CN111444783A

CN111444783A - Crop planting land parcel identification method and device based on pixel statistics

Info

Publication number: CN111444783A
Application number: CN202010164471.1A
Authority: CN
Inventors: 罗明; 陆洲; 吴学明; 梁爽; 徐飞飞
Original assignee: China United Property Insurance Co ltd; Zhongke Hexin Remote Sensing Technology Suzhou Co ltd; Institute of Geographic Sciences and Natural Resources of CAS
Current assignee: China United Property Insurance Co ltd; Zhongke Hexin Remote Sensing Technology Suzhou Co ltd; Institute of Geographic Sciences and Natural Resources of CAS
Priority date: 2020-03-11
Filing date: 2020-03-11
Publication date: 2020-07-24
Anticipated expiration: 2040-03-11
Also published as: CN111444783B

Abstract

The application relates to a crop planting plot identification method based on pixel statistics, image identification statistics is carried out on target crops based on the identification of plots, the accuracy of plot crop identification is guaranteed, and the reliability of remote sensing images on crop detection is improved by screening areas, subdividing non-single planted plots and verifying plots marked crops through pixel proportion.

Description

Crop planting land parcel identification method and device based on pixel statistics

Technical Field

The application belongs to the technical field of remote sensing image processing, and particularly relates to a crop planting land parcel identification method and device based on pixel statistics.

Background

With the rapid development of remote sensing technology, continuous surface sampling from local, regional to global scope can be provided, remote sensing data with a spatial resolution of 0.5 meter to tens of kilometers can be provided, and remote sensing observation of the ground from multiple spatial scales can be realized. Scale variation and sensitivity play an increasingly important role in analysis using telemetric data. In recent years, multi-scale remote sensing data is widely used for regional or global scale earth coverage mapping, and people pay more and more attention to the research on the scale effect of remote sensing data classification precision. With the improvement of spatial resolution, remote sensing is often higher in ground target identification and area estimation accuracy.

The remote sensing data is utilized to identify different crops and estimate the planting area of the crops, so that the crop cultivation can be conveniently monitored. The land parcel is the basic unit of agricultural planting management, and in the crop remote sensing identification process, how to accurately identify the land parcel planted by the crop, namely, how to identify the boundary of the area planted by a certain crop is the basis for accurately counting the planting area.

Disclosure of Invention

The technical problem to be solved by the invention is as follows: in order to solve the defects in the prior art, the crop planting land parcel identification method and device based on pixel statistics are high in identification precision.

The technical scheme adopted by the invention for solving the technical problems is as follows:

a crop planting land parcel identification method based on pixel statistics comprises the following steps:

s1: collecting remote sensing images of a plurality of time phases in a crop cultivation period of a certain area;

s2: identifying cultivated land plot information on the remote sensing image, dividing plots according to ridge roads of cultivated land, and counting the area of each plot;

s3: comparing the area of each land block with a first threshold value, selecting the land blocks which are greater than or equal to the first threshold value, and ignoring the land blocks which are smaller than the first threshold value;

s4: calculating and identifying whether the marked land is a single-planted land or a non-single-planted land by using the standard deviation of pixel statistics in the land when the crop key period image is used for the land which is larger than or equal to the first threshold;

s5: dividing the non-single-planted land parcel identified in the step of S4 into a plurality of sub-divided land parcels according to the types of crops, so that each land parcel is marked as one type of crop;

s6: analyzing the remote sensing images of a plurality of time phases in the crop cultivation period in the step S1, identifying pixels of target crops planted in each land, and labeling the land with the target crops;

s7: counting the number of pixels in each plot where the target crop is planted, calculating the pixel ratio in each plot, and judging whether the crop marking of the plots in the step S6 is correct or not;

s8: statistical analysis was performed on the correctly labeled plots in step S7.

