CN109034189B

CN109034189B - Forest type identification method based on high-resolution remote sensing image

Info

Publication number: CN109034189B
Application number: CN201810621443.0A
Authority: CN
Inventors: 张贵; 肖化顺; 张琦; 邱书志; 周璀
Original assignee: Central South University of Forestry and Technology
Current assignee: Central South University of Forestry and Technology
Priority date: 2018-06-15
Filing date: 2018-06-15
Publication date: 2022-04-08
Anticipated expiration: 2038-06-15
Also published as: CN109034189A

Abstract

The invention relates to a forest type identification method based on high-resolution remote sensing images, which comprises the following steps: preprocessing the high-resolution remote sensing image to obtain a panchromatic spectral image and a multispectral image; carrying out gray level extraction from the panchromatic spectral image, and carrying out vegetation index extraction from the panchromatic spectral image; analyzing the dominant tree species by combining the second-class survey data of the forest resources, and obtaining the gray value and the vegetation index value of the dominant tree species; dividing a research area into a forest land and a non-forest land according to a forest land classification standard, and respectively establishing a relevant forest type identification function model according to the correlation between the gray value and the vegetation index value; checking and evaluating the correlation function model; and applying the correlation function model to forest type identification. The forest type can be identified by acquiring the gray scale and the vegetation index by utilizing the texture features of the high-resolution second image, the field investigation work of forest resources is reasonably reduced, and the cost and the resources are saved.

Description

Forest type identification method based on high-resolution remote sensing image

Technical Field

The invention relates to a forest type identification method based on high-resolution remote sensing images.

Background

Forest type identification is an important content of forest resource management and monitoring. The traditional forest type identification method is relatively backward, and the forest type identification method relies on a large amount of manual work to carry out identification on the spot, so that the problems of high working strength, poor timeliness, inaccurate monitoring result, error, timeliness and the like exist, a large amount of manpower, material resources and financial resources are consumed, and the expected requirements cannot be met. With the change of time, the state of forest resources is dynamically changed, and the traditional means can not accurately master the change of forest types.

With the successful emission of domestic high-resolution series remote sensing satellites in China, high-resolution remote sensing images are widely applied to earth observation. The high-resolution remote sensing image data has clear texture, and the cost for acquiring the high-resolution remote sensing image data is greatly reduced due to the development of science and technology. The application of the high-resolution binary (GF-2) satellite image data means that the application of the remote sensing satellite in China enters a sub-meter high-resolution era.

If the high-resolution image processing technology can be fully utilized to identify the forest type, the strength and blindness of manual actual measurement in the field can be effectively reduced. The accuracy and the timeliness of forest type identification can be improved by analyzing and researching the extracted data of the high-resolution remote sensing image.

Therefore, a method for effectively identifying the forest type through high-resolution remote sensing images is needed.

Disclosure of Invention

According to the purpose of the invention, the forest type identification method based on the high-resolution remote sensing image comprises the following steps:

preprocessing the high-resolution remote sensing image to obtain a panchromatic spectral image and a multispectral image;

carrying out gray level extraction from the panchromatic spectral image, and carrying out vegetation index extraction from the panchromatic spectral image;

analyzing the dominant tree species by combining the second-class survey data of the forest resources, and obtaining the gray value and the vegetation index value of the dominant tree species;

dividing the research area into a forest land and a non-forest land according to a forest land classification standard, and dividing coniferous forests, broadleaf forests, coniferous commingled forests, broadleaf commingled forests and coniferous and broad commingled forests in the forest land to obtain gray values and vegetation index values of the coniferous forests, the broadleaf forests, the coniferous commingled forests, the broadleaf commingled forests and the coniferous and broad commingled forests;

respectively establishing function models for identifying related forest types according to the correlation relationship between the gray value and the vegetation index value;

checking and evaluating the correlation function model;

and applying the correlation function model to forest type identification.

According to one embodiment of the invention, the establishment of the correlation function model comprises the establishment of a forest type identification function model of the conifer forest, namely, during forest type identification, the high-score second-number image can be identified by the following function model relational expression according to the gray scale and the vegetation index:

y＝0.9061x+0.109

R²＝0.9677

wherein the formula x represents the vegetation index of the conifer forest, y represents the gray value of the coordinate point corresponding to the vegetation index, R²Indicating an accuracy of 96.77%, x is at 0,0.4]Values are taken over the interval.

According to one embodiment of the invention, the establishment of the correlation function model comprises the establishment of a forest type identification function model of the broad-leaved forest, namely, during forest type identification, the high-grade second image can be used for identification through the following function model relational expression according to the gray level and the vegetation index:

y＝32.289x³-22.443x²+5.9118x-0.3852

R²＝0.983

wherein the formula x represents the gray value of the broad-leaved forest, y represents the corresponding vegetation index in the corresponding broad-leaved forest, and the precision R²98.3% of the total amount of the compound, x being [0.25, 0.45 ]]Values are taken over the interval.

According to one embodiment of the invention, the establishment of the correlation function model comprises the establishment of a forest type identification function model of the pin leaf mixed forest, namely, during forest type identification, the high-score second image can be used for identification through the following function model relational expression according to the gray scale and the vegetation index:

y＝0.0548ln(x)+0.2879

R²＝0.9692

wherein x represents the vegetation index of the pin-leaved mixed forest, y represents the gray value of the coordinate point corresponding to the vegetation index, and R²Indicating an accuracy of 96.92%, x is at [0,0.4 ]]Values are taken over the interval.

According to one embodiment of the invention, the establishing of the correlation function model comprises establishing a forest type identification function model of the broad leaf mixed forest, and fitting the vegetation index and the gray level to obtain a function model relational expression:

y＝22.963x³-7.2359x²+1.6298x-0.0357

R²＝0.996

wherein x represents the gray level of the broad-leaved forest, y represents the vegetation index value of the specific coordinate point corresponding to the gray level, and R²Indicating an accuracy of 99.6%, x is [0,0.3 ]]Values are taken over the interval.

