CN107657264A - One kind carries out soil profile kind identification method based on KNN classification - Google Patents

Abstract

The invention discloses a soil profile type identification method based on KNN classification, which comprises the following steps: S1, extracting the trough characteristics of a spectral curve; S2, extracting the peak characteristics of a spectral curve; S3, training Collect and test set data, extract the trough characteristics of the soil spectral curve, and store them in the data set c1; S4, for the training set and test set data, after differential derivation processing, extract the peak and trough features of the first derivative curve of the soil spectrum, and store them in the data Set c2; S5, merge the data sets c1 and c2 to obtain the attribute data set L, after normalizing the attribute data set L, based on the KNN classification method, carry out the soil type identification of the soil spectral curve of the test set; S6, based on the maximum According to the proportion principle, the soil type of the soil profile of the test set is determined. This method has greater advantages in ultra-high-dimensional classification, and can be used in the case of high-dimensional sample data and many soil types.

Description

A Soil Profile Type Recognition Method Based on KNN Classification

technical field

The invention belongs to the field of soil remote sensing, and in particular relates to a method for identifying soil types based on K-Nearest Neighbor (KNN for short).

Background technique

Soil classification is based on the comprehensive differences in soil physical properties, classifying soil types at different levels, so as to promote agricultural technology and soil improvement technology according to local conditions. The identification of soil types needs to obtain a large amount of diagnostic information such as soil morphology, physics, chemistry, and even biology. Some of the information (such as morphology, bulk density, etc.) It can only be obtained after laboratory testing and analysis. Therefore, the identification and identification of soil types is usually costly and requires the involvement of soil taxonomy experts.

At present, in order to meet the rapidly developing demand for large-scale and high-precision soil information, rapid acquisition and continuous updating of soil information is one of the core research contents in the field of soil resource research. Due to its high cost and low efficiency, traditional methods usually have problems such as large sampling scale, sparse sampling density, and low survey frequency, making it difficult to meet the needs of dynamic, rapid, and low-cost acquisition and update of soil information.

The change of basic material composition of soil is not only the reflection of soil development process, but also the basis for determining soil diagnostic index and type classification, and also the main factor affecting soil reflectance spectrum. In soil survey and mapping, it is theoretically feasible to use spectral analysis technology to realize rapid, accurate and automatic identification of soil types. The key to realizing the above applications is to obtain soil spectral data under relatively standard measurement conditions, establish a complete soil spectral library, and then develop effective pattern recognition methods to achieve the goal of soil spectral identification and classification.

Liu Huanjun et al. (2008, 2017) based on the BP neural network method, obtained the accurate classification of five main soil types including black soil and chernozem in Nong'an County, Jilin Province (see "Research on Soil Classification Based on Reflectance Spectral Characteristics", Liu Huanjun, Zhang Bo , Zhang Yuanzhi, et al. Spectroscopy and Spectral Analysis, 2008,28(3):624-628.; Soil classification method using multi-layer perceptron neural network model combined with spectral characteristic parameters, Liu Huanjun, Zhang Xinle, etc. Chinese patent. CN 106650819A .2017-05-10.), there are fewer relevant soil types in this method, and the BP neural network method has certain advantages. However, when faced with many soil types and a large number of soil samples, the learning time of this method is too long, and may even fail to achieve the purpose of learning, so when faced with many soil types and a large number of soil sample data, the overall classification effect is not good. not ideal. In addition, only 0-20cm plow layer soil samples were used in related methods, and the soil spectral information at different depths in the same section was not fully utilized.

Contents of the invention

In view of the above-mentioned technical problems, the present invention aims at the soil spectral information at different depths in the soil profile, on the basis of extracting the characteristics of the soil spectral curves of the training set and the test set, and based on the KNN classification method, accurately and quickly perform the soil analysis of the soil profile of the test set. type identification.

In order to realize above-mentioned technical purpose, the present invention adopts following concrete technical scheme:

A method for identifying soil profile types based on KNN classification, comprising the following steps:

S1. Aiming at a spectral curve, extracting its trough features;

S2. Aiming at a spectral curve, extracting its peak features;

S3. According to the training set and test set data, extract the trough characteristics of the soil spectral curve, and store them in the data set c1;

S4. For the training set and the test set data, after differential derivation processing, the method of step S1 and step S2 is used to extract the peak and valley characteristics of the first order derivative curve of the soil spectrum, and store it in the data set c2;

S5. Combine the data sets c1 and c2 respectively obtained in step S3 and step S4 to obtain the attribute data set L. After normalizing the attribute data set L, based on the KNN classification method, carry out the soil type identification of the soil spectral curve of the test set ;

S6. Based on the principle of maximum proportion, determine the soil type of the soil profile of the test set.

