CN102867115A - Farmland division method based on fuzzy c-means clustering - Google Patents

Abstract

The invention discloses a farmland division method based on fuzzy c-means clustering. The farmland division method comprises the following steps of: carrying out spatial interpolation on sampling point farmland soil nutrient data so as to generate grid data, and organizing grid data of different nutrients into a sample matrix; and carrying out clustering analysis on the sample matrix in a given classification range by using the fuzzy c-means clustering, calculating the clustering effect index S of all clustering results in the classification range, and taking the clustering result corresponding to the minimal S value as the farmland division result. According to the farmland division method, a method for determining optimal farmland classification number is adopted for solving the defects of FPI (Fibrosis Probability Index) and NCE (Normal Curve Equivalent) evaluation functions, so that the discrimination on soil nutrient difference in the farmland division process is improved, therefore the requirements of the farmland precise management or crop growth information monitoring on the farmland division can be better met.

Description

A kind of farmland division methods based on Fuzzy c-means Clustering

Technical field

The present invention relates to a kind of farmland division methods, relate in particular to a kind of method of utilizing soil nutrient information that the farmland is divided.

Background technology

Accurately the farming management is by monitoring crop growth and farm environment information, according to farmland cell environment and crop growing state difference, implement location, quantitative, accurately management of agricultural regularly, under the prerequisite of assurance crop yield and quality, can obtain optimal economic and ecological benefits.The growing way difference of crop mainly is subjected to the impact of Soil Nutrients in Farmland difference, and soil nutrient spatially has correlativity, therefore utilize soil nutrient information that the farmland is divided, make the similar zone, farmland of nutrient be divided into a class, can reduce the quantity of monitoring point, farmland, quick, the low-cost monitoring that realizes plant growth information is had important realistic meaning.

Fuzzy c-means Clustering is because of it artificial intelligence characteristic that possesses, and become focus in the subregion research in farmland, and be widely used, but the optimal classification number still needs other evaluation function to determine when dividing the farmland with the method.

The evaluation function that generally uses has fuzzy performance index FPI and normalization classification entropy NCE.

Wherein, $\mathrm{FPI}=1-\frac{c}{c-1}(1-\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{c}{\mathrm{\&Sigma;}}}{u}_{\mathrm{<figure-callout\; id="2"\; label="ik"\; filenames="120828171002.png"\; state="\{\{state\}\}">ik</figure-callout>}}^2/n),$ Wherein c is number of categories, and n is sample number, u _IkBe sample X _kThe degree of membership value that belongs to the i class.The value of FPI is between 0 to 1, and shared data was few when this value represented cluster near 0, and the division of class is obvious, if this value, represents then that more shared data is arranged near 1, the division of class is not obvious, so the less Clustering Effect of FPI is better. $\mathrm{NCE}=[-\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{c}{\mathrm{\&Sigma;}}}{u}_{\mathrm{ik}}{\log }_a\left(u_{\mathrm{ik}}\right)]/(1-\frac{c}{n}),$ The truth of a matter a of logarithm can be any positive integer, and NCE is worth littlely, and classifying quality is better.General FPI and NCE need unite use, and namely getting FPI and NCE, to be the corresponding number of categories of minimum value be the optimal classification number.

The above-mentioned FPI of utilization and NCE evaluation function determine that the method for the optimal classification number in Fuzzy c-means Clustering division farmland has description: Li Yan in Publication about Document, Shi Zhou, Wu Cifang, Deng. based on the accurate management zone research in the field of fuzzy cluster analysis [J]. Scientia Agricultura Sinica, 2007 (1): 114-122; Moral F J, Terron J M, Silva J R.Delineation of management zones using mobile measurements of soil apparent electrical conductivity and multivariate geostatistical techniques[J] .Soil and Tillage Research, 2010,106 (2): 335-343.

There is defective in the method for determining the optimal classification number in Fuzzy c-means Clustering division farmland with FPI and NCE evaluation function: FPI and NCE are not high to the discrimination of farmland nutrient difference, caused final number of categories on the low side, and often have FPI and the inconsistent phenomenon of the corresponding number of categories of NCE minimum value to occur, cause determining suitable number of categories.

Summary of the invention

Technical matters to be solved by this invention be overcome above-mentioned when utilizing Fuzzy c-means Clustering that the farmland is divided, utilize FPI and NCE evaluation function to determine the defective of optimal classification number, a kind of more effective method of utilizing Fuzzy c-means Clustering that the farmland is divided is provided, improves farmland discrimination to soil nutrient when dividing.