According to one embodiment of the invention, the establishing of the correlation function model comprises establishing a forest type identification function model for broad mixed forests, and fitting the vegetation index and the gray level to obtain a function model relational expression:

y＝10.197x²-0.5278x+0.0077

R²＝0.9901

wherein x represents the gray scale of the mixed forest, y represents the gray scale value of the coordinate point corresponding to the gray scale of the mixed forest, R²Indicating an accuracy of 99.01%, x is [0,0.25 ]]Values are taken over the interval.

According to one embodiment of the invention, the establishment of the correlation function model comprises a non-forest land forest type identification function model, statistical analysis is carried out on data of non-forest land vegetation indexes, images of the non-forest land are acquired by adopting multispectral texture characteristics, and the correlation relation is as follows:

y＝0.1971e^5.633x

R²＝0.9261

wherein the formula x represents the vegetation index of the non-forest land, y represents the gray value of the coordinate point corresponding to the vegetation index of the non-forest land, R²Expressed accuracy, its value was 92.61%, x was [ -0.25,0.25 [ ]]Values are taken over the interval.

Preferably, the checking and evaluating of the correlation function model comprises checking and verifying by using typical representatives of thirty plaques in the checking area, judging a verification result according to a distribution curve of the function model, and comparing the verification result with second-class survey data of forest resources.

In order to further verify the accuracy of the experimental result, experimental sample points can be randomly selected in different longitudes and latitudes of the inspection area respectively, and further verification is performed by combining a function model.

Aiming at the difficulties faced by the forest type identification at present, the invention considers that a large amount of manpower, material resources and financial resources are needed to obtain the specific field investigation data in the field investigation and monitoring process, and the obtained forest type data is influenced by a plurality of factors, has low accuracy and low dynamics and is not beneficial to the management and monitoring of forest resources. The method comprises the steps of taking a high-score second image as a technical basis for forest classification, analyzing the best method for identifying the current forest type by combining a series of methods such as ARCGIS software, PIE image processing, ENVI vegetation index calculation and statistical samples and the like through data analysis, analyzing the correlation between the forest gray scale and the forest vegetation index in a research area, establishing a function model, further demonstrating the accuracy of an experiment by using other forest sample data, and finally applying the experiment result to forest type identification.

Drawings

FIG. 1 is a flow chart of forest type identification;

FIG. 2 is a flow chart of remote sensing image processing;

FIG. 3 is a forest type map of a research area;

FIG. 4 is a gray scale distribution diagram of conifer forest;

FIG. 5 is a broadleaf forest gray distribution graph;

FIG. 6 is a gray level distribution diagram of a pin leaf mixed forest;

FIG. 7 is a gray distribution diagram of broadleaf mixed forest;

FIG. 8 is a gray scale distribution plot for a pin-broad commingled forest;

FIG. 9 is a gray scale distribution plot for a non-forestry land;

FIG. 10 is a gray scale interval of a study area;

FIG. 11 is a conifer vegetation index;

FIG. 12 is broadleaf forest vegetation index;

FIG. 13 is a conifer commingled forest vegetation index;

FIG. 14 is a broad-leaved commingled forest vegetation index;

FIG. 15 is a vegetation index for Mirabilitum;

FIG. 16 is a non-forestry vegetation index;

FIG. 17 is a conifer gray scale distribution function formula;

FIG. 18 is a relation of index distribution function of conifer vegetation;

FIG. 19 is a relation of a conifer forest type identification function;

FIG. 20 is a broad-leaved forest gray scale distribution function formula;

FIG. 21 is a broad-leaved forest vegetation index distribution function formula;

FIG. 22 is a relation of broadleaf forest type identification functions;

FIG. 23 is a functional relation of gray scale distribution of conifer commingled forests;

FIG. 24 is a relationship of index distribution function of conifer commingled forest vegetation;

FIG. 25 is a functional relation for identifying forest type of a pin leaf mixed forest;

FIG. 26 is a plot of the gray scale distribution function of broadleaf commingled forests;

FIG. 27 is a graph of a broad-leaved mixed forest vegetation index distribution function;

FIG. 28 is a relation of broad leaf commingled forest type identification function;

FIG. 29 is a graph showing a relationship of gray scale distribution for the broad commingled forest;

FIG. 30 is a graph of the relationship between the distribution functions of vegetation indexes for Mirabilitum and broadside mixed forest;

FIG. 31 is a functional relation for broad mixed forest type identification;

FIG. 32 is a non-forest gray scale distribution function equation;

FIG. 33 is a non-forestry vegetation index profile;

FIG. 34 is a non-woodland type identification function relationship;

FIG. 35 shows a forest farm in the inspection area, such as a Wangchai forest farm and a ditch forest farm;

FIG. 36 is a representative gray scale 1 image of an inspection sample;

FIG. 37 is a representative 2-gray scale image of an inspection sample;

FIG. 38 is a table of test samples representing a 3-gray scale image;

FIG. 39 is a chart of a test sample table representing 1a vegetation profile;

FIG. 40 is a graph of a test sample representation 2 vegetation distribution;

FIG. 41 is a graph of a test sample representative 3 vegetation profile;

FIG. 42 is a chart of checkpoint positions;

FIG. 43 is a graph showing the results of domain inspection;

FIG. 44 is a chart of forest farm inspection area results;

FIG. 45 is a graph comparing the results of the two types of tests.

Detailed Description

As shown in fig. 1, a flow chart of forest type identification according to the present invention is obtained by preprocessing a high-resolution remote sensing image to obtain a panchromatic spectral image and a multispectral image, extracting gray levels from the panchromatic spectral image, extracting a vegetation index from the multispectral image, analyzing dominant tree species by combining forest resource secondary survey data, and obtaining gray levels and vegetation index values of the dominant tree species. According to the classification standard of the forest land, the research area is divided into the forest land and the non-forest land, the forest land is divided into coniferous forest, broadleaf forest, coniferous mixed forest, broadleaf mixed forest and coniferous mixed forest, correlation function models are respectively established, and the correlation function models are applied to forest type recognition after inspection and evaluation.