The step S1 specifically includes:

S1.1: Valley extraction, using the valley extraction algorithm to extract valleys and store them in the set G={g _b |b=0,...,M-1}, where M is the number of valleys, g _b ={r' _a |a=0,...,N-1} is a set of trough elements, and N is the number of points in the trough;

S1.2: Valley feature extraction, for a wave valley g _b , perform the following operations:

a) Extract the minimum value point r' _min meeting the following conditions, which is the bottom point of the valley; r' _min ≤ r' _c (c=0,...,N-1)

Wherein, _r'c is the spectral reflectance value at any point in the trough;

b) Calculate the depth H and width Q of the trough according to formula (1) and formula (2);

H＝(r' ₀ +r' _N-1 )/2-r' _min (1)

Q=N (2)

Among them, r' ₀ is the value of the starting point of the trough, and r' _N-1 is the value of the ending point of the trough.

c) cyclically execute step a) to step b), until the feature extraction of all valleys of the spectral curve is completed;

d) Save the valley point band position min, valley point value r' _min , valley depth H and valley width Q of different valleys into the matrix V _M*4 row by row.

Described step S2 specifically comprises:

S2.1: Peak extraction, use the peak extraction algorithm to extract the peaks, and store them in the set G'={g'_b'|b'=0,...,M'-1}, where M' is the number of peaks, g'_b' = {r” _a' |a'=0,...,N'-1} is a set of peak elements, and N' is the number of points in the peak;

S2.2: Peak feature extraction, for a peak g'_b' , perform the following operations:

a) Extract the maximum value point r" _max that meets the following conditions, and this point is the peak apex of the wave peak;

r" _max ≥ r" _c '(c'=0,...,N'-1)

Wherein, r"_c' is the spectral reflectance value at any point in the peak;

b) According to the formula (3), (4), calculate the depth H' and the width Q' of the peak;

H'＝(r” ₀ +r” _N'-1 )/2-r” _max (3)

Q'=N' (4)

Among them, r” ₀ is the value of the starting point of the peak, and r” _N’-1 is the value of the ending point of the peak.

c) cyclically execute step a) to step b), until the feature extraction of all peaks of the spectral curve is completed;

d) Save the peak apex band position max, peak apex value r” _max , peak height H' and peak width Q' of different peaks into the matrix V'_M'*4 row by row.

Described step S3 specifically comprises:

S3.1: Read the spectral data of any soil profile in the training set and test set, adopt the equal-area quadratic spline interpolation method, and perform depth interpolation processing for the reflectance values of each occurrence layer of the soil profile to obtain the preset distance interval, Soil spectral reflectance at different depths T={t _i,j |i=0,...,n-1; j=0,...,m-1}, where n represents the soil spectrum after interpolation The number of curves, m represents the number of soil spectral bands, t _i,j represents the reflectance value of the i-th spectral curve generated by interpolation at band j;

S3.2: data smoothing, for the soil spectral reflectance obtained through depth interpolation processing, perform 11-point moving average smoothing processing to obtain the smoothed data set T';

S3.3: Envelope removal processing, use the envelope removal method to normalize any spectral curve in T', and store the results in the set R={r _e |e=0,...,h-1 }; where r _e represents the normalized result of the reflectance of the spectral curve at the band e+350, and h is the number of bands of the spectral curve;

S3.4: trough feature extraction, perform the aforementioned step S3 on the interpolation soil layer data in the set R, obtain the characteristic matrix S _P*4 of the soil spectral curve of each interpolation soil layer, wherein, P represents the number of troughs of the interpolation soil layer spectral curve; Extract the first two trough features of the matrix S and save them in the set c1;

S3.5: cyclically execute steps S3.1 to S3.4 until the data processing of all soil spectral profiles is completed.