Preferably, in the method for identifying a crop planting plot based on pixel statistics of the present invention, in step S4, the method for identifying whether a plot is a single-planted plot or a non-single-planted plot by calculating the standard deviation of the pixel statistics in the plot when used as an image of the key period of an object is as follows:

extracting pixels at the boundary of the plot, taking the intersection point of diagonal lines of the pixels at the boundary as the central point of the pixels, and judging whether the central point is positioned in the plot; if the center point of the pixel is outside the land parcel, the pixel is not used as the pixel contained in the land parcel; if the center point of the pixel is in the land parcel, the pixel is taken as the pixel contained in the land parcel, and the pixel of which the center point is positioned at the boundary in the land parcel and the pixel in the boundary are taken as the pixels contained in the corresponding land parcel;

calculating the NDVI value of each pixel contained in the land parcel,

in the formula, B3 and B4 are respectively a red waveband and a near red waveband;

calculating the standard deviation of the NDVI value of each pixel in the land mass:

in the formula, x_iRepresenting the NDVI values of the picture elements comprised by the parcel,

the average value of the NDVI of the pixels contained in the land parcel is represented;

preferably, in the method for identifying the crop planting plots based on the pixel statistics, in the step S7, the specific judgment step is that the pixel proportion in each plot is calculated to be C/N × 100 percent,

in the formula, C represents the number of pixels in a single plot where a target crop is planted, and N represents the number of all pixels contained in the single plot; when the pixel occupation ratio is less than or equal to a third threshold value, the crop of the plot in the S6 is considered to be marked as an error; and when the pixel occupancy is larger than a third threshold value, the crop mark of the plot in the S6 is considered to be correct.

Preferably, in the method for identifying crop planting plots based on pixel statistics, in step S5, a marked training sample is newly selected for the non-single planting plot identified in step S4 as a new mask map layer, the marked plot is subdivided by using a neural network, the non-single planting plot is divided into a plurality of subdivided plots according to the types of crops, and inspection and manual correction are performed so that each plot is only marked as one crop.

Preferably, according to the crop planting land parcel identification method based on pixel statistics, the target crop is cotton, and the remote sensing images of late 4 month, late 5 month, middle 6 month, late 7 month, middle 8 month, late 9 month and late 9 month are respectively obtained in the step of S1;

and S6, when the picture elements of the target crops planted in each land parcel are identified, the following steps are adopted:

when any one of the following two conditions is met, judging that the field planting object is cotton;

the first condition is as follows:

NDVI value of the plot in the remote sensing image in late 4 th ten days is less than 0.19 and more than 0.04;

and is

The NDVI value of the plot in the remote sensing image in late 5-month ten-day is less than 0.18 and more than 0.06;

and is

NDVI value of the plot in the remote sensing image in 6 middle of the month is less than 0.44 and greater than 0.29;

and is

The NDVI value of the plot in the remote sensing image in late 7 th month is less than 0.51 and more than 0.36;

and is

NDVI value of the plot in the remote sensing image in 8 middle of the month is less than 0.69 and greater than 0.44;

and is

The NDVI value of the plot in the remote sensing image in late 9 th month is less than 0.66 and greater than 0.42;

and a second condition:

and is

The NDVI value of the plot in the remote sensing image in last 9 th ten days is less than 0.69 and greater than 0.44;

and is

The NDVI value of the plot in the remote sensing image in late 9 th month is less than 0.47 and greater than 0.15.

The invention also provides a device for identifying the crop planting land based on the pixel statistics, which comprises the following components:

the image acquisition module: the remote sensing image acquisition system is used for acquiring remote sensing images of a plurality of time phases in a crop cultivation period of a certain area;

an information identification module: the remote sensing image is used for identifying cultivated land plot information on the remote sensing image, dividing plots according to ridge roads of cultivated land, and counting the area of each plot;

area screening module: the area of each land is compared with a first threshold value, land which is larger than or equal to the first threshold value is selected, and land which is smaller than the first threshold value is ignored;

planting amount screening module: the standard deviation calculation module is used for calculating and identifying whether the marked land is a single-planted land or a non-single-planted land by using the standard deviation of pixel statistics in the land when the crop key period image is used for aiming at the land which is larger than or equal to the first threshold;

