(10): pp. 1644-1646). At the same time, the greenhouse is covered with film around late December, and the film is generally removed in late March. Basically, all the agricultural films are removed after the Tomb-Sweeping Festival ( "Yangtze River Vegetables", 2009(16): pp. 102-103). The planting time of autumn water bamboo is from mid-June to November. When planting autumn water bamboo, land preparation is required. Generally, field irrigation is carried out in May-June after the harvest of summer water bamboo, and then autumn water bamboo is planted from late June to early July. It is generally harvested in November. When harvesting, more strong autumn tillers should also be reserved to increase the quality and yield of early-maturing spring and summer water bamboo.

[0004] At present, the method for obtaining the water bamboo area mainly relies on the traditional manual investigation method, which not only takes a lot of time, but also requires a large amount of manpower and material resources. In the process of manual investigation, due to some subjective factors, such as statistical errors, inconsistent standards in different regions, different measurement tools, etc., the accuracy of statistical investigation will inevitably be affected. Remote sensing can conduct large-area synchronous monitoring, and has high timeliness. Using remote sensing to count the water bamboo planting area is not only more accurate than the traditional method in the statistical results, but also has less economic investment and is not restricted by regions.

[0005] At present, most of the research on Zizania latifolia is focused on physiological and biochemical aspects such as cultivation measures and occurrence of pests and diseases. There are few studies on remote sensing extraction of Zizania latifolia, and the extracted crops are mainly concentrated in field crops such as wheat, rice, and corn.

[0006] In terms of remote sensing classification methods, there are two types: supervised classification and unsupervised classification. Supervised classification is to select feature parameters based on samples provided by known training areas, use the trained feature parameters as decision rules, and classify each image to be classified according to this rule. Unsupervised classification is based on the class differences of different image objects in the feature space, and is a method of image classification according to known category standards. Quan Jing et al. used the unsupervised classification method to extract the distribution of rice planting area in the Dabie Mountain area. This method achieved real-time dynamic monitoring of the rice planting area in the study area (see "Chinese Agricultural Science Bulletin", 2019.35(01): pp. 104-111). Mai Li Su took Ulanhot City, Xing'an League, Inner Mongolia Autonomous Region as the research area, and compared three supervised classification methods, including support vector machine method, maximum likelihood method, and object-oriented classification method, with the hierarchical clustering method of unsupervised classification. The study showed that the supervised classification method has higher extraction accuracy than the unsupervised classification method ("Journal of Northern Agriculture", 2019.47(05): pp. 119-126). Li Zhongyuan and others used NDVI time series data and decision tree classification methods to monitor rapeseed area. The study showed that the identification results can effectively remove interference types such as non-vegetation and woodland, and then extract the planting area and distribution of rapeseed (Journal of Geo-Information Science, 2019).

[0007] In addition, most of the crop extraction features are identified using a single feature or a combination of multiple features. Zhao Zhenguo et al. analyzed the spectral characteristics of rice canopy and other types of ground features, and determined the best phase for remote sensing extraction of rice planting information by comparison (Modern Agricultural Science and Technology, 2015(15): pp. 258-259). However, data from other phases were not used, and further research is needed compared with the multi-phase method for rice extraction. Time series remote sensing images not only have the spectral information of a single phase image, but also have a series of time information, which is of great significance in the extraction of crop distribution information. At present, many scholars have used multi-phase data for remote sensing monitoring.

[0008] At present, the research on large-scale extraction of crop planting distribution mainly includes the following aspects: first, the research on crop remote sensing recognition ability under different remote sensing image conditions; second, the comparative study of crop recognition ability and accuracy based on different classification methods; third, the combination of multi-temporal images and multi-features to assist crop classification. However, remote sensing identification of crops is mainly focused on major crops, and there are few related studies on remote sensing identification of wild rice. The degree of fragmentation of cultivated land in Huangyan District is high, and the crop species are rich. Many crops have the phenomenon of "same object, different spectrum", which will interfere with the extraction of wild rice. In the past, the acquisition of crop information was mainly based on optical satellite data. For optical remote sensing, the crop distribution is judged by analyzing the changes in the spectral reflectance of the current image, and the crops are extracted by constructing the vegetation index. Its limitation is that in the same period, there is a phenomenon of "different objects, same spectrum" between wild rice and other crops. Simply based on single-temporal images and spectral features, it is not possible to distinguish between different objects well.