Described step S4 specifically comprises:

S4.1: Utilize the aforementioned steps S3.3 to obtain the smoothed data set T' of any soil profile spectral data;

S4.2: According to the formula (5), the data T' is differentially and amplified, and the result is stored in the set Y={y _{i, j} |i=0,...,n-1; j=0,... .,m-1} in;

Among them, D is the amplification factor;

S4.3: Envelope removal processing, use the envelope removal method to normalize any spectral curve in the set Y, and store the results in the set R={r _e |e=0,...,h-1 }; where r _e represents the normalized result of the reflectivity of the spectral curve at the band e+350, and h is the number of bands of the spectral curve;

S4.4: Wave trough feature extraction, perform the aforementioned step S4 on the interpolation soil layer data in the set R, and obtain the soil spectral curve characteristic matrix U _X*4 of each interpolation soil layer, where X represents the trough of the first-order derivative curve of the interpolation soil layer Quantity; select the bottom value and the bottom wave band of the first trough in each interpolation soil layer soil spectral curve characteristic matrix U _{X * 4} as the soil spectrum first derivative curve trough feature of the interpolation soil layer and save it in the set c2;

S4.5: Peak feature extraction, perform the aforementioned step S2 on any interpolation soil layer data in the set Y, and obtain the soil spectral curve characteristic matrix W _Z*4 of each interpolation soil layer, where Z represents the first-order derivative curve of the interpolation soil layer The number of peaks; select the peak value and peak band of the first peak in the soil spectral curve characteristic matrix W _Z*4 of each interpolation soil layer as the peak characteristic of the soil spectrum first derivative curve of the interpolation soil layer and save it in the set c2 ;

S4.6: cyclically execute steps S4.1 to S4.5 until the data processing of all soil spectral profiles is completed.

Described step S5 specifically comprises:

S5.1: Combine the data sets c1 and c2 respectively obtained in step S3 and step S4 to obtain attribute data set L={l _i,j |i=0,...,l1-1; j=0,... ,l2-1}, where, l1 represents the number of interpolated soil layers of the entire soil profile, and l2 represents the number of characteristics of each soil layer;

S5.2: According to the formula (6), perform normalization operation on all feature elements l _i,j in the data set L, and obtain the normalized feature set L';

l' _i,j ＝(l _i,j -l _min,j )/(l _max,j -l _min,j ) (6)

Among them, l _min,j is the minimum value of the jth feature of all interpolation soil layers, and l _max,j is the maximum value of the jth feature of all interpolation soil layers;

S5.3: Set the parameter k, call the KNN algorithm, and obtain the classification results of the interpolated soil layer spectral curves of the test set samples.

Described step S6 specifically comprises:

S6.1: Calculate the proportion λ(A) of each soil type in the classification results of each interpolated soil layer spectral curve in a soil profile; find the maximum value MAX _{λ(A) of λ(A)} , then consider the profile to be class A;

S6.2: Step S6.1 is executed cyclically until the soil type discrimination of all soil profiles in the test set is completed.

Compared with the existing soil identification technology, the present invention has the following beneficial effects:

The present invention proposes a method of comprehensively utilizing the characteristics of the soil spectral curve after envelope removal processing and the soil spectral curve characteristics after derivation processing, and using the KNN classification method to quickly identify the type of soil profile. The method mainly has the following characteristics:

1) Make full use of the spectral information of soil samples at different depths.

2) The characteristics of the soil spectral curve and the characteristics of the soil spectral curve after derivation processing are comprehensively used.

3) Compared with the BP neural network method, the KNN method used has greater advantages in ultra-high-dimensional classification, which can be used in the case of high-dimensional sample data and many soil types.

Description of drawings

Fig. 1 is the flowchart of the inventive method;

Detailed ways

The technical solutions of the present invention will be described in further detail below in conjunction with the accompanying drawings and embodiments.

The soil samples used in this example were collected from 649 profiles. These profiles are divided into 10 soil classes, 23 subclasses, 56 soil classes and 120 subclasses according to the Chinese Soil System Classification. A Cary 5000 spectrophotometer (Agilent Technologies) was used to measure the 350-2500 nm diffuse reflectance spectrum of the sample, and the spectral sampling interval was 1 nm.

In this embodiment, 553 soil spectral profile data are selected as a training set, and 96 soil spectral profile data are selected as a test set. For the whole process of data preprocessing, spectral feature extraction, spectral classification based on KNN, soil profile type identification, etc., the present invention is further described in detail. Fig. 1 is the flowchart of the method of the present invention.

1. Feature extraction of soil spectral curves.

1-1. Read the spectral data of any soil profile in the training set and test set, set the interpolation interval to 1cm, and use the equal-area quadratic spline interpolation method to perform depth interpolation processing for the reflectance values of each interpolated soil layer in the soil profile , obtain soil spectral reflectance T={t _i,j |i=0,...,73432; j=0,...,2150} at different depths at intervals of 1 cm, as shown in Table 1 below.