a land block subdivision module: the non-single planting plots identified by the planting quantity screening module are divided into a plurality of sub-divided plots according to the types of crops, so that each plot is only marked as one type of crop;

a crop labeling module: the remote sensing image acquisition module is used for analyzing the remote sensing images of a plurality of time phases in the crop cultivation period in the image acquisition module, identifying pixels of target crops planted in each land, and marking the land with the target crops;

planting amount screening module: the system is used for counting the number of pixels of a target crop planted in a single plot of a single plot, calculating the pixel ratio of each plot, and judging whether the crop marking of the plot in the crop marking module is correct or not;

a statistical analysis module: and carrying out statistical analysis on the land parcels marked correctly in the mark verification module.

Preferably, in the method for identifying a crop planting land parcel based on pixel statistics, the method for calculating and identifying whether the marked land parcel is a single-planted land parcel or a non-single-planted land parcel by using the standard deviation of the pixel statistics in the land parcel when the crop key period image is in the planting amount screening module comprises the following steps:

extracting a pixel at the boundary of the land parcel, taking the intersection point of diagonal lines of the pixel at the boundary as the central point of the pixel, judging whether the central point is positioned in the land parcel, if the central point of the pixel is outside the land parcel, the pixel is not taken as the pixel contained in the land parcel, if the central point of the pixel is in the land parcel, the pixel is taken as the pixel contained in the land parcel, and the pixel at the boundary of which the central point of the pixel is positioned in the land parcel and the pixel in the boundary are taken as the pixels contained in the corresponding land parcel;

calculating the NDVI value of each pixel contained in the land parcel,

and counting the standard deviations of all the plots, and determining a second threshold value, wherein the plots with the standard deviations larger than the second threshold value are non-single-planting plots, and the plots with the standard deviations smaller than or equal to the second threshold value are single-planting plots.

Preferably, in the method for identifying the crop planting plots based on the pixel statistics, the planting amount screening module specifically comprises the steps of calculating the pixel proportion of each plot as C/N × 100%,

in the formula, C represents the number of pixels in a single plot where a target crop is planted, and N represents the number of all pixels contained in the single plot; when the pixel occupation ratio is less than or equal to a third threshold value, the crops of the land blocks in the crop marking module are marked as errors; and when the pixel element ratio is larger than a third threshold value, the crop marking of the land parcel in the crop marking module is considered to be correct.

Preferably, in the method for identifying crop planting plots based on pixel statistics, the plots subdivision module takes the non-single planting plots identified in the planting amount screening module as new mask patterns, newly selects the marked training samples, subdivides the marked plots by using the neural network, divides the non-single planting plots into a plurality of subdivided plots according to the types of crops, and performs inspection and manual correction to make each plot only be marked as one type of crop.

Preferably, according to the crop planting land parcel identification method based on pixel statistics, a target crop is cotton, and remote sensing images of late 4 months, late 5 months, middle 6 months, late 7 months, middle 8 months, early 9 months and late 9 months are respectively acquired from the image acquisition module;

when the crop marking module identifies the pixels of the target crops planted in each land parcel, the following steps are adopted:

the first condition is as follows:

and is

and a second condition:

and is

The invention has the beneficial effects that:

the crop planting plot identification method based on the pixel statistics carries out image identification statistics on target crops based on the identification of plots, verifies the crops marked on the plots through screening the area, subdividing the non-single planted plots and comparing the plots by the pixels, ensures the accuracy of the identification of the plots, and improves the reliability of the remote sensing image on the detection of the crops.

Drawings

The technical solution of the present application is further explained below with reference to the drawings and the embodiments.