The research area is located in the wax mouth forest farm of the Ministry of forestry of the south of Gansu province, and the total area is 48560 hectare. The method comprises the steps of adopting high-resolution binary (GF-2) remote sensing image data, wherein the data level is L1A, the panchromatic image resolution is 0.8m, the multispectral image resolution is 3.2m, the data range is panchromatic/multispectral image data which takes decimal longitude and latitude (103.579E, 34.016N) as the center and has the radius of 60 kilometers, obtaining 60 images in total, and processing the 60 images in two steps. The remote sensing image processing process is shown in fig. 2: firstly, the multispectral image is subjected to radiation correction, image registration and orthorectification. The preprocessing of the high-resolution second remote sensing image comprises radiation correction and geometric correction, and the radiation correction and the geometric correction are used for correcting radiation deformation and geometric deformation of the image acquired from the sensor. Due to the influences of factors such as illumination difference in the image, change of a geometric field angle, difference of atmospheric conditions, noise of the sensor and the like, the remote sensing image acquired by the sensor has distortion, and forest type identification is influenced by shadows generated by landforms and tree crowns, so that preprocessing of the high-resolution second remote sensing image is particularly important for forest type identification. The invention obtains the full-color spectrum image map of the overlapped forestry office by performing radiation calibration on the full-color spectrum image map to obtain reflectivity, image registration and image fusion after orthorectification. According to the rich stratification degree of trees in the second type of forest resource survey data and the detailed degree of classification data of broad-leaved forest class, the fused panchromatic spectrum and multispectral image of the overlapped forest bureau are combined with the second type of survey data, the outline of a wax opening is cut, a study area image map is obtained, and fig. 3 is a study area forest zoning map.

Combining the tree species collection condition in the second type of forest resource investigation data, using Arcgis to perform data processing, dividing the data into six types of attribute statistical tables, respectively performing segmentation and mask processing, extracting vector data graphs of coniferous forest, broadleaf forest, coniferous mixed forest, broadleaf mixed forest and coniferous mixed forest, introducing the vector graphs into ENVI or PIE to process the vector data graphs to obtain the layers of the coniferous forest, broadleaf forest, coniferous mixed forest, broadleaf mixed forest and coniferous mixed forest, respectively randomly sampling the coniferous forest, broadleaf forest, coniferous mixed forest, broadleaf mixed forest and coniferous mixed forest by different colors, and randomly sampling 50 samples to perform gray value statistics and inspection. Fig. 4 is a gray scale distribution diagram of a conifer forest, fig. 5 is a gray scale distribution diagram of a broadleaf forest, fig. 6 is a gray scale distribution diagram of a conifer mixed forest, fig. 7 is a gray scale distribution diagram of a broadleaf mixed forest, fig. 8 is a gray scale distribution diagram of a broadleaf mixed forest, and fig. 9 is a gray scale distribution diagram of a non-forest land.

Importing a full-color spectral image map of a research area, processing image gray data by adopting ENVI or PIE, classifying and counting gray values by using Arcgis according to patches divided according to coniferous forests, broadleaf forests, coniferous mixed forests, broadleaf mixed forests, coniferous mixed forests, non-forest lands and the like in forest resource secondary survey data, and summarizing to obtain a gray value range of the whole research area between [0, 0.8], wherein the gray value range of the research area is shown in figure 10.

Arcgis is applied to classify the attribute table of the second-class survey data of forest resources in the research area, and the full-color spectral image is divided into five classes, namely coniferous forest, broadleaf forest, coniferous mixed forest, broadleaf mixed forest and coniferous mixed forest. The five classes of small shifts respectively obtain gray value intervals of the five classes of small shifts through superposition analysis and mask processing of the two image layers, and abnormal gray value values are eliminated so as to prevent influences on the result of forest type recognition in the later period.

The method comprises the steps of preprocessing, correcting and fusing a series of measures on an image map of a wax forest field in a research area, then cutting the image map to obtain a multispectral image map and a panchromatic spectral image map, distinguishing the multispectral image map of coniferous forests, broadleaf forests, coniferous mixed forests, broadleaf mixed forests and coniferous mixed forests from the multispectral image map in the research area, and obtaining the multispectral image map of the coniferous forests by using a superposition and masking method. In ENVI, the most suitable vegetation index is selected by utilizing the spectral values of 4 wave bands to identify the forest type and carry out program operation.

As shown in fig. 11, the data map obtained after calculation of the conifer vegetation index shows that the vegetation index range value of the conifer is between 0 and 0.2, and the number of points after 0.2 is rarely distributed. It can be seen from the figure that the vegetation index fluctuates greatly at each numerical point, and the variation range of the point number is small. The influence of natural factors is eliminated, and the plaque of the conifer forest can be preliminarily judged through image processing.

The data graph obtained after calculating the broadleaf forest vegetation index is shown in fig. 12, and it can be known that the vegetation index range value of the broadleaf forest is between-0.2 and 0.3, and the point number distribution after 0.3 is very small. NDVI is more than or equal to-1 and less than or equal to 1, the place where the vegetation is sparse is not a negative number, and the NDVI is the negative number because the surface is covered by clouds or snow, and the spectral curve of the NDVI is inconsistent with that of the area with the vegetation. 0 represents rock or bare earth, etc., and NIR and R are approximately equal; a positive value, indicating vegetation coverage, and increasing with increasing coverage; therefore, the vegetation index range of broad-leaved forest is between 0 and 0.3 when the periphery is removed with some interference points and the number of points is the largest. The observation of the graph of the vegetation index shows that the fluctuation of the vegetation index is large at each numerical point, and the variation range of the point number is large. Eliminating the influence of natural factors, and preliminarily judging the plaques of the broad leaf forests through image processing.