Table 1

1-2. For the soil spectral reflectance t _i,j obtained through depth interpolation, perform 11-point moving average smoothing to obtain the smoothed data set T', as shown in Table 2 below.

Table 2

1-3. Valley feature extraction. Perform envelope removal processing on any interpolated soil layer data in the set T′, and perform the aforementioned step S1 to obtain the soil spectral curve characteristic matrix of each interpolated soil layer. As shown in Table 3 below;

table 3

2. Feature extraction of soil spectral first derivative curve.

2-1. Using the aforementioned step S3.3 to obtain the smoothed data set T′ of the soil profile spectral data (as shown in Table 2);

2-2. According to the formula (5), the data T' is differentially and amplified, and the result is stored in the set Y, as shown in Table 4 below;

Table 4

2-3. Valley feature extraction: Perform envelope removal processing on any interpolated soil layer data in the set Y, and perform the aforementioned step S1 to obtain the valley feature matrix U of the first-order derivative curve of the soil spectrum of each interpolated soil layer. Select the valley bottom value and the valley bottom wave band of the first valley in the first derivative curve valley feature matrix U of each interpolation soil layer soil spectrum as the first derivative curve valley characteristics of the interpolation soil layer and save them in the set c2;

2-4. Peak feature extraction: Perform envelope removal processing on any interpolated soil layer data in the set Y, and perform the aforementioned step S2 to obtain the peak characteristic matrix W of the first-order derivative curve of the soil spectrum of each interpolated soil layer. Select the peak value and peak band of the first peak in the peak feature matrix W of the soil spectrum first-order derivative curve of each interpolation soil layer as the peak feature of the first-order derivative curve of the interpolation soil layer and save them in the set c2;

2-5. Perform steps 2-1 to 2-4 in a loop until the data processing of all soil spectral profiles in the training set and test set is completed, as shown in Table 5 below;

table 5

3. Soil type identification of test set soil interpolation soil layer spectral curve.

3-1. Merge the data sets c1 and c2 to obtain the feature data set L. The curve features involved in the calculation mainly include: soil spectral curve difference depth UpperDepth, LowerDepth, the position of the bottom of the first absorption valley Position1, the bottom value Value1, and the bottom of the wave Depth1, trough width xDistance1, the position of the bottom of the second absorption valley Position2, the value of the bottom of the valley Value2, the depth of the trough Depth2, the width of the trough xDistance2, the position of the bottom of the first absorption valley of the soil spectrum first derivative curve P1, the bottom of the valley value V1, the first The peak position P2 and the peak value V2 of an absorption peak; as shown in Table 6 below;

Table 6

3-2. According to the formula (6), perform normalization operation on all feature elements l _{i, j} in the data set L to obtain the normalized feature set L′, as shown in Table 7 below;

Table 7

3-3. Set the parameter k=201, use R.NET to call the KNN algorithm, and obtain the classification results of each interpolated soil layer spectral curve of the test set sample, as shown in Table 8 below;

Table 8

4. Soil type discrimination of test set soil spectral profile

4-1. Calculate the proportion λ(A) of each soil type in the classification results of the spectral curves of the differential soil layers in the first soil profile in the test set, as shown in Table 9 below, and judge that the profile is dry sandy new soil ;

Group CST Percent dry sandy new soil 46.67% Simplified moist ferroaluminite 10.83% Aluminous normal wet prototype soil 0.01% submerged hydroponic artificial soil 0.05% Ginger moist rudimentary soil 2.50% wet normal new soil 15.83% Iron infiltration hydroponic man-made soil 9.17% ferro-moist leaching soil 5.83% purple moist rudimentary soil 3.33%

Table 9

4-2. Repeat step 4-1 until the soil type discrimination of all soil profiles in the test set is completed, and the results are shown in Table 10 below;

Table 10

In the above application examples, only the identification of soil types is performed. This method is also applicable to the identification of soil types at different levels such as soil classes, subclasses, and subclasses; and in this embodiment, the identification of soil types is only based on visible-near-infrared diffuse reflectance spectral data, and this method is also applicable to Other types of soil spectral data such as mid-infrared diffuse reflectance spectroscopy.

5. Test analysis.

It can be seen from the above-mentioned examples that: at the level of soil type identification of the soil profile, the identification method is suitable for light-colored moist rudimentary soil, simplified hydroponic artificial soil, iron-accumulating hydroponic artificial soil, viscose moist iron-rich soil, gleation hydroponic artificial soil, etc. The recognition rate of soil type has reached more than 60%. In this embodiment, the test set data covers 6 provinces, involves 21 soil types and 96 soil samples, and the overall classification accuracy reaches 54.8%. With the continuous enrichment of profile data in the training set and the continuous balance of different types of soil profile data, the recognition accuracy will continue to improve.