FIG. 1 is a flowchart of a method for identifying a crop planting plot based on pixel statistics according to an embodiment of the present application;

FIG. 2 is a diagram of elevation and location of an experimental area in an effect example;

FIG. 3 is a flow chart of cotton identification in an effect embodiment;

FIG. 4 is a distribution diagram of sampling points of pixel identification of a cotton planting field in an effect embodiment;

FIG. 5 is a graph of growth of various plants in the same period in an effect example;

FIG. 6 is a histogram of the area of statistical plots in the effect example;

FIG. 7 is a plot correction technique route in an effect embodiment;

FIG. 8 is a diagram of the effect of the parcel correction process in the effect embodiment;

FIG. 9 is a diagram showing the result of intersection analysis of pixels and plots in an effect example;

FIG. 10 is a flowchart of marked parcel refinement correction in an effect embodiment;

FIG. 11 is a flow diagram of marked parcel refinement overlay in an effect embodiment;

FIG. 12 is a schematic diagram of analysis of superposition of land parcel and pixel in effect embodiment;

FIG. 13a shows the extraction result of cotton pixels nested in the effect example;

FIG. 13b shows the result of extraction of cotton pixels in the effect example;

FIG. 13c shows the result of extracting cotton pixels after superposition of nested cropping and single cropping in the effect embodiment;

FIG. 14 is a diagram showing the results of correction of plots in the effect example;

FIG. 15 is a diagram of the result of the superposition of the image plots in the effect embodiment.

Detailed Description

It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict.

The technical solutions of the present application will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.

Example 1

The embodiment provides a method for identifying crop planting land blocks based on pixel statistics, which comprises the following steps as shown in fig. 1:

the identification method can utilize a convolution algorithm of machine learning to identify the farmland plot information, adopts a multi-scale segmentation method to identify ridge roads of the farmland, and combines a high-resolution Google earth image to carry out manual quality inspection and correction on an identification result, so as to finally divide the plot;

the specific method for calculating and identifying whether the marked land is a single-planted land or a non-single-planted land by using the standard deviation of pixel statistics in the land when the crop key period image is taken comprises the following steps:

extracting a pixel at the boundary of the plot, taking the intersection point of diagonal lines of the image capturing element as the central point of the pixel, judging whether the central point is positioned in the plot, if the central point of the pixel is outside the plot, not taking the pixel as the pixel contained in the plot, if the central point of the pixel is inside the plot, taking the pixel as the pixel contained in the plot, and taking the pixel at the boundary with the central point of the pixel positioned in the plot and the pixel in the boundary as the pixel C contained in the corresponding plot;

calculating the NDVI value of each pixel contained in the land parcel,

Determination of the second threshold: and acquiring the standard deviation of the NDVI values of all the plots, carrying out natural breakpoint classification on the standard deviation according to the value, dividing the plots into four stages from small to large according to the value range, reserving the plots with the smaller standard deviation of the third stage, and setting the second threshold as the standard deviation of the boundary of the third stage and the fourth stage.

S5: dividing the non-single-planted plots identified in step S4 into a plurality of sub-divided plots according to the type of crop, such that each plot (a plot is understood to be a single-planted plot as well as a sub-divided plot) is labeled as a crop only;

the specific identification method comprises the steps of taking the non-single planted land blocks identified in the step S4 as a new mask image layer, reselecting a mark training sample, subdividing the marked land blocks by using a neural network again, dividing the non-single planted land blocks into a plurality of small land blocks according to planting conditions, and carrying out inspection and manual correction to ensure that each land block is only marked as a crop;

s6: analyzing the remote sensing images of a plurality of time phases in the crop cultivation period in S1, identifying pixels of target crops planted in each land plot (the land plot is also understood as a single planted land plot and a subdivided land plot), and labeling the land plot with the target crops;

s7: counting the number of pixels of the target crops planted in a single plot of a single plot, calculating the pixel ratio of each plot, and judging whether the crop mark of the plot is correct in S5;

the specific judging steps are as follows: calculating the pixel occupation ratio in each land block:

Determination of the third threshold: and acquiring pixel occupation ratios of all the plots, grading the pixel occupation ratios according to values, dividing the plots into four stages from small to large according to value ranges, reserving the plots with the larger pixel occupation ratios of the three stages, and setting a third threshold as the pixel occupation ratio of the boundary of the first stage and the second stage.