[0009] How to effectively and accurately identify the distribution and area of ​​wild rice stems to provide support for vegetable planting subsidies and production layout planning is an urgent problem facing this field. Summary of the invention

[0010] In order to solve the above-mentioned deficiencies in the prior art, the object of the present invention is to provide a method for extracting the double-season water chestnut planting area, which can accurately obtain the area information and spatial distribution of water chestnuts.

[0011] Specifically, the present invention provides a method for extracting wild rice stem planting area based on multi-temporal remote sensing images, comprising the following steps:

[0012] S1 obtains remote sensing multispectral images of Zizania latifolia during its main growth period in the study area, extracts images containing red, green, blue, and near-infrared bands, and then performs atmospheric correction on the images;

[0013] S2 constructs a decision tree classification model for classification based on the differences in spectral index characteristics, vegetation pattern characteristics, and color characteristics between wild rice and confusing crops;

[0014] S3 builds a recognition model to distinguish wild rice stems and estimate their area.

[0015] In the above method, preferably, the remote sensing multispectral image of the main growth period of Zizania latifolia in the study area obtained in S1 is image data provided by the Sentinel-2 satellite; more preferably, the central wavelength of the blue light band is 0.49 μm, and the resolution is better than 10 m; the central wavelength of the green light band is 0.56 μm, and the resolution is better than 10 m; the central wavelength of the red light band is 0.665 μm, and the resolution is better than 10 m; the central wavelength of the near-infrared band is 0.842 μm, and the resolution is better than 10 m.

[0016] In Huangyan District, wild rice is generally planted in two seasons. The wild rice harvested in summer is called "summer wild rice", and the wild rice harvested in autumn is called "autumn wild rice". The identification basis of the vegetation pattern characteristics described in S2 includes:

[0017] (1) For plots of land planted with summer wild rice, the fields should be covered with greenhouse plastic film from November of the previous year to March or April of the following year before the summer wild rice is harvested;

[0018] (2) After the summer rice harvest and before the autumn rice planting, it is the field irrigation stage;

[0019] (3) During the autumn rice harvest stage, the color characteristics of the rice are greener than those of the late rice harvested at the same time.

[0020] Further, according to the color features of wild rice, rice, fruit trees, and woodlands in the sentinel image, the identification of the color features in S2 includes:

[0021] Visual interpretation was performed using the grayish-white color of the wild rice field in February;

[0022] Extract all plots with water and covered with duckweed in June from the images in June;

[0023] The dark green index characteristics of water chestnuts in the field in October were used to further screen out water chestnut-planting plots, and non-water chestnut-planting plots were excluded based on the October images.

[0024] Furthermore, the spectral index characteristics of S2 include:

[0025] Calculate the NDVI and NDWI vegetation indices on images from February to November;

[0026] Among them, NDVI is the normalized vegetation index, which is the ratio of the difference between the near-infrared band and the visible red band and the sum of the values ​​of these two bands; NDVI can reflect the background influence of the plant canopy and is related to vegetation coverage;

[0027] The calculation formula of NDVI is: NDVI = (NIR-R) / (NIR+R);

[0028] NDWI is a normalized ratio index based on the green band and the near-infrared band. It is used to extract water information from images. The calculation formula of NDWI is: NDWI = (G-NIR) / (G+NIR).

[0029] Among them, NIR is the infrared band value, R is the visible light red band value, and G is the green band value.

[0030] Furthermore, the step S3 of constructing the recognition model includes:

[0031] S31: Use NDVI to exclude non-vegetation and man-made buildings;

[0032] S32: Set thresholds using NDWI and NDVI to extract the water body range in June;

[0033] S33: Combine with the NDVI in October to distinguish water bamboo;

[0034] More preferably, the construction of the recognition model in S3 includes:

[0035] S31 Extract the image map in September, set the threshold NDVI > 0.6. When the condition is false, the pixel is marked as non-vegetation; when the condition is true, the pixel is marked as vegetation, and then enter the judgment of S32;

[0036] S32: Extract the image map in June, set the threshold 0.4 < NDVI < 0.6. When the condition is false, the pixel is marked as other vegetation; when the condition is true, then enter the judgment of S33;

[0037] S33: Extract the image map in October, set the threshold NDVI > 0.6. When the condition is false, the pixel is marked as other vegetation; when the condition is true, the pixel is marked as water bamboo.

