CN116205678A - Agricultural land physical quantity investigation and value quantity estimation method based on remote sensing automatic interpretation - Google Patents

Abstract

The invention discloses an agricultural land real object investigation and value estimation method based on remote sensing automatic interpretation, which comprises the steps of obtaining an agricultural land satellite remote sensing image of a research area, and preprocessing the image; automatically identifying agricultural land physical quantity information in the preprocessed image based on a machine learning method; collecting price signal data; and establishing a price system, and finally calculating the spot value of the agricultural map. Compared with the traditional statistical method, the method for acquiring the agricultural physical quantity based on the remote sensing image data acquisition area has the advantages that the agricultural spatial position is accurate, the boundary is clear, the physical quantity is accurate, and the cheapness of acquiring the remote sensing data can be realized, so that the acquisition of multi-time sequence agricultural physical quantity data in a shorter time can be realized.

Description

Agricultural land physical quantity investigation and value quantity estimation method based on remote sensing automatic interpretation

Technical Field

The invention relates to a land asset checking and evaluating technology, in particular to a method for checking real object quantity and calculating economic value quantity of agricultural land assets.

Background

1. In recent years, the earth observation technology is developed rapidly, the defects of the traditional investigation means are overcome to a great extent by the appearance and application of satellite remote sensing images, the speed and precision of acquiring the earth surface target object are improved, and the rapid acquisition and effective coverage of large-scale high-resolution remote sensing images can be realized. The existing regional agricultural land area, quantity and other physical quantity attribute information acquisition is based on statistical data of professional management departments, the difficulty of data acquisition exists in a barrier, and the difficulty of agricultural land data acquisition is reduced by acquiring agricultural land physical quantity information based on remote sensing images, and dynamic adjustment of agricultural land classification can be performed according to research requirements.

2. The agricultural land reference land price results and other evaluation results which are dominant by the existing natural resource departments do not correspond well in the classification system, the agricultural land reference land price results reference land price classification system does not comprise other agricultural land types except cultivated lands such as garden lands and cultivation pits, the classification system is unhooked from the cultivated lands and other evaluation results, the agricultural land price based on the reference land price evaluation is difficult to accurately reflect the total amount of the agricultural land economic value in a region to a certain extent, and the evaluation result is rarely reflected in space.

Disclosure of Invention

The invention aims to: the main problems to be solved by the invention are as follows: firstly, how to timely survey and timely acquire attribute information such as agricultural land distribution, type and area based on a remote sensing big data automatic interpretation method; and secondly, a set of agricultural land evaluation price system matched with the coverage type and the real object type is specified, so that the agricultural land value is estimated more accurately.

The technical scheme is as follows: the invention discloses an agricultural land real object investigation and value quantity estimation method based on remote sensing automatic interpretation, which comprises the following steps:

s1, acquiring an agricultural ground satellite remote sensing image of a research area, and preprocessing the image; the invention carries out image preprocessing based on the professional remote sensing images such as ENVI and the like, and comprises geometric correction, calibration, atmospheric correction and the like;

step S2, automatically identifying agricultural physical quantity information in the preprocessed image based on a machine learning method, namely: firstly, establishing an interpretation mark based on the analysis of a sentinel-2A image and the combination of field investigation according to land utilization classification standards (see standard number GB/T21010-2007 for details), and then, obtaining an agricultural land distribution range based on interpretation of professional remote sensing image processing software such as ENVI and the like;

s3, collecting price signal data;

the price signal data comprises (1) a use right value or a benefit value estimated by using the years, (2) a quality grade evaluation result of the agricultural land in a research area, (3) according to the collected spatial distribution of different grades of the agricultural land, the agricultural land transaction sample point data existing in each grade distribution area are collected in a grading manner;

the data of the agricultural transaction sample points comprise transaction case data (preferably selecting transaction cases of nearly three years) such as agricultural contract, subcontracting and renting, and the like, the insufficient transaction sample points are supplemented by sample point input and output data (the data is an average value of nearly three years), all sample points need to represent the average level of national agricultural land prices in different grade areas, and all the agricultural land types, cultivated land quality and the like must be covered by the sample points;