The data map obtained after calculating the vegetation index of the coniferous commingled forest is shown in fig. 13, and it can be known that the vegetation index range value of the coniferous commingled forest is between-0.4 and 0.6. NDVI is more than or equal to-1 and less than or equal to 1, the place where the vegetation is sparse is not a negative number, and the NDVI is the negative number because the surface is covered by clouds or snow, and the spectral curve of the NDVI is inconsistent with that of the area with the vegetation. 0 represents rock or bare earth, etc., and NIR and R are approximately equal; a positive value, indicating vegetation coverage, and increasing with increasing coverage; therefore, the place where the interference points are removed from the periphery and the number of the points is the maximum is the vegetation index range of the coniferous mixed forest. The vegetation index occurs most at points between 0 and 0.161811, and thus it is seen that the main grey values of the pin-leaved commingled forest are concentrated between 0.099213 and 0.161811.

The data obtained after calculating the vegetation index of the broadleaf mixed forest is shown in figure 14, and the vegetation index range value of the broadleaf mixed forest is between-0.4 and 0.365385. NDVI is more than or equal to-1 and less than or equal to 0, the place where the vegetation is sparse is not a negative number, and the NDVI is the negative number because the surface is covered by clouds or snow, and the spectral curve of the NDVI is inconsistent with that of the area with the vegetation. 0 represents rock or bare earth, etc., and NIR and R are approximately equal; a positive value, indicating vegetation coverage, and increasing with increasing coverage; the curve represents the vegetation index range of the broad-leaved mixed forest where the number of points is the greatest. The vegetation index is the largest in number of points between 0 and 0.69658, and it is also seen that the main grey values of broadleaf commingled forests are concentrated between 0.020370 and 0.069658.

The data plot obtained after calculating the vegetation index for the pin-wide commingled forest is shown in fig. 15, and it can be determined that the vegetation index for the pin-wide commingled forest ranges from-0.326923 to 0.653846. NDVI is less than or equal to-1 and less than or equal to 1, and the NDVI is a negative number because the surface is covered by clouds or snow, and the spectral curve of the NDVI is inconsistent with that of a vegetation area. The curve represents the place with the most points, namely the vegetation index range of the broad commingled forest. The areas with the highest number of points appearing between 0 and 0.653846 for vegetation indices are between 0 and 0.102163, and it is thus also seen that the main grey values for pin-wide commingled forests are concentrated between 0.020370 and 0.069658.

The data plot obtained after calculating the non-forestry vegetation index is shown in fig. 16, and it can be determined that the vegetation index range value of the non-forestry vegetation index is between-0.388091 and 0.679558. NDVI is more than or equal to-1 and less than or equal to 0, and the NDVI is a negative number because the surface is covered by clouds or snow, and the spectral curve of the NDVI is inconsistent with that of a vegetation area. 0 represents rock or bare earth, etc., and NIR and R are approximately equal; a positive value, indicating vegetation coverage, and increasing with increasing coverage; the vegetation index in the non-forest land is positive, which indicates that the vegetation is present in the non-forest land, and the vegetation has the vegetation index value, so the value of the non-forest land is positive in the calculation of the vegetation index.

Forest type identification is an important content of forest resource management and monitoring. Forest resource investigation requires a great deal of manpower, material resources and financial resources. The forest type is divided according to different classification methods, classification systems and technical standards, so that different application requirements can be met. The invention provides forest type identification by utilizing gray values and vegetation indexes. Randomly extracting sample points, carrying out classification statistics on the gray values and the vegetation index values, and analyzing the characteristics of the samples to classify coniferous forests, broadleaf forests, coniferous mixed forests, broadleaf mixed forests and coniferous mixed forests.

And according to the randomly extracted sample points, counting by adopting a sample point selection method aiming at the research area, and establishing a relevant forest type identification function model according to the gray value and the vegetation index characteristics. There is a correlation between the vegetation index and the grey value. Because the gray scale and the multispectral have rich image texture characteristics, the comprehensive change information of the gray scale of the high-resolution image in the direction, the distance and the amplitude is reflected, and the forest type can be more accurately identified by utilizing the gray scale value and the multispectral information of each wave band.

Establishment of conifer forest function model

Coniferous pure forest (the proportion of single coniferous tree species is more than or equal to 90%) and coniferous relatively pure forest (the proportion of single coniferous tree species is 65% -90%) are classified as coniferous forest. According to the value of the randomly selected forest sample points, a random sampling method is adopted, abnormal values are eliminated firstly by collecting the gray value and the vegetation index value of the conifer, and then the relation between the curve matching trend of the gray function of the conifer and the distribution points is analyzed. The gray function relation of conifer forest is established as shown in FIG. 17. The method specifically comprises the following steps:

y＝-1E+06x⁵+3E+06x⁴-2E+06x³+594190x²-90066x+5093.8

R²＝0.9452

wherein x represents the gray value of the conifer forest, y represents the frequency of the corresponding gray value in the corresponding conifer mixed forest, R²The accuracy was shown to be 94.52% with the conifer gray scale profile showing a substantially symmetrical distribution with the peak at the center.

The relation of the conifer vegetation index distribution function is shown in fig. 18. The method specifically comprises the following steps:

y＝-1E+06x⁶+1E+06x⁵-436981x⁴-92565x³+58634x²-6449x+174.47

R²＝0.8797

wherein x represents the vegetation index of the conifer forest, y represents the frequency of emergence of the vegetation index value in the corresponding conifer forest, R²Expressed in accuracy, its value was 87.97%

The relation of the conifer forest type identification function is shown in fig. 19. The method specifically comprises the following steps:

y＝0.9061x+0.109

R²＝0.9677

wherein x represents the vegetation index of the conifer forest, y represents the gray value of the coordinate point corresponding to the vegetation index, and R²Indicating an accuracy, the value was 96.77%. X is in [0,0.4 ]]Values are taken over the interval.

Establishment of broad-leaved forest function model

Broad-leaved pure forest (the proportion of single broad-leaved tree species is more than or equal to 90%) and broad-leaved relative pure forest (the proportion of single broad-leaved tree species is 65% -90%) are classified as broad-leaved forest. A hardwood forest gray function relation as shown in fig. 20 was established. The method specifically comprises the following steps:

y＝24833x³-22578x²+6215.4x-460.4

R²＝0.8543

wherein x represents the gray value of the broad-leaved forest, y represents the frequency of occurrence of the corresponding gray value in the corresponding broad-leaved forest, and R²Indicating an accuracy, the value was 85.43%.