[0038] Furthermore, in the above method, it also includes using the confusion matrix method to evaluate the position accuracy; preferably, the use of the confusion matrix method to evaluate the position accuracy includes: creating random verification points in the study area, with 900 random points. Visually interpret each random point in combination with high-resolution images and field conditions, and compare it with the remote sensing extraction results to form a confusion matrix, and calculate its user accuracy, mapping accuracy, and Kappa coefficient respectively.

[0039] The present invention provides a method that uses remote sensing time-series images covering the entire growth period of the cover crop, processes and generates an NDVI time-series spectrum. According to the growth characteristics of water bamboo, combined with the decision tree classification method, it can effectively distinguish water bamboo from other ground objects, and the overall accuracy of crop area extraction is relatively high. The verification results show that by combining planting characteristics and multi-temporal spectral index characteristics, the decision tree classification results based on these two characteristics are more in line with the actual ground object distribution, verifying the feasibility of remote sensing extraction of water bamboo.

[0040] The method provided by the present invention provides relevant basis for the investigation of water bamboo using high-resolution remote sensing data, and provides support for vegetable planting subsidies and production layout planning.

[0041] The decision tree model constructed by combining multi-temporal spectral characteristics provided by the present invention can effectively identify the water bamboo planting area. The user accuracy and mapping accuracy of water bamboo are 96.00% and 94.00% respectively, and the Kappa coefficient is 0.93. The user accuracy and mapping accuracy of other ground objects are 95.00% and 93.00% respectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The technical solution of the present application is further described below in conjunction with the accompanying drawings and embodiments.

[0043] Figure 1 is a location map of the research area in the embodiment of the present application;

[0044] Figure 2 This is a diagram of the wild rice stem planting pattern in the embodiment of the present application;

[0045] Figure 3 This is a color characteristic diagram of double-season wild rice stem in the embodiment of the present application;

[0046] Figure 4 It is a multi-temporal NDVI curve diagram of vegetation from February to November in the embodiment of the present application;

[0047] Figure 5 It is a multi-temporal NDWI curve diagram of vegetation from February to November in the application embodiment, wherein 5(a) is a multi-temporal NDWI curve diagram of various vegetation, and 5(b) is a multi-temporal NDWI curve diagram of wild rice stem;

[0048] Figure 6 This is a technical flow chart of Zizania latifolia extraction.

[0049] Figure 7 : is a spatial distribution diagram of wild rice stem in the embodiment. Figure 7 (a) is the overall distribution of wild rice stem within the scope of the embodiment, Figure 7 (b) is an enlarged view of a local area within the scope of the embodiment, Figure 7 (c) is the planting situation of wild rice stem in the local area.

[0050] Figure 8 It is the user accuracy and mapping accuracy map of wild rice. DETAILED DESCRIPTION

[0051] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application may be combined with each other.

[0052] Example

[0053] The following is the eastern part of Huangyan District, Taizhou City, Zhejiang Province, with longitude and latitude 28°35′N-28°45′N, 121°7′E-121°12′E, located in Toudao Town and Xinqian Street as the research area (such as Figure 1 As shown), the technical scheme and beneficial effects of the present invention are illustrated by using multi-view Sentinel-2 images from February 2020 to November 2020.

[0054] Huangyan District is located in the central part of Taizhou City. The terrain of the area is high in the west and low in the east. There are many high mountains in the west and some plains in the east, which belong to the Wenhuang Plain. The total area of ​​the district is 988 square kilometers. The climate is mild and humid, with abundant rainfall, suitable sunlight and four distinct seasons. It belongs to the subtropical monsoon climate. The average temperature of the district is 17℃ for many years, the continuous sunshine days are 247.9 days, the average annual sunshine hours are 1955 hours, and the rainfall distribution is from west to east, decreasing with the decrease of altitude.

[0055] S1 obtains remote sensing multispectral images of Zizania latifolia during its main growth period in the study area, extracts images containing red, green, blue, and near-infrared bands, and then performs atmospheric correction on the images;

[0056] Sentinel-2 carries a multispectral imager that covers 13 spectral bands, ranging from visible light, near infrared to shortwave infrared. It is the only satellite with three bands in the red edge range, with a maximum spatial resolution of 10m and a swath width of 290km. When the two satellites work together, the temporal resolution can be increased to 5d.