S8: statistical analysis was performed on the land parcel marked correctly in S7. The content of the statistical analysis may be the planting area of the target crop, the percentage of the target crop to the total cultivation area, and the like.

The target crop is cotton, and remote sensing images of late 4 month, late 5 month, middle 6 month, late 7 month, middle 8 month, early 9 month and late 9 month are respectively obtained in the step of S1;

in the step S5, the following steps are adopted when the planted crops in the land parcel are identified:

when any one of the following two conditions is met, the field cultivated object is judged to be cotton, and the two conditions can completely identify the cotton planting;

the first condition is as follows: (to identify the plot of cotton for a single crop)

and is

and a second condition: (to identify the plot of relay intercropping cotton)

and is

Example 2

The embodiment provides a crop planting plot identification device based on pixel statistics, which corresponds to the method of the embodiment and comprises the following steps:

Preferably, in the method for identifying a crop planting land parcel based on pixel statistics of the present invention, the method for calculating and identifying whether the marked land parcel is a single-planted land parcel or a non-single-planted land parcel by using the standard deviation of the pixel statistics in the land parcel when the crop key period image is displayed in the planting amount screening module comprises:

extracting a pixel at the boundary of the land parcel, taking the intersection point of diagonal lines of the pixel at the boundary as a central point of the pixel, judging whether the central point is positioned in the land parcel, if the central point of the pixel is outside the land parcel, the pixel is not taken as the pixel contained in the land parcel, if the central point of the pixel is in the land parcel, the pixel is taken as the pixel contained in the land parcel, and the pixel at the boundary of which the central point of the pixel is positioned in the land parcel and the pixel in the boundary are taken as the pixels contained in the corresponding land parcel;

calculating the NDVI value of each pixel contained in the land parcel,

wherein B3 and B4 are red respectivelyBand, near red band;

and counting the standard deviations of all the plots, and grading the standard deviations according to natural breakpoints, wherein the plots with the standard deviations larger than a second threshold value are non-single-planted plots, and the plots with the standard deviations smaller than or equal to the second threshold value are single-planted plots.

Preferably, the planting amount screening module specifically judges that the ratio of pixels in each plot is calculated to be C/N × 100%,

in the formula, C represents the number of pixels in a single plot where a target crop is planted, and N represents the number of all pixels contained in the single plot; when the pixel occupation ratio is less than or equal to a third threshold value, the crops of the land blocks in the land block subdivision module are marked as errors; and when the pixel proportion is larger than a third threshold value, the crop mark of the land parcel in the land parcel subdivision module is correct.

Determination of the third threshold: and acquiring pixel occupation ratios of all the plots, grading the pixel occupation ratios according to values, dividing the plots into four levels according to value ranges, reserving the plots with the last three levels of pixel occupation ratios, and setting a third threshold as the pixel occupation ratio of a boundary between the first level and the second level.

Preferably, in the plot subdividing module, the non-single-planted plots identified in the planting amount screening module are used as new mask patterns, the marked training samples are newly selected, the marked plots are subdivided by using a neural network, the non-single-planted plots are divided into a plurality of subdivided plots according to the types of crops, and the subdivided plots are checked and manually corrected, so that each plot is only marked as one type of crop.

Preferably, the target crop is cotton, and the image acquisition module acquires remote sensing images of late 4 months, late 5 months, middle 6 months, late 7 months, middle 8 months, early 9 months and late 9 months respectively;

the first condition is as follows:

and is

and a second condition:

and is

Effects of the embodiment

The leaf city county is used as an experimental area, is located in the southwest border of the Uygur autonomous region in Xinjiang, belongs to the karshi area, is located in the Kunlun mountain and Kunlun mountain, is connected with the open plain in the north, is connected with the Takrama desert in the south, is located at the upstream of the Henan of the leaf Notopterygium and is low in the south, the north and the south of the terrain. Belongs to a continental drought climate zone in a warm temperature zone, the average annual air temperature in the plain in the north is 11.3 ℃, the average annual precipitation is 54mm, and the average annual frost-free period is 228 days. The leaf city county plants cotton, wheat, corn, mung bean, melons, fruits, vegetables and other crops throughout the year, and produces large-seed sweet pomegranate, thin-skinned walnuts, black-leaf apricots, yellow-pulp peaches, chessboard pears and the like.

