CN113920262A - Mining area FVC calculation method and system for enhancing edge sampling and improving Unet model - Google Patents
Mining area FVC calculation method and system for enhancing edge sampling and improving Unet model Download PDFInfo
- Publication number
- CN113920262A CN113920262A CN202111201530.9A CN202111201530A CN113920262A CN 113920262 A CN113920262 A CN 113920262A CN 202111201530 A CN202111201530 A CN 202111201530A CN 113920262 A CN113920262 A CN 113920262A
- Authority
- CN
- China
- Prior art keywords
- image
- area
- model
- sampling
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000005070 sampling Methods 0.000 title claims abstract description 134
- 238000005065 mining Methods 0.000 title claims abstract description 81
- 238000004364 calculation method Methods 0.000 title claims abstract description 28
- 230000002708 enhancing effect Effects 0.000 title claims abstract description 13
- 238000012549 training Methods 0.000 claims abstract description 84
- 238000003062 neural network model Methods 0.000 claims abstract description 65
- 238000000034 method Methods 0.000 claims abstract description 63
- 238000012795 verification Methods 0.000 claims description 60
- 238000011160 research Methods 0.000 claims description 56
- 238000012545 processing Methods 0.000 claims description 31
- 238000011176 pooling Methods 0.000 claims description 19
- 238000005520 cutting process Methods 0.000 claims description 15
- 238000011156 evaluation Methods 0.000 claims description 13
- 238000006243 chemical reaction Methods 0.000 claims description 8
- 238000013075 data extraction Methods 0.000 claims description 6
- 238000013210 evaluation model Methods 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 238000007670 refining Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 abstract description 3
- 238000011161 development Methods 0.000 abstract description 2
- 239000002689 soil Substances 0.000 abstract description 2
- 238000012360 testing method Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 238000005259 measurement Methods 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000009826 distribution Methods 0.000 description 5
- 238000000605 extraction Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- 239000011800 void material Substances 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010422 painting Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000005728 strengthening Methods 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
- 238000009827 uniform distribution Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/05—Geographic models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Forestry; Mining
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Computer Hardware Design (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Business, Economics & Management (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Agronomy & Crop Science (AREA)
- Animal Husbandry (AREA)
- Marine Sciences & Fisheries (AREA)
- Mining & Mineral Resources (AREA)
- Mathematical Analysis (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
Abstract
The invention discloses a mining area FVC calculation method for enhancing edge sampling and improving a Unet model, comprising the following steps of A, acquiring ground vegetation parameter data in a mining area scene; B. constructing a sample data set based on an expanded selected area cross overlapping method; C. constructing and training an improved Unet neural network model; D. and calculating the vegetation coverage of the mining area by utilizing the improved Unet neural network model. The method comprises the steps of firstly acquiring remote sensing data and ground soil vegetation coverage through unmanned aerial vehicle flight, laying ground control points to construct a vegetation related data acquisition system, providing a data base for vegetation coverage calculation, then segmenting training sample data by using an expanded selective area alternate overlapping sampling method to extract and construct a vegetation coverage sample database, finally performing model training and constructing a vegetation coverage network relation model by using an improved Unet network model, further accurately deducing vegetation coverage centimeter-level information data, and providing powerful data support for mining area ecological environment monitoring management and mining area mining development plans.
Description
Technical Field
The invention relates to the mining field, the artificial intelligence field, the ecology field, the remote sensing field and the geographic information field, in particular to a mining area FVC calculation method for enhancing edge sampling and improving a Unet model.