Establishing an index distribution function relation of broadleaf forest vegetation as shown in figure 21, which specifically comprises the following steps:

y＝29308x³-23113x²+4983.4x-156.88

R²＝0.9731

wherein x represents the vegetation index of the hardwood forest and y represents the frequency of occurrence of the corresponding vegetation index in the corresponding hardwood forest. R²Indicating an accuracy of 97.31%.

The establishment of the broadleaf forest type identification function relation shown in fig. 22 specifically comprises the following steps:

y＝32.289x³-22.443x²+5.9118x-0.3852

R²＝0.983

wherein x represents the gray value of the broad-leaved forest, y represents the index of the vegetation corresponding to the broad-leaved forest, and R²Indicating an accuracy, the value was 98.3%. x is in [0.25, 0.45 ]]Values are taken over the interval.

Establishment of conifer blending forest function model

According to the numerical value of the randomly selected forest sample points, a random sampling method is adopted to obtain the gray value and the vegetation index of the pin leaf mixed forest, the abnormal value is eliminated, and then the relation among the gray function curve, the matching trend and the distribution point number of the pin leaf mixed forest is analyzed. FIG. 23 shows a relationship of the gray-scale function of the conifer forest.

The coniferous commingled forest is defined to be composed of two or more coniferous tree species, and the proportion of the coniferous tree species is more than or equal to 65 percent.

The gray function relation of the coniferous commingled forest is specifically as follows:

y＝1E+07x⁶-1E+07x⁵+4E+06x⁴-569147x³+45782x²-1646.3x+22.765

R²＝0.8525

wherein x represents the gray value of the blending forest of coniferous needles, and y represents the occurrence of the gray value corresponding to the blending forest of coniferous needlesFrequency of R²Representing an accuracy, the value was 85.25%. x is in [0,0.3 ]]Values are taken in the interval, the curve fluctuation is large, the frequency distribution between 0.15 and 0.2 is concentrated, the distribution trend of a sextic equation regression model is met, and the distribution is in parabolic asymmetric distribution.

The statistical analysis is carried out according to the points of the statistical vegetation indexes, the image texture collected by the vegetation indexes is clearer, the texture is reflected by the spectral characteristics of a plurality of optical bands, and the correlative relation formula of the vegetation indexes of the coniferous mixed forest obtained by utilizing the multispectral characteristics is shown in figure 24 and specifically comprises the following steps:

y＝-435205x⁶+423958x⁵-150950x⁴+23835x³-1607.5x²-4.5084x+14.742

R²＝0.8913

wherein x represents the vegetation index of the coniferous commingled forest, y represents the frequency of the emergence of the vegetation index value in the corresponding coniferous commingled forest, R²Indicating an accuracy of 89.13%. x is in [0, 0.35 ]]Values are taken over the interval. And (3) when the vegetation index of the coniferous mixed forest is processed, firstly removing abnormal values, sequencing the data, and obtaining a vegetation index function model with a trend close to a fitting curve of the sextic function, wherein the curve distribution is reduced along with the increasing frequency of the vegetation index value, and the distribution points are fewer.

By fitting and comparing the vegetation index value and the gray value, and performing data processing and correlation analysis on the two function relations of the gray value and the vegetation index, a function model of the coniferous mixed forest in the forest type identification about two influence factors is obtained, namely, the coniferous mixed forest type identification function relation shown in the figure 25 can be obtained by utilizing a high-score second image according to the gray value and vegetation index relation in the forest type identification, and the specific formula is as follows:

y＝0.0548ln(x)+0.2879

R²＝0.9692

wherein x represents the vegetation index of the pin-leaved mixed forest, y represents the gray value of the coordinate point corresponding to the vegetation index, and R²Indicating an accuracy of 96.92%. x is in [0,0.4 ]]Values are taken over the interval.

Establishment of broad leaf mixed forest function model

Randomly selecting broad-leaved forest data points in forest sample points, also adopting a random sampling method, respectively collecting broad-leaved forest gray values and vegetation index values, respectively establishing a functional relation between the broad-leaved forest vegetation index values and the gray values, removing abnormal values by fitting a function model of the gray values and the vegetation index, and then fitting the vegetation index and the gray of the broad-leaved forest.

The broad-leaved mixed forest is defined to be composed of two or more broad-leaved tree species, and the proportion of the broad-leaved tree species is more than or equal to 65 percent. The broadleaf-mixed forest gray function relation is shown in FIG. 26. The method specifically comprises the following steps:

y＝136276x⁴-73272x³+13045x²-954.18x+41.049

R²＝0.8565

wherein x represents the gray value of the broad-leaved mixed forest, y represents the frequency of the corresponding gray value in the corresponding broad-leaved mixed forest, R²Indicating an accuracy of 85.65%. x is in [0,0.3 ]]And (3) taking values in an interval, wherein the curve fluctuation ratio is gentle, the frequency distribution between 0 and 0.15 is concentrated, and the distribution trend of linear descending is in line after 0.15, and conforms to the distribution trend of a quadratic equation regression line model.

The relation of the vegetation index function of the broad-leaved mixed forest is shown in figure 27, and specifically comprises the following steps:

y＝-21127x⁴+19534x³-5664.7x²+501.95x+6.1831

R²＝0.8766

wherein x represents the vegetation index of the broad-leaved mixed forest, y represents the frequency value of the vegetation index value in the corresponding broad-leaved mixed forest, R²Indicating an accuracy, the value was 87.66%. x is in [0,0.3 ]]Values are taken over the interval. When the vegetation index data of the broad-leaved mixed forest is processed, the abnormal values are eliminated, data sorting is carried out, the trend of the obtained vegetation index function model is close to the fitting curve of the sextic function, the curve distribution is mainly concentrated between 0 and 0.1, and the distribution straight line is reduced from 0.1.