Agronomic data

The growing period of the cotton in the leaf city county is from sowing and transplanting in 4 months to picking in 11 months, and the growing period is as follows:

satellite image data

Remote sensing image data adopts 10m resolution data of European Space Administration (ESA) Sentinel 2 (Sentinel-2). The 10m resolution wave band comprises 4 wave bands of blue (B2: 458-523 mm), green (B3: 543-578 mm), red (B4: 650-680 mm) and near infrared (B8: 785-900 mm). The Sentinel-2A satellite was launched in 2015, the Sentinel-2B satellite was launched in 2017, and the two-satellite network was revisited for 5 days. And acquiring images in four blocks according to the county area range of the leaf county, the planting area and the width of the sentinel image. 28 scenes of images covering a research area are obtained from an European Bureau distribution system (https:// scihub. copernius. eu/dhus/#/home), the time is 4 months from 2018 to 9 months from 2018, the research area is large and is divided into 4 areas, which is shown in Table 2.

TABLE 2 image time phase corresponding to different block regions

Time phase 1, time phase 2, time phase 3, time phase 4, time phase 5, time phase 6, and time phase 7 respectively represent late 4, late 5, middle 6, late 7, middle 8, early 9, and late 9.

Data of land parcel

In order to ensure the integrity and classification precision of the extracted land parcel information, the farmland land parcel information is identified by using a convolution algorithm of machine learning based on a GF-2 0.8m resolution image, a multi-scale segmentation method is adopted to identify ridge roads of the farmland, and the identification result is manually inspected and corrected by combining a high resolution Google earth image. 115869 symbiotic land blocks, 1068840.7 mu area and 9.2 mu average land block size.

According to the analysis of the various point data, thresholds of all periods are limited, a multi-time-phase decision-making judgment model is constructed to extract cotton planting pixels, and single cropping cotton and relay cropping cotton are distinguished and identified because the planting fields of the single cropping cotton (the whole field is only cotton) and the relay cropping cotton (the field in which the cotton is planted and the fruit trees are interplanted) are obviously different in spectrum.

NDVI0423 represents the NDVI value of the corresponding land parcel of the remote sensing image of 23 days in 4 months.

Correction of land mass

Currently, the cultivated land parcel is identified and extracted only according to one scene image, and the growth condition of the target land feature is not considered. When the plots are automatically produced, due to the influence of current-period images, finely-divided plots (undersized plots) or large plots (oversized plots) with unobvious ridges can appear, and a plot correction method based on pixel statistics is provided to further improve the accuracy of plot identification.

For the produced cultivated land plots, deleting the undersized plots by using the area of the plots, calculating and identifying the non-single-planting plots by using the variance of pixel statistics in the image plots used as the object key period, and finally subdividing the non-single-planting plots again.

According to the planting condition and field investigation of a research area, the cotton plots exceed 1 mu, so the threshold value of the set area is 1 mu, and the fine ground plots with the area less than 1 mu are directly deleted. An oversized plot is one in which a plurality of plots are put together without distinction; for the image of the vigorous growth period, a crop is planted in a plot, the spectral values of all the pixels contained in the plot are concentrated, namely the standard deviation of all the pixel values is small; if a plurality of crops are planted in a land, the spectrum values of all the pixels contained in the land are scattered, namely the standard deviation of all the pixel values is large.

(1) NDVI feature image conversion

By using remote sensing images of various regions in vigorous growth periods, plots for planting different crops can be effectively distinguished, and a cross-border image of 5, 23 and 2018 is selected in the research. NDVI (normalized difference vegetation index) can reflect the coverage of vegetation, and the NDVI values of different crops in the same period are obviously different, so that the land parcel is analyzed by utilizing the statistics of the NDVI values.