Compared with the method of Liu Huanjun et al. (2008, 2017), this method has the following characteristics:

1) Make full use of the spectral information of soil samples at different depths;

2) Utilize the characteristics of the soil spectral curve and the characteristics of the soil spectral curve after derivation processing;

3) The KNN method used has greater advantages in ultra-high-dimensional classification than the BP neural network method, and can be used in the case of high-dimensional sample data and many soil types.

Claims (7)

1. one kind carries out soil profile kind identification method based on KNN classification, it is characterised in that including following steps：

S1, for a curve of spectrum, extract its trough feature；

S2, for a curve of spectrum, extract its crest feature；

S3, for training set and test set data, extract soil spectrum curve trough feature, deposit data acquisition system c1；

S4, for training set and test set data, after difference derivation is handled, using step S1 and step S2 methods, extraction soil Earth spectrum first derivative curve Wave crest and wave trough feature, deposit data acquisition system c2；

Data acquisition system c1, c2 that S5, combining step S3 and step S4 are respectively obtained, attribute data collection L is obtained, to attribute data collection After operation is normalized in L, based on KNN sorting techniques, the soil types identification of test set soil spectrum curve is carried out；

S6, based on maximum accounting principle, determine the soil types of test set soil profile.

2. according to claim 1 carry out soil profile kind identification method based on KNN classification, it is characterised in that described Step S1 is specifically included：

S1.1：Trough extracts, and extracts trough using trough extraction algorithm, and be stored in set G={ g_b| b=0 ..., M-1 } in, its In, M is trough quantity, g_b={ r'_a| a=0 ..., N-1 } it is a trough element set, N is the quantity of the point in the trough；

S1.2：Trough feature extraction, for a trough g_b, perform following operate：

A) the minimum point r' for meeting following conditions is extracted_min, the point is the lowest point point of the trough；

r'_min≤r'_c(c=0 .., N-1)

Wherein, r'_cFor the spectral reflectance values at any point in the trough；

B) according to formula (1) and formula (2), the depth H and width Q of the trough are calculated；

H=(r'₀+r'_N-1)/2-r'_min (1)

Q=N (2)

Wherein, r'₀Point value, r' are originated for trough_N-1Point value is terminated for trough.

C) circulation performs step a) to step b), until completing the feature extraction of all troughs of the curve of spectrum；

D) by the lowest point point band po sition min of different troughs, the lowest point point value r'_min, trough depth H and trough width Q are preserved line by line To matrix V_M*4In.

3. according to claim 1 carry out soil profile kind identification method based on KNN classification, it is characterised in that described Step S2 is specifically included：

S2.1：Crest extracts, and extracts crest using crest extraction algorithm, and be stored in set G'={ g'_b'| b'=0 ..., M'-1 } In, wherein M' is crest quantity, g'_b'={ r "_a'| a'=0 ..., N'-1 } it is a crest element set, N' is in the crest Point quantity；

S2.2：Crest feature extraction, for a crest g'_b', perform following operate：

A) the maximum of points r " for meeting following conditions is extracted_max, the point is the peak maximum of the crest；r”_max≥r”_c'(c'= 0,..,N'-1)

Wherein, r "_c'For the spectral reflectance values at any point in the crest；

B) according to formula (3) and formula (4), calculate the crest depth H ' and width Q'；

H'=(r "₀+r”_N'-1)/2-r”_max (3)

Q'=N'(4)

Wherein, r "₀Point value, r " are originated for crest_N'-1Point value is terminated for crest.

C) circulation performs step a) to step b), until completing the feature extraction of all crests of the curve of spectrum；

D) by the peak maximum band po sition max of different crests, summit point value r "_max, crest height H' and wave peak width Q' are protected line by line Be stored to matrix V '_M'*4In.