Establishing a broad-leaved mixed forest zoning model, performing fitting comparison on the two function relations of the vegetation index and the gray value, and performing data processing and correlation analysis on the two function relations of the gray value and the vegetation index. A functional model relation shown in fig. 28 is obtained, specifically:

y＝22.963x³-7.2359x²+1.6298x-0.0357

R²＝0.996

wherein x represents the gray level of broad-leaved forest, y represents the vegetation index value of specific coordinate point corresponding to the gray level, and R²Indicating an accuracy of 99.6%. x is in [0,0.3 ]]Values are taken over the interval.

Establishment of broad mixed forest function model

And (3) adopting a random sampling method, eliminating abnormal values by collecting the gray value and the vegetation index value of the pin-wide mixed forest, and then analyzing the relation between the curve matching trend and the distribution point of the gray function of the pin-wide mixed forest. The needle and broad leaf mixed forest is defined as the needle leaf tree species or the broad leaf tree species ratio is 35-65%. Establishing a gray function relation of the pin-wide mixed forest as shown in fig. 29, specifically:

y＝-2E+08x⁶+1E+08x⁵-3E+07x⁴+4E+06x³-261707x²+7332.6x-70.938

R²＝0.9753

wherein x represents the gray value of the pin-wide commingled forest, y represents the frequency of occurrence of the corresponding gray value in the pin-wide commingled forest, R²Indicating an accuracy of 97.53%. x is in [0,0.25 ]]Values are taken in the interval, the frequency distribution of the curve fluctuation between 0.1 and 0.15 is concentrated, the distribution trend of linear descending between 0.05 and 0.1 and between 0.15 and 0.25 is in accordance with the distribution trend of a sextic equation curve model, the curve is almost close to symmetrical distribution, the curve is relatively gentle, and the ascending and descending frequency is very stable.

And (3) carrying out statistical analysis on the data of vegetation indexes of the broad mixed forest, wherein the image texture acquired by adopting multispectral texture characteristics is clearer, and the distribution interval of the mixed forest is reflected by the spectral characteristics of a plurality of optical bands. Establishing a vegetation index functional relation of the pin-wide mixed forest as shown in fig. 30, specifically:

y＝2769.2x³-1484.2x²+164.34x+9.4291

R²＝0.8831

wherein x represents the vegetation index of the mixed forest, y represents the frequency value of the vegetation index in the mixed forest, R²Indicating an accuracy of 88.31%. x is in [0,0.4 ]]Values are taken over the interval. The vegetation index model obtained by the method of random sampling in the data processing process of the vegetation index of the broad-mixed forest has a trend close to the equation of the cubic function. The curve distribution is mainly concentrated between 0.05 and 0.1, the distribution straight line descends from 0.1, and the distribution frequency shows a rising trend at 0.3.

Fitting and comparing the two function relations of the vegetation index and the gray value, and performing data processing and correlation analysis on the two function relations of the gray value and the vegetation index of the mixed forest to obtain a forest zoning function relation for the broad mixed forest as shown in fig. 31, wherein the method specifically comprises the following steps:

y＝10.197x²-0.5278x+0.0077

R²＝0.9901

wherein x represents the gray scale of the pin-wide mixed forest, y represents the gray scale value of the coordinate point corresponding to the gray scale of the pin-wide mixed forest, R²Indicating an accuracy, the value was 99.01%. x is in [0,0.25 ]]Values are taken over the interval.

Establishment of non-woodland function model

The invention establishes a non-forest land gray scale function relation as shown in fig. 32, which specifically comprises the following steps:

y＝694565x⁶-1E+0.6x⁵+895921x⁴-303566x³+49796x²-3485.4x+81.603

R²＝0.8215

wherein x represents the gray scale value of the non-forest land, y represents the frequency of the gray scale value in the non-forest land, and R²Indicating an accuracy, the value was 82.15%. x is in [0, 0.6 ]]The values are taken in the interval, the frequency distribution of the curve fluctuation between 0.15 and 0.2 is concentrated, the front section is symmetrically distributed, the gray level changes greatly in the rear section, the distribution trend of the sextic equation curve model is met,almost close to a symmetrical distribution.

And carrying out statistical analysis on the data of the non-forest land vegetation index, acquiring the image of the non-forest land by adopting multispectral texture characteristics, wherein the texture is clearer, and the spectral characteristics of a plurality of optical wave bands reflect the distribution interval of the non-forest land. As shown in fig. 33, is an index profile of non-forestry vegetation.

Establishing a non-forest land forest type identification function relation shown in fig. 34, specifically:

y＝0.1971e^5.633x

R²＝0.9261

wherein x represents the vegetation index of the non-forest land, y represents the gray value of the coordinate point corresponding to the vegetation index of the non-forest land, R²Indicating an accuracy of 92.61%. x is [ -0.25,0.25 [)]Values are taken over the interval.

The invention mainly relates to establishing function models according to characteristic research of gray scale and vegetation index, and the forests can be classified into coniferous forests, broadleaf forests, coniferous mixed forests, broadleaf mixed forests and coniferous mixed forests through the mathematical models. And selecting experimental sample points to operate in the function model, wherein the experimental result accords with which type of model and has the highest precision, and the experimental result belongs to which type of the five major types. The forest type can be identified according to the fitting function model obtained through experiments, and the accuracy averagely reaches about 85%. Firstly, randomly selecting 1000 experimental data points in each divided area, excluding abnormal values, extracting a gray value and a vegetation index value corresponding to each point, counting the distribution frequency of the vegetation index and the gray value, establishing the relationship among the gray, the vegetation index and the frequency, and finally obtaining an ideal function model. The method comprises the steps of establishing a function model of gray scale and gray scale frequency, wherein the significance of the function model of the vegetation index and the vegetation index frequency lies in that the peak value of the gray scale and the vegetation index and the trend of a function curve in a certain interval range can be used for judging the forest type, the vegetation indexes and the gray scale values of different tree species are different, and the vegetation index values obtained by different tree species in different light wave intervals are different, so that the establishment of a fitting function model between the gray scale and the vegetation index and the frequency is used for identifying the forest type. Randomly selecting 200 hectares of forest lands from high-resolution image data to perform experiments, respectively extracting gray scales and vegetation indexes from 1000 experimental points, verifying by using the function model obtained by the method, verifying the gray scales and the vegetation indexes of the same point in 6 fitting function models, determining which forest type can be determined by which fitting function model is closer, and determining the accuracy of the experiments only when the 1000-point inspection precision is higher than 85%.