s4, establishing a price system (comprising sampling point price information data acquisition, sampling point land price calculation, statistical test and determination of regional average price);

the agricultural land price is a use weight value or a income value estimated by using a year according to land class (namely, cultivated land quality and the like) on the premise of determining a unified reference time point and an established use, meanwhile, the land price of a rented transaction sample point is calculated by a income reduction method (see formulas (1) and (2) for details), and then, the average value of the land price of each land class (cultivated land quality and the like) is statistically checked and used as the county average price of each land class (cultivated land quality and the like); evaluating the price of the investigation transaction sample points by adopting a return restoration method according to the circulation lease level of the investigation transaction sample points;

s5, calculating the spot value of the agricultural map;

map spot economic value = check price x map spot land area, land area of each agricultural map spot is obtained by calculating ellipsoid area based on ArcGIS software.

Further, the details of the step S2 are,

step S2.1, extracting a gray level co-occurrence matrix of the preprocessed satellite remote sensing image: extracting texture features of each satellite remote sensing image by calculating a conditional probability density function between image gray levels, and further forming a gray level co-occurrence matrix GLCM of a 3 multiplied by 3 window; meanwhile, calculating a normalized vegetation index NDVI wave band and a normalized water body index NDWI; texture features here include energy (ASM), contrast (CON), correlation (CORRLN), and Entropy (ENT);

NDVI, NDWI and related texture bands are specifically expressed as follows:

wherein: b5 is near infrared band, B4 is red band, B5 is Vegetation Red Edge band, and B3 is green band;

wherein P is _ij Representing pixel gray values of the corresponding positions of the windows of 3 multiplied by 3; i and j respectively represent grid row and column coordinate values of pixels in a window, and N is the row and column length of the window, wherein N=3 in the window of 3×3;

s2.2, synthesizing data to be classified, which contains 16 wave band characteristics;

b2, B3, B4, B5, B6, B7, B8A, B11, NDVI, NDWI, DEM, ASM, CON, CORRLN, ENT; b2, B3, B4 respectively represent visible light wave bands, B5, B6, B7 represent Vegetation Red Edge wave bands of different wavelength ranges, B8A represent two near infrared wave bands, and B11 represents a short wave infrared wave band; synthesizing the 16 wave bands obtained by calculation into data to be classified in multiple wave bands by using a layer stacking tool of ENVI;

s2.3, determining the types of agricultural land in a research area, wherein the types of agricultural land comprise paddy fields, water-irrigated lands and dry lands in cultivated lands, orchards, tea gardens, rubber gardens and other gardens in plantation lands, pit water surfaces in water areas and water conservancy facility lands, and facility agricultural lands in other lands;

s2.4, establishing interested ROIs on the original B4, B3 and B2 pseudo-color display images according to the interpretation marks, wherein the number of the ROIs of each type of farm land is required to be not less than 20 and uniformly distributed in a regional range; the specific establishment method of the ROI comprises the steps of opening a natural color image with research area wave band combinations of B4, B3 and B2 in ENVI, clicking an image layer on the right button, selecting and establishing the ROI, selecting a region of interest ROI corresponding to the agricultural type on the image according to the image characteristics of the established agricultural interpretation marks, and requiring that the selected samples are uniform and accurate, wherein the sizes of the samples cannot differ too much;

s2.5, training a classification model by utilizing samples acquired by the ROI, wherein 80% of the samples are used as training sets, and 20% of the samples are used as verification sets; selecting a machine learning model of a random forest method for automatic classification; and performing prediction classification after training, performing accuracy verification by using a verification set according to the result of model taxonomy, and evaluating interpretation accuracy.