Background
Mining in mining area seriously influences the growth of vegetation, and timely accurate detection mining area vegetation growing situation utilizes vegetation coverage data to study mining area vegetation, can make clear and determine the distribution and the growing situation of mining area ecosystem vegetation, provides data basis for mining area follow-up management exploitation. The existing methods for acquiring vegetation coverage by remote sensing monitoring can be mainly divided into the following three types: firstly, a statistical regression model: the vegetation coverage of a wider area is deduced by establishing linearity (Graetz 1988, Peter 2002) and a nonlinear model (Dymond 1992) of the vegetation coverage actually measured on the earth surface and remote sensing information, but the model is only suitable for a specific area and a specific vegetation type and is not suitable for complex scenes such as mining areas and the like with strong spatial heterogeneity and mixed vegetation types; II, pixel binary model (Qi J2000, Cao 2019): the model simply assumes that the pixel of the remote sensing image is only composed of vegetation covered earth surface and non-vegetation covered earth surface, and in the aspect of data, the spatial resolution is not equal to 30m-8km based on the traditional satellite remote sensing data, but under the condition of equal spatial resolution, the vegetation earth surface coverage of the mining area has extremely strong spatial heterogeneity and the simple assumption condition of the pixel binary model has contradiction; third, traditional machine learning models (Boyd 2002, zhao jian 361911; however, the vegetation generally at the edge of the image has poor calculation accuracy and is limited by calculation efficiency. Therefore, the existing vegetation coverage calculation technology is difficult to be generally applied in mining area scenes.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a mining area FVC calculation method for enhancing edge sampling and improving a Unet model, which utilizes an unmanned aerial vehicle image to construct a high-resolution mining area vegetation image sample data set and solves the problems of strong surface heterogeneity and sparse vegetation distribution and difficult recognition in a mining area scene; high-precision digital elevation model data are introduced, topographic feature information under a mining area scene is supplemented, multidimensional features of sample data are perfected, and a method and a theoretical basis are provided for constructing a mine vegetation sample data set.
The purpose of the invention is realized by the following technical scheme:
a mining area FVC calculation method for enhancing edge sampling and improving a Unet model comprises the following steps:
A. collecting ground vegetation parameter data in a mining area scene: determining a mining area research area, carrying out overlapped orthographic aerial photography on the mining area research area by adopting an unmanned aerial vehicle according to drawing precision to obtain an orthographic image set, wherein the lowest aerial photography overlapping proportion is 32-38%, then carrying out image splicing to obtain an orthographic image of the research area, then establishing an air-to-three calculation model (SFM) and generating dense point cloud data, and obtaining high-precision digital geographic elevation model data through conversion;
B. constructing a sample data set based on an enlarged selective area cross overlapping method, wherein the method comprises the following steps:
b1, dividing the orthographic image set of the research area into a training set and a verification set, respectively carrying out data annotation on the training set and the verification set by adopting a labelimg tool, marking a vegetation sample in the image, wherein the annotation result is a json file, and then converting the json file into an annotated image file with the same format as that of the orthographic image, so as to obtain an annotated training set and an annotated verification set;
b2, simultaneously, cross-overlapping and sampling the orthoimage in the training set and the labeled image file labeled with the training set, wherein the moving window of the sample sampling model is as follows:
W=2n(ii) a Wherein W is the image side length of a single sample data, and n is a positive integer;
establishment 2n×2nThe moving window carries out global traversal on the orthoimages in the training set and the marked image files marked with the training set according to the step length L, wherein the step length L is less than 2nCutting according to the size of the moving window in each traversal to obtain a sample data image block, and completing the traversal to obtain a training input data set;
b3, performing cross overlapping sampling and traversing cutting on the orthoimages in the verification set and the marked image files of the marked verification set according to the method B2 to obtain a verification input data set;
b4, performing data enhancement processing on the training input data set and the verification input data set in sequence by adopting rotation, mirror image, color transformation, contrast variation and Gaussian noise based on the python platform, wherein the training input data set and the verification input data set after the enhancement processing form an input database together;
C. constructing an improved Unet neural network model, wherein the improved Unet neural network model comprises a down-sampling system and an up-sampling system, the down-sampling system comprises a plurality of down-sampling modules, the up-sampling system comprises a plurality of up-sampling modules, and the down-sampling module is composed of a module structure of cavity convolution support constructed by two cavity convolution layers and a maximum pooling layer;
c1, taking the training input data set after the enhancement processing in the step B and the high-precision digital geographic elevation model data in the step A as an input layer of the improved Unet neural network model, training the improved Unet neural network model, extracting the characteristics of the hole convolution layer of the improved Unet neural network model by using hole convolution, and finally obtaining the characteristic images after the hole convolution as follows:wherein p (i) represents the feature value extracted at position i, k (f) is the parameter value of the convolution kernel at position f, r is the hole convolution rate, and x (i + r f) is the image value of the corresponding receptive field position;
then, pooling the characteristic images according to the following formula and further refining the characteristic points: p is a radical ofm1Max-pool (p); wherein P ism1The characteristic image is a characteristic image after one layer of pooling, and p is a characteristic image after cavity convolution;
D. the vegetation coverage of the mining area is calculated by the following method:
d1, loading the high-precision digital geographic elevation model data in the step A or the input database in the step B into a trained improved Unet neural network model for classification and identification processing to obtain a classification result, and splicing the classification result images through an Envi platform to obtain a vegetation classification result of the research area;
d2, constructing a fishing net coverage research area with single grid dimension of m multiplied by n, and counting the area of vegetation coverage under the single grid through an Arcgis platform and recording the area as SvegeThen the vegetation coverage of the research area is determinedObtained by the following formula:
wherein FVC is the vegetation coverage of the study area, SvegeThe area of vegetation coverage under a single grid, k is the number of the grids covered in the study area, SallIs the total area of the ground in the area under study.