The present invention is for solving the problems of the technologies described above by the following technical solutions:

A kind of farmland division methods based on Fuzzy c-means Clustering, adopt following steps:

Step (1) according to soil nutrient data and the positional information thereof of farmland sampled point, is done respectively the Ordinary Kriging Interpolation interpolation to different soil nutrient data, obtains the raster data of each soil nutrient; Then the raster data of Different Soil Nutrients is formed data pair by corresponding grid positions, obtain the sample matrix of all soil nutrients;

Step (2) is according to the maximum number of categories C in the given farmland of soil nutrient difference _Max, C _MaxFor greater than 2 natural number;

Step (3) utilizes Fuzzy c-means Clustering that sample matrix is carried out cluster analysis, and the number of categories of cluster analysis is [2, C _Max], obtain C _Max-2+1 kind cluster result;

Step (4) is calculated Clustering Effect index S, min cluster efficiency index S to every kind of cluster result respectively _MinThe corresponding number of categories of value be the optimal classification number, the cluster result that the optimal classification number is corresponding is that the result is divided in the farmland.

Further, a kind of farmland division methods based on Fuzzy c-means Clustering of the present invention, the sample matrix of described step (1) is the capable l of n row, wherein n be data to number, i.e. sample number; L is soil nutrient index number, and namely the every row in the sample matrix represent a kind of soil nutrient index.

Further, a kind of farmland division methods based on Fuzzy c-means Clustering of the present invention, the described Fuzzy c-means Clustering that utilizes of step (3) to the concrete steps that sample matrix carries out cluster analysis is:

1) establishing target function:

$J_m(U,V)=\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{c}{\mathrm{\&Sigma;}}}{\left(u_{\mathrm{ik}}\right)}^m{\left(d_{\mathrm{ik}}\right)}^2---\left(1\right)$

Wherein, n is sample number, and c is number of categories, and m is the FUZZY WEIGHTED index, u _IkK sample x among the expression sample matrix X _kBelong to i cluster centre ν in the cluster centre matrix V _iThe degree of membership value, U is the degree of membership matrix, d _Ik ²Be sample x _kWith ν _iSquare distance on proper vector, i, k are natural number;

2) by iteration, minimize (1) formula J _m(U, V), wherein u _IkReplaced d by (2) formula _Ik ²Calculate required ν _iReplaced by (3) formula,

$u_{\mathrm{ik}}={\left[\underset{j=1}{\overset{c}{\mathrm{\&Sigma;}}}{\left(\frac{{d_{\mathrm{ik}}}^2}{{d_{\mathrm{jk}}}^2}\right)}^m\right]}^{-1}---\left(2\right)$

$v_{\mathrm{ik}}=\frac{\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left(u_{\mathrm{ik}}\right)}^m{x}_k}{\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left(u_{\mathrm{ik}}\right)}^m}---\left(3\right)$

3) if the variation of cluster centre then stops iteration less than the threshold value of setting in twice circulation, otherwise repeating step 2), until the variation of cluster centre is less than the threshold value of setting or reach the iterations of setting, wherein said threshold value gets 10 ^-3-10 ^-5, iterations is got 50-500 time.

Further, a kind of farmland division methods based on Fuzzy c-means Clustering of the present invention, described step (3) Clustering Effect index S computing formula be,

$S=\frac{\underset{i=1}{\overset{c}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{l}{\mathrm{\&Sigma;}}}\frac{{\mathrm{CV}}_{\mathrm{ij}}}{n_{\mathrm{ij}}}}{\frac{1}{n}\underset{f=1}{\overset{l}{\mathrm{\&Sigma;}}}{\mathrm{CV}}_{\mathrm{vf}}};$

Wherein, c is number of categories, c ∈ [2, C _Max], l is soil nutrient index number, the i.e. columns of sample matrix; N is the line number of sample matrix, i.e. the sample number of soil nutrient, n _IjBe the data amount check of nutrient j in classification i, CV _IjBe i the coefficient of variation of classifying of nutrient j, i ∈ [1, c], j ∈ [1, l], CV _VfBe the coefficient of variation of calculating by all cluster centres of nutrient f, f ∈ [1, l].

The present invention adopts above technical scheme compared with prior art, has following technique effect:

The invention solves the defective of FPI and NCE in the background technology, provide a kind of Fuzzy c-means Clustering the farmland to be divided the method for determining the optimal classification number, improve farmland discrimination to soil nutrient difference when dividing, can satisfy preferably accurately management or plant growth information monitoring needs that the farmland is divided of farmland.

Description of drawings

Fig. 1 is that process flow diagram is divided in farmland of the present invention.

Fig. 2 is the Ordinary Kriging Interpolation interpolation result figure of soil nutrient; Wherein (a) is content of organic matter interpolation result figure, (b) is total nitrogen content interpolation result figure, (c) is quick-acting potassium content interpolation result figure, (d) is available phosphorus contents interpolation result figure.

Fig. 3 is Clustering Effect index S curve map in each classification situation.