In the formula, B3 and B4 are respectively a red waveband and a near red waveband.

(2) Plot pixelation

Due to the fact that the remote sensing image used for producing the land parcel and the image resolution used for analyzing the image elements are different, and due to the existence of the mixed image elements, the phenomenon that the image elements are cut by the land parcel can occur at the boundary of the land parcel. When pixel analysis is performed by using the transit image of 5, 23 and 2018, the pixels contained in the plot need to be determined first by overlaying the produced plots. The pixels in the remote sensing image are regular squares with the side length of 10 meters. Taking the intersection point of the diagonal lines of the image capturing elements as the center point of the image elements, and performing superposition analysis to judge whether the center point is positioned in the ground block; if the center point of the pixel is outside the land, the pixel is not used as the pixel contained in the land; if the pixel center point is in the land, the pixel is used as the pixel contained in the land, as shown in fig. 9 b.

(3) Pixel NDVI value standard deviation statistics in land mass

And (4) counting the NDVI values of the image elements contained in the land parcel, and calculating the standard deviation of the NDVI values.

In the formula, x_iRepresenting the NDVI value of the picture elements contained in the parcel,

and taking the plots as basic units, counting the standard deviations of all the plots, grading the standard deviations according to natural break points, and screening out the plots with larger standard deviations, namely the plots for planting various crops.

(4) Fine correction of land parcel

And taking the land blocks marked with the various crops as new mask image layers, reselecting the marked training samples, subdividing the marked land blocks by utilizing the neural network again, dividing one land block planted with the various crops into a plurality of land blocks according to the planting condition, and carrying out inspection and manual correction.

Superposition analysis of pixels and plots

In agriculture, crop planting takes a land block as a unit, and in order to obtain the accurate planting area of crops, pixel extraction results need to be superposed and associated to the corrected land block. And determining the crop planting land block by analyzing the number of the pixels contained in the land block and the number of the pixels of which the pixel extraction result falls in the land block. And (3) analyzing and counting the land areas planted by the target crops by using the land areas as basic units and utilizing the pixel occupation ratio.

Wherein C represents the number of pixels of the pixel extraction result in a single plot, and N represents the number of all pixels contained in the single plot (pixels at all boundaries plus pixels in all boundaries);

superposing the multi-temporal decision extraction result to the corrected land parcel, counting the number of pixels of the extraction result falling into the land parcel and all the pixels of the land parcel, calculating the pixel occupation ratio of the land parcel, dividing the land parcel into four stages according to a natural breakpoint method, and setting a threshold value of the pixel occupation ratio according to the actual planting condition; when the pixel occupation ratio is less than or equal to the threshold value, the pixels extracted by crop planting cover less land blocks, the land blocks are not target crop planting land blocks and are misclassified of pixels, and the land blocks are directly deleted; when the pixel proportion is larger than the threshold value, more pixels extracted by crop planting cover the land parcel, which indicates that the land parcel is a target crop planting land parcel, and the attribute is directly reserved and marked.

Evaluation of accuracy

And carrying out precision evaluation by using the area and the confusion matrix, wherein the evaluation indexes comprise: area accuracy, overall classification accuracy, and Kappa coefficient.

In the formula, S_rFor remote-sensing of monitored area, S_tReporting the statistical area;

where r is the number of rows in the error matrix, x_iiIs the value, x, in i rows and i columns (main diagonal)_i+And x_+iRespectively, the sum of the ith row and the sum of the ith column, wherein N is the total number of sampling points;

results and precision evaluation

Result of pixel extraction

Evaluation of accuracy

Surface cotton plot confusion matrix evaluation

Kappa is 0.908, the remote sensing monitoring area is 25.46 ten thousand mu (16973.33 hectare), and the statistical area is 26.5 ten thousand mu (17666.67 hectare).

Table method accuracy comparison

As can be seen from the above table, the accuracy is improved to a greater extent.