Further, the interpretation process is specifically as follows:

first, the establishment and selection of a region of interest ROI

Opening a natural color image of which the band combinations of the research areas are B4, B3 and B2 in ENVI, clicking an image layer on the right key, selecting and creating an ROI, and selecting a region of interest ROI corresponding to the agricultural type on the image according to the image characteristics of the established agricultural interpretation marks of different types, wherein the selected samples are required to be uniform and accurate, and the sizes of the samples cannot differ too much;

secondly, using the established ROI, taking eighty percent of the ROI as an automatic classification sample, selecting a method for supporting vector machine classification to train a model for classification, and automatically interpreting the farm land of a research area by using the trained model;

finally, post-classification processing and accuracy verification, since the automatically interpreted agricultural map extraction result is generally a preliminary result, some small-area image spots are inevitably generated in the classification result. From the practical application point of view, the small image spots must be removed or reclassified, and the interpretation preliminary results are classified and processed by using methods such as major/minor analysis, clustering (Clump), filtering (Sieve) and the like; after the post-classification treatment, calculating kappa coefficients by adopting a confusion matrix method, and carrying out precision evaluation on the interpretation results to determine the classification precision and reliability;

verifying by using the remaining 20% of the ROI selected before as a verification sample area to obtain a precision index;

wherein p is _o Is the overall classification accuracy; p is p _e The product of the number of real sample pixels of each class and the number of predicted sample pixels of each class is used for summing the calculation results of all classes and then the square ratio of the sum to the total number of pixels.

Further, the specific process of calculating the price of the rental transaction sample by the return method in step S4 is as follows:

a) Determining net land benefit

Taking the rental gold of the circulation years as an index for measuring net income of the sample points belonging to the rental property;

b) Determining land reduction rate

Setting the land reduction rates of cultivated lands, plantation lands, cultivation ponds and facility agricultural lands respectively by adopting a safe interest rate and risk adjustment value method;

c) Calculating and correcting the land price of the sample point

Calculating the land price of the lease transaction sample point by a return method, wherein the land price is shown in a formula (1);

P0＝v/r*[1-1/(1+r)^30] (1)

wherein: p (P) ₀ Sample point price for trade; v is yearly rent; r is soilEarth reduction rate;

for investigation spots of historical transactions, further carrying out date correction on spot land prices, and uniformly correcting spot land prices during transactions to reference days, wherein the reference days are shown in a formula (2);

Pt＝P0*Kt (2)

wherein: p (P) _t The price of the sample point is the point to be estimated; p (P) ₀ Sample point price for trade; kt is a date correction factor.

The beneficial effects are that: the invention acquires the agricultural physical quantity based on remote sensing image interpretation survey and establishes an agricultural price system by combining price signal data so as to carry out agricultural asset accounting, can better account the agricultural value of an accounting area and display the spatial distribution of the agricultural economic value.

Compared with the prior art, the method has the following advantages:

1. the agricultural land physical quantity based on the remote sensing image data acquisition area is accurate in spatial position, clear in boundary and accurate in physical quantity compared with the traditional statistical method of field investigation. And the cheapness of remote sensing data acquisition can meet the requirement of acquiring multi-time-series agricultural land physical quantity data in a shorter time.

2. The method is equivalent to a general agricultural price system, and the established agricultural evaluation price system is better corresponding to the physical quantity data according to price signals collected by different grades and different types of agricultural lands, so that the agricultural value accounting result is more accurate.

Drawings

FIG. 1 is a schematic diagram of a farm-oriented interpretation of a sign in an embodiment;

FIG. 2 is a schematic view of the spatial distribution of agricultural land obtained by the interpretation of the examples;

FIG. 3 is a schematic view of the spatial distribution of different types of agricultural land samples collected in the examples;

FIG. 4 is a schematic view of the spatial distribution of the agricultural land value amounts calculated by the examples;

fig. 5 is an overall flow chart of the present invention.

Detailed Description

The technical scheme of the present invention is described in detail below, but the scope of the present invention is not limited to the embodiments.