Result schematic diagram is divided in Fig. 4 farmland.

Embodiment

Below in conjunction with accompanying drawing technical scheme of the present invention is described in further detail:

As shown in Figure 1, farmland of the present invention division methods specifically comprises following step:

1. with soil nutrient data and the positional information input GIS software of farmland sampled point, different nutrient datas is done respectively the Ordinary Kriging Interpolation interpolation, interpolation result is output as the raster data of ASCII fromat.The soil nutrient data of the same grid positions of different ascii text files form data to as a sample, and all samples are organized into sample matrix.Sample matrix is the capable l of n row, n be data to number, i.e. sample number, l is soil nutrient index number, the every row of sample matrix represent a kind of soil nutrient index.

2. the maximum number of categories C in given farmland _Max, C _MaxUsually get about 10.

3. utilize conventional Fuzzy c-means Clustering that sample matrix is carried out cluster analysis, the number of categories of cluster analysis is [2, C _Max], obtain C _Max-2+1 kind cluster result calculates respectively Clustering Effect index S to every kind of cluster result, min cluster efficiency index S _MinThe corresponding number of categories of value be the optimal classification number, the cluster result that the optimal classification number is corresponding is that the result is divided in the farmland.

In the technique scheme, the concrete steps that Fuzzy c-means Clustering is divided the farmland are:

1) establishing target function

$J_m(U,V)=\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{c}{\mathrm{\&Sigma;}}}{\left(u_{\mathrm{ik}}\right)}^m{\left(d_{\mathrm{ik}}\right)}^2---\left(1\right)$

Wherein, n is sample number, and c is number of categories, and m is the FUZZY WEIGHTED index, usually gets m=1.3, u _IkK sample x among the expression sample matrix X _kBelong to i cluster centre ν in the cluster centre matrix V _iThe degree of membership value, U is the degree of membership matrix.d _Ik ²Be sample x _kWith ν _iSquare distance on proper vector.

2) by iteration, minimize (1) formula J _m(U, V), wherein u _IkReplaced d by (2) formula _Ik ²Calculate required ν _iReplaced by (3) formula,

$u_{\mathrm{ik}}={\left[\underset{j=1}{\overset{c}{\mathrm{\&Sigma;}}}{\left(\frac{{d_{\mathrm{ik}}}^2}{{d_{\mathrm{jk}}}^2}\right)}^m\right]}^{-1}---\left(2\right)$

$v_{\mathrm{ik}}=\frac{\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left(u_{\mathrm{ik}}\right)}^m{x}_k}{\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left(u_{\mathrm{ik}}\right)}^m}---\left(3\right)$

If the variation of cluster centre then stops iteration less than the threshold value of setting in twice circulation, otherwise repeating step 2), until the variation of cluster centre less than the threshold value of setting or reach the iterations of setting.Threshold value gets 10 usually ^-3-10 ^-5, iterations is got 50-500 time usually.

In the technique scheme, the computing method of Clustering Effect index S are:

$S=\frac{\underset{i=1}{\overset{c}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{l}{\mathrm{\&Sigma;}}}\frac{{\mathrm{CV}}_{\mathrm{ij}}}{n_{\mathrm{ij}}}}{\frac{1}{n}\underset{f=1}{\overset{l}{\mathrm{\&Sigma;}}}{\mathrm{CV}}_{\mathrm{vf}}}---\left(4\right)$

Wherein, c is number of categories, c ∈ [2, C _Max], l is soil nutrient index number, and namely the columns of sample matrix is generally 3-10, and n is the line number of sample matrix, i.e. the sample number of soil nutrient, its value is determined by actual conditions, n _IjBe the data amount check of nutrient j in classification i, CV _IjBe i the coefficient of variation of classifying of nutrient j, i ∈ [1, c], j ∈ [1, l], CV _VfBe the coefficient of variation of calculating by all cluster centres of nutrient f, f ∈ [1, l].

Specific embodiment:

1) with Rugao City, Jiangsu Province soil organic matter content, total nitrogen content, quick-acting potassium content and available phosphorus contents and co-ordinate position information input ArcGIS 9.2,4 nutrients of soil are done respectively the Ordinary Kriging Interpolation interpolation, interpolation result is seen Fig. 2.Wherein (a) is content of organic matter interpolation result figure, (b) is total nitrogen content interpolation result figure, (c) is quick-acting potassium content interpolation result figure, (d) is available phosphorus contents interpolation result figure.Interpolation result is with the formal output of ascii text file, and the soil nutrient Organization of Data of same grid positions in 4 ascii text files is become a sample, obtains the sample matrix of capable 4 row of n, and n is sample number, and the every row of sample matrix represent a kind of soil nutrient data.

2) the maximum number of categories C in given farmland _Max=10.