[0057] The data products provided by the Sentinel-2 satellite are Level-1C and Level-2A. Level-1C is the apparent reflectance data of the top of the atmosphere after radiation correction and geometric correction, and Level-2A is the surface reflectance data after atmospheric correction. This embodiment uses the atmospherically corrected Level-2A data to effectively eliminate the influence of factors such as clouds, atmosphere and illumination on the reflection of ground objects. The image phase includes the main growth period of Zizania latifolia, from February 2020 to November 2020, with a total of 10 images. Four bands are selected in this paper, and the range and resolution of each band are shown in Table 1.

[0058] Table 1 Sentinel-2 image bands and resolutions selected in this embodiment

[0059] Band Central wavelength(um) Resolution(m) Blu-ray 0.490 10 Green Light 0.560 10 Red light 0.665 10 Near Infrared 0.842 10

[0060] S2 constructs a decision tree classification model for classification based on the differences in spectral index characteristics, vegetation pattern characteristics, and color characteristics between wild rice and confusing crops;

[0061] Firstly, the planting pattern characteristics of Zizania latifolia were analyzed. On this basis, the spectral index characteristics, vegetation pattern characteristics, color characteristics, etc. of Zizania latifolia were analyzed to provide a basis for feature selection for extracting Zizania latifolia. Then, according to the differences in spectral reflectance characteristics and vegetation index characteristics between Zizania latifolia and confusing crops, a decision tree classification model was constructed for classification.

[0062] Planting pattern characteristics

[0063] In Huangyan District, wild rice is usually planted in two seasons. The wild rice harvested in summer is called "summer wild rice", and the wild rice harvested in autumn is called "autumn wild rice". The planting time of summer wild rice is from November of the previous year to early April of the following year. It adopts the technology of covering cultivation with agricultural film and cultivating soil to protect wild rice production. It is generally harvested around the Qingming Festival. At the same time, the greenhouse film covering time is in late December, and the film is generally removed in late March. Basically, all agricultural films are removed after Qingming Festival. The planting time of autumn wild rice is from mid-June to November. When planting autumn wild rice, land preparation is required. Generally, field irrigation is carried out in May and June after the harvest of summer wild rice, and then autumn wild rice is planted from late June to early July. It is generally harvested in November. When harvesting, more strong autumn tillering seedlings should be retained to increase the quality and yield of early-maturing spring and summer wild rice.

[0064] The process of planting wild rice Figure 2 As shown in the figure, it can be divided into the rice stem planting stage and the land preparation stage. The following three points are summarized from the rice stem planting pattern as the basis for visual interpretation and feature recognition. They are: First, for the plots planted with summer rice stems, the fields are covered with greenhouse agricultural film from November of the previous year to March-April of the following year before the summer rice stems are harvested; second, after the summer rice stems are harvested and before the autumn rice stems are planted, it is the field irrigation stage; third, during the autumn rice stem harvest stage, compared with the late rice harvested at the same time, the color characteristics of the autumn rice stems are greener than those of the late rice.

[0065] Color Features

[0066] The main crops in Huangyan District are rice and wild rice, and the main fruit trees are orange trees and loquat trees. Rice in Huangyan District is generally planted in the morning and evening, and there is a "different species with the same spectrum" phenomenon with wild rice in most periods. The color characteristics of fruit trees and woodlands do not change much throughout the year. According to the color characteristics of wild rice, rice, fruit trees, and woodlands in the Sentinel images, it is concluded that the plots where wild rice was planted in February were covered with agricultural film, and they appeared grayish white from the image. In February, early rice had not yet been planted, and the fields were bare in a soil-gray state. Fruit trees and woodlands are evergreen all year round, and the woodlands are greener than fruit trees. Therefore, the grayish-white characteristics of the wild rice plots in February can be used for visual interpretation.

[0067] In June, the summer wild rice was just harvested and appeared dark green on the remote sensing images. The color of fruit trees and woodlands was not much different from that in February. The images in June were the key phase for identifying wild rice fields. The characteristics of the wild rice fields covered with duckweed during land preparation and irrigation were used to extract all the plots with water and duckweed in June. The extraction results at this time included the wild rice planted plots and other plots with water in the fields in June. In October, the wild rice was dark green in the field, which was obviously greener than the late rice that was in the mature harvest period. At this time, the wild rice planted plots can be further screened out based on the results extracted in June and the index characteristics in October, and the non-wild rice planted plots can be excluded based on the October images.

[0068] Spectral index characteristics

[0069] Based on the multi-temporal Sentinel-2 multispectral reflectance images, combined with Google images and the results of actual field surveys, a total of 2103 sample points were selected in the study area, including 177 fruit tree sample points, 673 buildings, 790 wild rice sample points, 289 woodlands, and 174 rice fields. The sample points are distributed as follows: Figure 1 As shown in Figure 2, buildings can be removed using the Normalized Difference Vegetation Index.