The invention discloses an agricultural land real object investigation and value quantity estimation method based on remote sensing automatic interpretation, which comprises the following steps:

s1, acquiring an agricultural ground satellite remote sensing image of a research area, and preprocessing the image;

step S2, automatically identifying agricultural physical quantity information in the preprocessed image based on a machine learning method, namely: firstly, according to land utilization classification standards, analyzing target texture features to establish interpretation marks based on sentinel-2A image analysis and field investigation, and then acquiring agricultural land distribution ranges based on interpretation of ENVI and other professional remote sensing image processing software;

s3, collecting price signal data;

the price signal data comprises (1) a use right value or a benefit value estimated by using the years, (2) a quality grade evaluation result of the agricultural land in a research area, (3) according to the collected spatial distribution of different grades of the agricultural land, the agricultural land transaction sample point data existing in each grade distribution area are collected in a grading manner;

the data of the agricultural transaction sample points comprise transaction case data (preferably selecting transaction cases of nearly three years) such as agricultural contract, subcontracting and renting, and the like, the insufficient transaction sample points are supplemented by sample point input and output data (the data is an average value of nearly three years), all sample points need to represent the average level of national agricultural land prices in different grade areas, and all the agricultural land types, cultivated land quality and the like must be covered by the sample points;

s4, establishing a price system (comprising sampling point price information data acquisition, sampling point land price calculation, statistical test and determination of regional average price);

the agricultural land price is a use weight value or a income value estimated by using a year according to land class (namely, cultivated land quality and the like) on the premise of determining a unified reference time point and an established use, meanwhile, the land price of a rented transaction sample point is calculated by a income reduction method (see formulas (1) and (2) for details), and then, the average value of the land price of each land class (cultivated land quality and the like) is statistically checked and used as the county average price of each land class (cultivated land quality and the like); evaluating the price of the investigation transaction sample points by adopting a return restoration method according to the circulation lease level of the investigation transaction sample points;

s5, calculating the spot value of the agricultural map;

plaque economic value = inventory price x plaque land area.

Example 1

In this embodiment, taicang city is taken as an example, and the technical scheme of the present invention is further described in detail, and the specific flow is as follows:

and step 1, downloading the sentinel No. 2 satellite remote sensing image data which is closest to the evaluation time node and covers Taicang city from websites such as the geospatial cloud data. As the sentinel-2A carries one multispectral imager, 13 spectral bands can be covered, and the width reaches 290 km. Spatial resolution of 10 meters and revisitation period of 10 days. From visible and near infrared to short wave infrared, with different spatial resolutions, among optical data, sentinel-2A data is the only data containing three bands in the red range, which is particularly effective for the covered ground type identification of vegetation. After downloading the image, preprocessing the image, including geometric correction, calibration and atmospheric correction, is carried out to complete the image preprocessing.

And 2, establishing a Taicang city agricultural land interpretation mark according to land utilization classification standards based on satellite remote sensing images and actual surveys, wherein the interpretation is as shown in figure 1, and acquiring an agricultural land (without forest and grass wet) distribution range. The agricultural land types of the Taicang are known according to the data and expert opinions, and specifically include cultivated lands (the secondary class is paddy field and water irrigation land), plantation land and pit water surface. And after the post-classification processing, calculating kappa coefficients by adopting a confusion matrix method, and carrying out precision evaluation on the interpretation results to determine the precision and reliability of classification. The method mainly uses the remaining 20% of the ROI selected before as verification sample areas to verify, and obtains the precision index. And evaluating whether the interpretation accuracy meets the requirement. The result of the interpretation of the spatial distribution of the Taicang agricultural land is shown in figure 2.

And 3, collecting price signal data. According to the quality and other evaluation results of the agricultural land of Taicang city, taicang covers five, six, seven and other agricultural lands, and transaction case data of the Taicang agricultural land contractual, subcontracting, renting and the like are collected in the last three years. The samples are to represent the average level of land prices for national agriculture in areas of different grades and must cover all types of agriculture, cultivated quality etc. The spatial distribution of different agricultural land samples collected in taicang city is shown in fig. 3.