In the experimental operation process, the abnormal points are removed, and the parts with the NDVI smaller than 0 in the vegetation index selection are removed. The experimental operation inspection precision is high, and a new method is provided for forest resource classification and forest resource monitoring.

Forest type identification and evaluation

In order to verify the experimental result, a flourishing forest farm and a watershed ditch forest farm of the overlapped forestry bureau are selected as the inspection areas, the flourishing forest farm and the watershed ditch forest farm are located in the northeast of the overlapped forestry bureau, and the error of the experimental result caused by the difference of regional factors is eliminated. Firstly, image preprocessing (atmospheric correction, orthorectification, radiation correction, mosaic, fusion and cutting) is carried out on an inspection area, then a panchromatic spectral image and a multispectral image are extracted, then ninety blocks are randomly selected to serve as experimental samples, and the extraction of gray scales and vegetation indexes is carried out on 90 patches respectively. The gray scale range of 30 plaques and the gray scale value of a specific corresponding point can be directly counted by the ARCGIS operation. And importing the multispectral image into ENVI, and calculating the detected patch vegetation index interval and the vegetation index of the corresponding point by utilizing the normalized vegetation index. As shown in FIG. 35, the test areas are the Wangchai forest land and the Poisson ditch forest land.

The interval range of the numerical distribution of 90 plaques and the trend of the fitted model are compared by extracting the gray scale and the vegetation index of the plaques. Image analysis of partial plaques were selected, which are representative of 90 plaques, and which are representative of 90 plaques and reflect the results of the experiment.

FIG. 36 is a gray distribution graph of selected patches, which is examined as conifer forest in combination with the previously established model, the gray value range is 0.161198 to 0.697568, the point distribution of the main gray concentration is 0.295291 to 0.386263, and the gray distribution trend of the conifer forest gray function model is met.

FIG. 37 is a graph showing the gray distribution of selected patches, which is tested as broadleaf forest in combination with the previously established model, the gray value range is 0.119767-0.458825, the point distribution of the main gray concentration is 0.204531-0.292820, and the trend of the gray distribution of the broadleaf forest gray model is met.

Fig. 38 is a gray distribution diagram of the patch, which is examined as a pin-wide mixed forest by combining the previously established model, and the gray distribution frequency of the patch is observed, and the distribution curve of the gray is found to be matched with the trend of the gray model of the pin-wide mixed forest, so that the pin-wide mixed forest is determined.

FIG. 39 is a distribution diagram of the vegetation index of the plaque, which is examined as conifer by combining the previously established model, the vegetation index ranges from 0.032738 to 0.563758, the main distribution ranges from 0.210878 to 0.295575, and the distribution trend of the conifer vegetation index function model is met.

FIG. 40 is a distribution diagram of the vegetation index of the plaque, which is proved to be broad-leaved forest by combining the previously established model, the vegetation index ranges from 0.003684 to 0.631579, the vegetation index is mainly distributed between 0.210878 and 0.295575, and the distribution trend of the vegetation index function model of the broad-leaved forest is met.

FIG. 41 is a distribution diagram of the vegetation index of the plaque, which is tested as a blend forest with a needle and broad in combination with the previously established model, the vegetation index ranges from 0010989 to 0.841176, the vegetation index is mainly distributed between 0.152778 and 0.260365, and the distribution trend of the vegetation index function model of the blend forest with needle and broad is met.

Forest type recognition result and precision evaluation

And (3) carrying out inspection and verification by adopting typical representatives in 30 plaques in the inspection area, judging that the verification result approximately accords with the experimental result according to the function model distribution curve, and almost agreeing with the second-class survey data of forest resources. In order to further verify the accuracy of the experimental result, experimental sample points are randomly selected in different longitudes and latitudes of the inspection area respectively, and further verification is performed by combining a function model.

FIG. 42 is a plot of experimental spot locations. Selecting a Wangchan forest field, a Poison ditch forest field and a Luoda forest field in a research area, randomly selecting sample points in an inspection area, respectively taking each sample point as an inspection patch, selecting 30 points in each sample, respectively extracting the gray scale and vegetation index of the point selected by each inspection position, analyzing the curve trend for preliminary judgment, then calculating by using a forest type identification function, and identifying the forest type by using a function model with the highest calculation result precision. The inspection results show that white represents non-forest land, green represents broadleaf mixed forest, yellow represents coniferous mixed forest, red represents mixed forest, light green represents coniferous forest and blue represents broadleaf forest. The result shows that the patches approximately accord with the functional relation, the forest type can be accurately identified, and the individual patches have deviation in the calculation functional relation. And (3) error analysis: the error may be caused by an abnormal data value during point selection, which is not removed completely, or by different brightness of each pixel point during high-resolution image shooting during gray scale extraction, or by mutual interference of spectra between plants, which may cause some errors during vegetation index extraction.

Fig. 43 is a diagram of inspection results of randomly selected sample areas, in order to further verify the feasibility of the forest type identification function model, a forest area is randomly selected, the patch verification result is broadleaf forest, the area is used for calculating the function model, the gray value difference is large due to the sun exposure and the sun exposure, the gray value and the vegetation index value are calculated in a segmented manner, the inspection result is shown in fig. 44, the brown part is other, the black area is an abnormal value, in addition, the two areas with obvious comparison are respectively the sun exposure area and the sun exposure area of the broadleaf forest, both accord with the calculation result of the broadleaf forest function model, and the selected point is found to be the broadleaf forest type by checking the second-class investigation data of forest resources.