[0070] NDVI (Normalized Difference Vegetation Index) is defined as the ratio of the difference between the near infrared band (NIR) and the visible red band (RED) to the sum of the values ​​of these two bands. It can reflect the background influence of the plant canopy and is related to vegetation coverage. NDWI (Nornalized Difference Waterter Index) is a normalized ratio index based on the green band and the near infrared band, which is better for extracting water information from images.

[0071] The normalized vegetation index from February to November was calculated according to the calculation formula of vegetation index NDVI and NDWI. According to the multi-phase NDVI curve, the NDVI of various objects is between 0.2-0.9, while the NDVI of wild rice is between 0.4 and 0.8. Wild rice in Huangyan District is generally planted in two seasons, and there are fields with wild rice all year round. Even when there is no wild rice planted in the field, the NDVI of the wild rice plot is not low because duckweed covers the water during land preparation and irrigation. The NDVI of the wild rice plot is between 0.4 and 0.6 from February to July, and the NDVI of wild rice gradually increases from July to November. It is generally stable between 0.7 and 0.8 during the harvest season. In October, compared with rice in the same period, the NDVI of wild rice is higher than that of rice. Figure 4 The NDVI of mid-season rice from February to November showed an overall trend of decline-increase-decline-increase-decline. The NDVI declined from February to April and dropped to about 0.2 in April. The NDVI from April to July showed a trend of first rising and then falling. The late rice from August to November also showed a trend of first rising and then falling. The rising and falling trends of forest land and fruit trees were similar.

[0072] NDWI is generally used to extract water body information from images. Figure 5 It can be seen that the NDWI of various features is generally between -0.8 and -0.2. The NDWI curve of Zizania latifolia generally shows a trend of first rising and then falling. From June to July, the NDWI of Zizania latifolia is significantly different from other features, such as Figure 5As shown in b, at this time, the NDWI of wild rice is greater than -0.4, while the other ground objects are less than -0.4. The NDWI of rice generally shows a trend of rise-fall-rise-fall.

[0073] S3 builds a recognition model to distinguish wild rice stems and estimate their area:

[0074] The main crop in the study area is wild rice, and there is also a small amount of rice planting. The method of the present invention aims to distinguish double-season wild rice and estimate its area, so all the features in the study area are not subdivided, and other features can be removed using mask files.

[0075] After repeated comparisons and tests, NDVI can effectively eliminate non-vegetation and man-made buildings; using the NDWI sensitive to water bodies and the NDVI sensitive to vegetation to set the threshold, the water body range in June can be extracted; combined with the NDVI in October, the wild rice can be distinguished. Based on the NDVI of vegetation in September being greater than 0.6, the threshold NDVI>0.6 is set to screen out green vegetation; when the double-season wild rice is irrigated and prepared in June, the NDVI of the wild rice plot is <0.6, and the NDWI>0.4 can exclude other green vegetation and extract the water area. Figure 3 This indicates that the NDVI of double-season wild rice in October is significantly different from other vegetation, and the NDVI of double-season wild rice in October is > 0.6; when this condition is met, it means that the object is wild rice, otherwise it is other objects. The specific classification flow chart is as follows Figure 6 The classification results are shown in Figure 7 As shown, Figure 7 In a, the distribution of Zizania latifolia is relatively concentrated, and is basically distributed in the relatively flat areas between valleys. The spatial distribution monitoring results are basically consistent with the on-site investigation.

[0076] Evaluation of model accuracy

[0077] In order to evaluate the accuracy of the model more realistically and objectively, the present invention uses the confusion matrix method to evaluate the position accuracy. Random verification points are created in the study area, with a total of 900 random points. Each random point is visually interpreted in combination with high-resolution images and field conditions, and compared with the remote sensing extraction results to create a confusion matrix, and its user accuracy, mapping accuracy and Kappa coefficient are calculated respectively. The results are shown in Figure 2. Figure 8 As shown. The Kappa coefficient is calculated to be 0.93, the user accuracy and mapping accuracy of wild rice are 96.00% and 94.00% respectively, and the user accuracy and mapping accuracy of other objects are 95.00% and 93.00% respectively, indicating that the misclassification rate and omission rate are low. The two accuracy verification results based on area and confusion matrix both verify that the recognition accuracy of the wild rice remote sensing extraction model in this paper is high, and remote sensing monitoring of the spatial distribution of wild rice in the region can be realized.