In light of the foregoing description of the preferred embodiments according to the present application, it is to be understood that various changes and modifications may be made without departing from the spirit and scope of the invention. The technical scope of the present application is not limited to the contents of the specification, and must be determined according to the scope of the claims.

As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

Claims

1. A crop planting land parcel identification method based on pixel statistics is characterized by comprising the following steps:

s3: comparing the area of each land block with a first threshold value, selecting land blocks which are greater than or equal to the first threshold value, and ignoring land blocks which are smaller than the first threshold value;

2. A method for identifying a crop planting land parcel based on pixel statistics as claimed in claim 1, wherein in step S4, the method for identifying whether the land parcel is a single-planted land parcel or a non-single-planted land parcel by using the standard deviation of the pixel statistics in the land parcel when the crop key period image is taken comprises:

calculating the NDVI value of each pixel contained in the land parcel,

indicating the land parcel containsAverage value of pixel NDVI size;

3. The method for identifying a crop planting land mass based on pixel element statistics as claimed in claim 1, wherein in the step of S7, the specific determination step is to calculate the pixel element ratio in each land mass as C/N × 100%,

4. A pixel statistics based crop planting land parcel identification method according to claim 1, characterized in that in the step S5, for the non-single planting land parcel identified in the step S4 as a new mask image layer, a new mark training sample is selected, the marked land parcel is subdivided by using a neural network, the non-single planting land parcel is divided into a plurality of subdivided land parcels according to the crop species, and the inspection and manual correction are carried out, so that each land parcel is marked as only one crop.

5. The method for identifying crop planting land parcels based on pixel statistics as claimed in claim 1, wherein the target crop is cotton, and in step S1, remote sensing images of late 4, late 5, middle 6, late 7, middle 8, early 9, and late 9 are respectively obtained;

the first condition is as follows:

and is

and a second condition:

and is

6. The utility model provides a crop planting plot recognition device based on pixel statistics which characterized in that includes:

area screening module: the area of each land is compared with a first threshold value, land blocks larger than or equal to the first threshold value are selected, and land blocks smaller than the first threshold value are ignored;

planting amount screening module: the standard deviation calculation method is used for calculating and identifying whether the marked land is a single-planted land or a non-single-planted land by using the standard deviation of pixel statistics in the land when the crop key period image is used for aiming at the land which is larger than or equal to the first threshold;

7. The method for identifying crop planting land parcels based on pixel statistics of claim 6, wherein the method for identifying whether the land parcels are single-planted or non-single-planted by calculating the standard deviation of the pixel statistics in the land parcels when the crop key period images are used in the planting amount screening module comprises:

calculating the NDVI value of each pixel contained in the land parcel,

8. The method for identifying crop planting plots based on pixel statistics as claimed in claim 6, wherein the planting amount screening module specifically judges the steps of calculating the pixel proportion of each plot as C/N × 100%,

in the formula, C represents the number of pixels in a single plot where a target crop is planted, and N represents the number of all pixels contained in the single plot; when the pixel occupation ratio is less than or equal to a third threshold value, the crops of the land parcel in the crop marking module are marked as errors; and when the pixel element ratio is larger than a third threshold value, the crop marking of the land parcel in the crop marking module is considered to be correct.

9. A pixel statistics based crop planting land parcel identification method according to claim 6, characterized in that in the land parcel segmentation module, the non-single planting land parcel identified in the planting amount screening module is used as a new mask image layer, a mark training sample is newly selected, the marked land parcel is segmented by using a neural network, the non-single planting land parcel is divided into a plurality of segmented land parcels according to the crop types, and the inspection and manual correction are carried out, so that each land parcel is only marked as a crop.

10. The method for identifying the crop planting land parcel based on the pixel statistics as claimed in claim 6, wherein the target crop is cotton, and the image acquisition module acquires remote sensing images of late 4 month, late 5 month, middle 6 month, late 7 month, middle 8 month, late 9 month and late 9 month respectively;

the first condition is as follows:

and is

and a second condition:

and is