And 4. Establishing a price system. The reference day estimated in this embodiment is 12 months 31 days 2020, the agricultural land use right period is 30 years, the land type (cultivated quality etc.) is calculated as the land price of the sample point, the lease transaction sample point is calculated by the return reduction method, the specific method is shown in formulas (1) and (2), the statistical test is performed, and the average value of the land price of each land type (cultivated quality etc.) is taken as the county average price of each land type (cultivated quality etc.). And evaluating the price of the investigation transaction sample points by adopting a return restoration method according to the circulation lease level of the investigation transaction sample points. The results of the Taiku city agricultural land price system are shown in Table 1:

TABLE 1

And 5, calculating the spot value of the agricultural map. Spot economic value = inventory price x spot land area according to the formula. The calculation result of the value quantity of the agricultural land in Taicang city is shown in figure 4.

The invention combines vegetation index, water body index, DEM and texture characteristic wave band, has abundant characteristic information participating in classification, and is beneficial to improving classification precision.

Claims (4)

1. The agricultural land real object quantity investigation and value quantity estimation method based on remote sensing automatic interpretation is characterized by comprising the following steps of:

s1, acquiring an agricultural ground satellite remote sensing image of a research area, and preprocessing the image;

step S2, automatically identifying agricultural physical quantity information in the preprocessed image based on a machine learning method, namely: analyzing texture characteristics of a target object based on sentinel-2A image analysis and field investigation according to land utilization classification standards to establish an interpretation mark, and then acquiring an agricultural land distribution range based on interpretation of ENVI remote sensing image processing software;

s3, collecting price signal data;

the price signal data comprises (1) a use right value or a benefit value estimated by using the years, (2) a quality grade evaluation result of the agricultural land in a research area, and (3) agricultural land transaction sample point data existing in each grade distribution area are collected in a grading manner according to the collected spatial distribution of different grades of the agricultural land;

step S4, establishing a price system

The agricultural land price is calculated according to the use right value or the income value estimated by the land use period on the premise of determining the unified reference time point and the set use, and meanwhile, the land price of the leased transaction sample point is calculated through a income reduction method, then the average value of the land price of each land sample point is used as the average price of each land county level through statistical inspection, and finally the checking price of the land sample point is estimated through the income reduction method according to the circulating lease level of the investigated transaction sample point;

s5, calculating the spot value of the agricultural map;

plaque economic value = inventory price x plaque land area.

2. The agricultural land physical quantity investigation and value quantity estimation method based on remote sensing automation interpretation according to claim 1, wherein the details of the step S2 are as follows:

step S2.1, extracting a gray level co-occurrence matrix of the preprocessed satellite remote sensing image: extracting texture features of each satellite remote sensing image by calculating a conditional probability density function between image gray levels, and further forming a gray level co-occurrence matrix GLCM of a 3 multiplied by 3 window; meanwhile, calculating a normalized vegetation index NDVI wave band and a normalized water body index NDWI; the texture features here include energy ASM, contrast CON, correlation CORRLN and entropy ENT; NDVI, NDWI and related texture bands are specifically expressed as follows:

wherein, B5 is near infrared band, B4 is red band, B5 is Vegetation Red Edge band, B3 is green band;

/>

wherein P is _ij The pixel gray values of the corresponding positions of the 3 x 3 window are represented, i and j respectively represent grid row and column coordinate values of pixels in the window, and N is the row and column length of the window, wherein N=3 in the 3 x 3 window;

s2.2, synthesizing data to be classified, which contains 16 wave band characteristics;

b2, B3, B4, B5, B6, B7, B8A, B11, NDVI, NDWI, DEM, ASM, CON, CORRLN, ENT, B2, B3, B4 respectively represent visible light bands, B5, B6, B7 represent Vegetation Red Edge bands of different wavelength ranges, B8A represent two near infrared bands, B11 represents a short wave infrared band, and the 16 bands obtained by calculation are synthesized into multi-band data to be classified by using a layer stacking tool of ENVI;