FIG. 45 is a graph comparing the results of the two types of tests. The forest farm inspection area results are shown in fig. 40: according to the operation method of the invention, the forest type identification is carried out in the selected forest field inspection area by adopting a random point selection method. During verification, 80 inspection sample points are selected, the gray level and the vegetation index of each sample point are respectively extracted, the final result is divided into 6 forest stand types such as coniferous forest, broadleaf forest, coniferous mixed forest, broadleaf mixed forest, coniferous mixed forest, non-forest land and the like, calculation and model inspection are carried out on the selected sample areas, and if the calculated result is close to the forest type of the same function model, the calculated result is automatically classified into the corresponding forest type. And performing forest type identification and classification in the ARCGIS, and finding that the identification result is approximately consistent with the second-class survey data of forest resources, and the abnormality occurs in part of sample points. Aiming at the purpose of forest type identification, the invention proposes to take more points at the edge when the gray scale and the vegetation index are counted when the check point is selected, gradually enlarges the verification area, repeatedly operates, starts to mark the boundary line of the forest type if the change result appears, determines the position of the dividing patch boundary line, and finally determines the boundary line of the forest type through the function one time of test operation.

In order to further verify the precision of the experimental result, the invention finally combines the second-class survey data of the forest resources with the experimental result for analysis and comparison, the second-class survey data of the forest resources substantially accord with the experimental result, 72 sample monitoring points in 80 sampling points are completely matched, and the precision reaches 90%.

The forest type recognition function model provided by the invention can be accurately used for recognizing forest types. Thus, these models can be applied to the identification of forest types.

Aiming at the difficulty in dividing forest lands in the current forest resource investigation, the invention considers that the specific field investigation data can be obtained only by spending a large amount of manpower, material resources and financial resources in the field investigation monitoring process, and the obtained forest type data is influenced by a plurality of factors, so that the forest type identification can not be accurately carried out, and the management and the monitoring of forest resources are not facilitated. The method is characterized in that a high-grade second-image forest classification technology is used as a technical basis, a series of methods such as ARCGIS software, PIE image processing, ENVI vegetation index calculation, statistical samples and the like are combined through data processing and analysis to research a scientific method for recognizing the current forest type, the correlation between the gray value of a research area and the forest vegetation index is analyzed by combining the gray value of the research area and the forest vegetation index, a function model is established, finally, image preprocessing is carried out by using another forest field, the accuracy of an experiment is further demonstrated, and finally, the experiment result is applied to forest type recognition, so that the forest type can be recognized by acquiring the gray value and the vegetation index by using the texture characteristics of the high-grade second-image, the field investigation work of forest resources is reasonably reduced, and the cost and the resources are saved.

The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims

1. A forest type identification method based on high-resolution remote sensing images is characterized by comprising the following steps:

dividing the research area into a forest land and a non-forest land according to a forest land classification standard, dividing the forest land into a coniferous forest, a broadleaf forest and a mixed forest, and obtaining gray values and vegetation index values of the coniferous forest, the broadleaf forest and the mixed forest;

respectively establishing a relevant forest type identification function model according to the correlation between the gray value and the vegetation index value:

(1) establishing a forest type recognition function model of the coniferous commingled forest, namely, during forest type recognition, recognizing the coniferous commingled forest by using the following function model relational expression in a high-grade second image according to the gray level and the vegetation index:

y＝0.0548ln(x)+0.2879

R²＝0.9692

wherein x represents the vegetation index of the pin-leaved mixed forest, y represents the gray value of the coordinate point corresponding to the vegetation index, and R²Indicating an accuracy of 96.92%, x is at [0,0.4 ]]Taking values in the interval;

(2) establishing a forest type recognition function model of the broad-leaved mixed forest, and fitting the vegetation index and the gray level to obtain a function model relation formula:

v＝22.963x³-7.2359x²+1.6298x-0.0357

R²＝0.996

wherein x represents the gray level of the broad-leaved forest, y represents the vegetation index value of the specific coordinate point corresponding to the gray level, and R²Indicating an accuracy of 99.6%, x is [0,0.3 ]]Taking values in the interval;

(3) establishing a forest type recognition function model of the mixed forest, and fitting the vegetation index and the gray level to obtain a function model relation formula:

y＝10.197x²-0.5278x+0.0077

R²＝0.9901

wherein x represents the gray scale of the mixed forest, y represents the gray scale value of the coordinate point corresponding to the gray scale of the mixed forest, R²Indicating an accuracy of 99.01%, x is [0,0.4 ]]Taking values in the interval;

(4) establishing a forest type identification function model of the non-forest land, counting data of the vegetation index of the non-forest land for statistical analysis, acquiring an image of the non-forest land by adopting multispectral texture characteristics, wherein the correlation relation is as follows:

v＝0.1971e^5.633x

R²＝0.9261

wherein the formula x represents the vegetation index of the non-forest land, y represents the gray value of the coordinate point corresponding to the vegetation index of the non-forest land, R²Expressed accuracy, its value was 92.61%, x was [ -0.25,0.25 [ ]]Taking values in the interval;

(5) establishing a forest type recognition function model of the conifer forest, wherein the correlation relation is as follows:

y＝0.9061x+0.109

R²＝0.9677

wherein the formula x represents the vegetation index of the conifer forest, y represents the gray value of the coordinate point corresponding to the vegetation index, R²Indicating an accuracy of 96.77%, x is at [0,0.4 ]]Taking values in the interval;

(6) establishing a forest type recognition function model of the broad-leaved forest, wherein the correlation relation is as follows:

y＝32.289x³-22.443x²+5.9118x-0.3852

R²＝0.983

wherein the formula x represents the gray value of the broad-leaved forest, y represents the corresponding vegetation index in the corresponding broad-leaved forest, R²Indicating an accuracy of 98.3%, x is [0.25, 0.45 ]]On the interval;

checking and evaluating the correlation function model;

and applying the correlation function model to forest type identification.

2. The method as claimed in claim 1, wherein the checking and evaluating of the correlation function model comprises checking and verifying by using a representative in thirty small shifts in a checking area, judging a verification result according to a distribution curve of the function model, and comparing with forest resource secondary survey data;