[0078] Based on the above ideal embodiments of this application, the relevant staff can make various changes and modifications without departing from the technical concept of this application through the above description. The technical scope of this application is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A method for extracting wild rice stem planting area based on multi-temporal remote sensing images, characterized in that: It includes the following steps: S1: Obtain the remote sensing multispectral images of the main growth periods of water bamboo in the study area, extract the images containing the red, green, blue, and near-infrared bands, and then perform atmospheric correction processing on the images; S2: Construct a decision tree classification model for classification according to the differences in spectral index characteristics, vegetation pattern characteristics, and color characteristics between water bamboo and confused crops; S3: Construct an identification model to distinguish water bamboo and estimate its area; Among them, the identification basis of the vegetation pattern characteristics described in S2 includes: (1) For the plots planted with summer water bamboo, from November of the previous year to before the harvest of summer water bamboo in March - April of the following year, the fields are covered with greenhouse plastic films; (2) After the harvest of summer water bamboo and before the planting of autumn water bamboo, it is the field irrigation stage; (3) During the harvest stage of autumn water bamboo, compared with the late rice harvested in the same period, it shows greener than the late rice in terms of color characteristics; The identification of the color characteristics described in S2 includes: Use the feature that the water bamboo plots show grayish-white in February for visual interpretation; Extract all the plots with water and covered with azolla in June from the images in June; Then use the index feature that water bamboo shows dark green in the field in October to further screen out the water bamboo planting plots, and exclude the non-water bamboo planting plots according to the images in October; The spectral index characteristics described in S2 include: Calculate the NDVI and NDWI vegetation indices on the images from February to November; Among them, NDVI is the normalized vegetation index, which is the ratio of the difference between the near-infrared band and the visible light red band values to the sum of these two band values. The calculation formula is: NDVI = (NIR - R) / (NIR + R); NDWI is the normalized ratio index based on the green band and the near-infrared band. The calculation formula is: NDWI = (G - NIR) / (G + NIR); Where NIR is the value of the infrared band, R is the value of the visible light red band, and G is the value of the green band.

2. The method according to claim 1, characterized in that The remote sensing multispectral images of the main growth periods of water bamboo in the study area described in S1 are obtained through the image data provided by Sentinel - 2 satellite.

3. The method according to claim 2, characterized in that The central wavelength of the blue light band is 0.49μm, and the resolution is better than 10m; the central wavelength of the green light band is 0.56μm, and the resolution is better than 10m; the central wavelength of the red light band is 0.665μm, and the resolution is better than 10m; the central wavelength of the near-infrared band is 0.842μm, and the resolution is better than 10m.

4. The method according to claim 1, characterized in that: The construction of the identification model described in S3 includes: S31: Use NDVI to exclude non-vegetation and man-made buildings; S32: Use NDWI and NDVI to set thresholds to extract the water body range in June; S33: Then combine the NDVI in October to distinguish water bamboo.

5. The method according to claim 1, characterized in that The construction of the identification model described in S3 includes: S31: Extract the image in September, set the threshold NDVI > 0.

6. When the condition is false, the pixel is marked as non-vegetation; when the condition is true, the pixel is marked as vegetation, and then enter the judgment of S32; S32: Extract the image in June, set the threshold 0.4 < NDVI < 0.

6. When the condition is false, the pixel is marked as other vegetation; when the condition is true, then enter the judgment of S33; S33: Extract the image map of October, set the threshold NDVI>0.6, when the condition is false, the pixel is marked as other vegetation; when the condition is true, the pixel is marked as wild rice.

6. The method according to any one of claims 1 to 5, characterized in that: It also includes the use of confusion matrix method to evaluate position accuracy.

7. The method according to claim 6, characterized in that: The method of using confusion matrix to evaluate position accuracy includes: creating 900 random verification points in the study area, visually interpreting each random point in combination with high-resolution images and field conditions, and comparing them with remote sensing extraction results to make a confusion matrix, and calculating its user accuracy, mapping accuracy and Kappa coefficient respectively.

Citation Information

Patent Citations

  • Apple identification method based on time sequence high-resolution remote sensing data

    CN111079846A

  • Soybean and corn planting area identification method in Jianghuai region based on Sentinel-2 image

    CN111462223A

  • Overwintering rape and wheat remote sensing identification method and device

    CN111860149A