4. according to claim 1 carry out soil profile kind identification method based on KNN classification, it is characterised in that described Step S3 is specifically included：

S3.1:Any soil profile spectroscopic data in training set and test set is read, using equivalance Quadratic Spline Interpolation method, pin The reflectance value that layer occurs each to soil profile, carries out depth interpolation processing, obtains pre-determined distance interval, the soil at different depth Earth spectral reflectivity T={ t_i,j| i=0 ..., n-1；J=0 ..., m-1 }, wherein, n represents the soil spectrum curve after interpolation Quantity, m represent soil spectrum wave band number, t_i,jReflectance value of i-th curve of spectrum that expression interpolation is generated at wave band j；

S3.2:Data smoothing, for the soil spectrum reflectivity obtained through depth interpolation processing, carry out 11 rolling averages and put down Sliding processing, obtain smooth rear data acquisition system T'；

S3.3:Envelope place to go is handled, and using envelope removal method, any curve of spectrum in T' is normalized, and is tied Fruit deposit set R={ r_e| e=0 ..., h-1 } in；Wherein, r_eRepresent the normalizing of curve of spectrum reflectivity at wave band e+350 Change result, h is the curve of spectrum wave band quantity；

S3.4:Trough feature extraction, abovementioned steps S3 is performed to interpolation soil layer data in set R, obtains each interpolation soil layer soil Curve of spectrum eigenmatrix S_P*4, wherein, P represents the trough quantity of the interpolation soil layer curve of spectrum；Extract matrix S the first two Trough feature, preserve into set c1；

S3.5:Circulation performs step S3.1 to step S3.4, until completing the data processing of all soil spectrum sections.

5. according to claim 1 carry out soil profile kind identification method based on KNN classification, it is characterised in that described Step S4 is specifically included：

S4.1:Data acquisition system T' after obtaining any soil profile spectroscopic data smoothly using abovementioned steps S3.3；

S4.2:According to formula (5), difference and enhanced processing are carried out to data T', are as a result stored in set Y={ y_i,j| i=0 ..., n-1；J=0 ..., m-1 } in；

Wherein, D is amplification factor；

S4.3:Envelope removal is handled, and using envelope removal method, any curve of spectrum in set Y is normalized, As a result set R={ r are stored in_e| e=0 ..., h-1 } in；Wherein, r_eRepresent that curve of spectrum reflectivity at the wave band e+350 is returned One changes result, and h is the curve of spectrum wave band number；

S4.4:Trough feature extraction, abovementioned steps S4 is performed to interpolation soil layer data in set R, obtains each interpolation soil layer soil Curve of spectrum eigenmatrix U_X*4, wherein, X represents the trough quantity of the interpolation soil layer first derivative curve；Choose each interpolation soil layer Soil spectrum curvilinear characteristic matrix U_X*4In first trough the lowest point value and soil spectrum of the lowest point wave band as the interpolation soil layer First derivative curve trough feature is preserved into set c2；

S4.5:Crest feature extraction, abovementioned steps S4 is performed to any interpolation soil layer data in set Y, obtains each interpolation soil layer Soil spectrum curvilinear characteristic matrix W_Z*4, wherein, Z represents the crest quantity of the interpolation soil layer first derivative curve；Choose each interpolation Soil layer soil spectrum curvilinear characteristic matrix W_Z*4In first crest summit value and soil of the summit wave band as the interpolation soil layer Spectrum first derivative curve crest feature is preserved into set c2；

S4.6:Circulation performs step S4.1 to S4.5, until completing the data processing of all soil spectrum sections.

6. according to claim 1 carry out soil profile kind identification method based on KNN classification, it is characterised in that described Step S5 is specifically included：

S5.1：Data acquisition system c1, c2 that combining step S3 and step S4 are respectively obtained, obtain attribute data collection L={ l_i,j| i= 0,...,l1-1；J=0 ..., l2-1 }, wherein, l1 represents whole soil profile interpolation soil layer quantity, and l2 represents each soil layer Feature quantity；

S5.2：According to formula (6), to all characteristic element l in data set L_i,jOperation is normalized, it is special after being normalized L' is closed in collection；

l'_i,j=(l_i,j-l_min,j)/(l_max,j-l_min,j) (6)

Wherein, l_min,jFor the minimum value of all j-th of features of interpolation soil layer, l_max,jFor all j-th of features of interpolation soil layer most Big value；

S5.3：Arrange parameter k, KNN algorithms are called, obtain the classification results of each interpolation soil layer curve of spectrum of test set sample.

7. according to claim 1 carry out soil profile kind identification method based on KNN classification, it is characterised in that described Step S6 is specifically included：

S6.1：Calculate the proportion λ (A) of each soil types in each interpolation soil layer curve of spectrum classification results in a soil profile；Look for To λ (A) maximum MAX_λ(A), then it is assumed that the section is class A；

S6.2：Circulation performs step S6.1, until the soil types for completing all soil profiles in test set differentiates.