3) cluster analysis.By conventional Fuzzy c-means Clustering sample matrix is carried out cluster analysis, in the parameter of Fuzzy c-means Clustering operation, Fuzzy Weighting Exponent m=1.3, iterations gets 100, and threshold value is 10 ^-5, number of categories is [2,10], obtains 9 kinds of cluster results through computing.

4) determine farmland division result.9 kinds of cluster results that step 3) is obtained calculate respectively Clustering Effect index S, the results are shown in Figure 3.As shown in Figure 3, the S value is minimum when dividing 8 class, and namely the optimal classification number is 8, and number of categories 8 corresponding clusters are the farmland and divide the result, see Fig. 4.

Claims (4)

1. farmland division methods based on Fuzzy c-means Clustering is characterized in that adopting following steps:

Step (1) according to soil nutrient data and the positional information thereof of farmland sampled point, is done respectively the Ordinary Kriging Interpolation interpolation to different soil nutrient data, obtains the raster data of each soil nutrient; Then the raster data of Different Soil Nutrients is formed data pair by corresponding grid positions, obtain the sample matrix of all soil nutrients;

Step (2) is according to the maximum number of categories C in the given farmland of soil nutrient difference _Max, C _MaxFor greater than 2 natural number;

Step (3) utilizes Fuzzy c-means Clustering that sample matrix is carried out cluster analysis, and the number of categories of cluster analysis is [2, C _Max], obtain C _Max-2+1 kind cluster result;

Step (4) is calculated Clustering Effect index S, min cluster efficiency index S to every kind of cluster result respectively _MinThe corresponding number of categories of value be the optimal classification number, the cluster result that the optimal classification number is corresponding is that the result is divided in the farmland.

2. a kind of farmland division methods based on Fuzzy c-means Clustering according to claim 1 is characterized in that: the sample matrix of described step (1) is the capable l row of n, wherein n be data to number, i.e. sample number; L is soil nutrient index number, and namely the every row in the sample matrix represent a kind of soil nutrient index.

3. a kind of farmland division methods based on Fuzzy c-means Clustering according to claim 1, it is characterized in that: the described Fuzzy c-means Clustering that utilizes of step (3) to the concrete steps that sample matrix carries out cluster analysis is:

1) establishing target function:

$J_m(U,V)=\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{c}{\mathrm{\&Sigma;}}}{\left(u_{\mathrm{ik}}\right)}^m{\left(d_{\mathrm{ik}}\right)}^2---\left(1\right)$

Wherein, n is sample number, and c is number of categories, and m is the FUZZY WEIGHTED index, u _IkK sample x among the expression sample matrix X _kBelong to i cluster centre ν in the cluster centre matrix V _iThe degree of membership value, U is the degree of membership matrix, d _Ik ²Be sample x _kWith the square distance of ν i on proper vector, i, k are natural number;

2) by iteration, minimize (1) formula J _m(U, V), wherein u _IkReplaced d by (2) formula _Ik ²Calculate required ν _iReplaced by (3) formula,

$u_{\mathrm{ik}}={\left[\underset{j=1}{\overset{c}{\mathrm{\&Sigma;}}}{\left(\frac{{d_{\mathrm{ik}}}^2}{{d_{\mathrm{jk}}}^2}\right)}^m\right]}^{-1}---\left(2\right)$

$v_{\mathrm{ik}}=\frac{\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left(u_{\mathrm{ik}}\right)}^m{x}_k}{\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left(u_{\mathrm{ik}}\right)}^m}---\left(3\right)$

3) if the variation of cluster centre then stops iteration less than the threshold value of setting in twice circulation, otherwise repeating step 2), until the variation of cluster centre is less than the threshold value of setting or reach the iterations of setting, wherein said threshold value gets 10 ^-3-10 ^-5, iterations is got 50-500 time.

4. a kind of farmland division methods based on Fuzzy c-means Clustering according to claim 1 is characterized in that described step (3) Clustering Effect index S computing formula is,

$S=\frac{\underset{i=1}{\overset{c}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{l}{\mathrm{\&Sigma;}}}\frac{{\mathrm{CV}}_{\mathrm{ij}}}{n_{\mathrm{ij}}}}{\frac{1}{n}\underset{f=1}{\overset{l}{\mathrm{\&Sigma;}}}{\mathrm{CV}}_{\mathrm{vf}}};$

Wherein, c is number of categories, c ∈ [2, C _Max], l is soil nutrient index number, the i.e. columns of sample matrix; N is the line number of sample matrix, i.e. the sample number of soil nutrient, n _IjBe the data amount check of nutrient j in classification i, CV _IjBe i the coefficient of variation of classifying of nutrient j, i ∈ [1, c], j ∈ [1, l], CV _VfBe the coefficient of variation of calculating by all cluster centres of nutrient f, f ∈ [1, l].

Images