s2.3, determining the types of agricultural land in a research area, wherein the types of agricultural land comprise paddy fields in cultivated lands, water-poured lands, dry lands, orchards in plantation lands, tea gardens, rubber gardens, other gardens, pit water surfaces in water areas and water conservancy facility lands, and facility agricultural lands in other lands;

s2.4, establishing interested ROIs on the original B4, B3 and B2 pseudo-color display images according to the interpretation marks, wherein the number of the ROIs of each type of farm land is required to be not less than 20 and uniformly distributed in a regional range; the specific establishment method of the ROI comprises the steps of opening a natural color image with research area wave band combinations of B4, B3 and B2 in ENVI, clicking an image layer on the right button, selecting and establishing the ROI, selecting a region of interest ROI corresponding to the agricultural type on the image according to the image characteristics of the established agricultural interpretation marks, and requiring that the selected samples are uniform and accurate, wherein the sizes of the samples cannot differ too much;

s2.5, training a classification model by utilizing samples acquired by the ROI, wherein 80% of the samples are used as training sets, and 20% of the samples are used as verification sets; selecting a machine learning model of a random forest method for automatic classification; and performing prediction classification after training, performing accuracy verification by using a verification set according to the result of model taxonomy, and evaluating interpretation accuracy.

3. The agricultural land physical quantity investigation and value quantity estimation method based on remote sensing automation interpretation according to claim 1, wherein: the interpretation process in the step S2 is specifically as follows:

first, the creation and selection of a region of interest ROI

Opening a natural color image of which the band combinations of the research areas are B4, B3 and B2 in ENVI, clicking an image layer on the right button, selecting and creating the ROI, and selecting a region of interest ROI corresponding to the type of the agricultural land on the image according to the image characteristics of the established different types of agricultural land interpretation marks;

secondly, using the established ROI, taking eighty percent of the ROI as an automatic classification sample, selecting a method for supporting vector machine classification to train a model for classification, and automatically interpreting the farm land of a research area by using the trained model;

finally, post-classification processing and accuracy verification

Classifying and post-processing the interpretation preliminary result by using a quality/quality analysis, a clustering processing Clump and a filtering processing Sieve, calculating kappa coefficients by using a confusion matrix method after the classifying and post-processing to evaluate the accuracy of the interpretation result, and determining the classification accuracy and reliability: verifying by using the remaining 20% of the ROI selected before as a verification sample area to obtain a precision index;

wherein p is _o Is the overall classification accuracy; p is p _e The product of the number of real sample pixels of each class and the number of predicted sample pixels of each class is used for summing the calculation results of all classes and then the square ratio of the sum to the total number of pixels.

4. The agricultural land physical quantity investigation and value quantity estimation method based on remote sensing automation interpretation according to claim 1, wherein: the specific process of calculating the price of the lease transaction point by the return method in the step S4 is as follows:

a) Determining net land benefit

Taking the rental gold of the circulation years as an index for measuring net income of the sample points belonging to the rental property;

b) Determining land reduction rate

Setting the land reduction rates of cultivated lands, plantation lands, cultivation ponds and facility agricultural lands respectively by adopting a safe interest rate and risk adjustment value method;

c) Calculating and correcting the land price of the sample point

Calculating the land price of the lease transaction sample point by a return method, wherein the land price is shown in a formula (1);

P0＝v/r*[1-1/(1+r)^30] (1)

wherein: p (P) ₀ Sample point price for trade; v is yearly rent; r is the land reduction rate;

for investigation spots of historical transactions, further carrying out date correction on spot land prices, and uniformly correcting spot land prices during transactions to reference days, wherein the reference days are shown in a formula (2);

Pt＝P0*Kt (2)

wherein: p (P) _t The price of the sample point is the point to be estimated; p (P) ₀ Sample point price for trade; k (K) _t The coefficients are modified for the future date.

Images