In order to better realize the invention, the method for calculating the FVC of the mining area by enhancing the edge sampling and improving the Unet model comprises the following steps:
A. collecting ground vegetation parameter data in a mining area scene: determining a mining area research area, carrying out overlapped orthographic aerial photography on the mining area research area by adopting an unmanned aerial vehicle according to drawing precision to obtain an orthographic image set, wherein the drawing precision is centimeter-level, the lowest aerial photography overlapping proportion is 32-38%, then carrying out image splicing to obtain an orthographic image of the research area, then establishing an air-to-three calculation model (SFM) and generating dense point cloud data, and obtaining high-precision digital geographic elevation model data through conversion;
B. constructing a sample data set based on an enlarged selective area cross overlapping method, wherein the method comprises the following steps:
b1, dividing the orthographic image set of the research area into a training set and a verification set, respectively carrying out data annotation on the training set and the verification set by adopting a labelimg tool, marking a vegetation sample in the image, wherein the annotation result is a json file, and then converting the json file into an annotated image file with the same format as that of the orthographic image, so as to obtain an annotated training set and an annotated verification set;
b2, simultaneously, cross-overlapping and sampling the orthoimage in the training set and the labeled image file labeled with the training set, wherein the moving window of the sample sampling model is as follows:
W=2n(ii) a Wherein W is the image side length of a single sample data, and n is 9;
establishing a 512 x 512 moving window to perform global traversal on the orthoimage in the training set and the labeled image file labeled with the training set according to a step length L, wherein the step length L is 212, cutting is performed on each traversal according to the size of the moving window to obtain a sample data image block A, the slice size of each sample data image block A is 512 x 512, and after the traversal is completed, data in a 300 x 300 pixel range are selected from the central area of the sample data image block A to form a training input data set;
b3, cross-overlapping and sampling the ortho-image in the verification set and the labeled image file in the labeled verification set, wherein the moving window of the sample sampling model is as follows:
W=2n(ii) a Wherein W is the image side length of a single sample data, and n is 9;
establishing a 512 x 512 moving window to perform global traversal on the orthoimage in the verification set and the labeled image file in the labeled verification set according to a step length L, wherein the step length L is 212, cutting is performed on each traversal according to the size of the moving window to obtain a sample data image block B, the slice size of each sample data image block B is 512 x 512, and after the traversal is completed, data in a 300 x 300 pixel range are selected from the central area of the sample data image block B to form a verification input data set;
b4, performing data enhancement processing on the training input data set and the verification input data set in sequence by adopting rotation, mirror image, color transformation, contrast variation and Gaussian noise based on the python platform, wherein the training input data set and the verification input data set after the enhancement processing form an input database together;
C. constructing an improved Unet neural network model, wherein the improved Unet neural network model comprises a down-sampling system and an up-sampling system, the down-sampling system comprises four down-sampling modules, the up-sampling system comprises four up-sampling modules, and the up-sampling module is formed by convolution with a corresponding transposition of the down-sampling modules; the down-sampling module consists of two cavity convolution layers and a module structure of cavity convolution support constructed by a maximum pooling layer;
c1, taking the training input data set after the enhancement processing in the step B and the high-precision digital geographic elevation model data in the step A as an input layer of the improved Unet neural network model, training the improved Unet neural network model, extracting the characteristics of the hole convolution layer of the improved Unet neural network model by using hole convolution, and finally obtaining the characteristic images after the hole convolution as follows:
wherein p (i) represents the feature value extracted at position i, k (f) is the parameter value of the convolution kernel at position f, r is the hole convolution rate, and x (i + r f) is the image value of the corresponding receptive field position;
then, pooling the characteristic images according to the following formula and further refining the characteristic points: p is a radical ofm1Max-pool (p); wherein P ism1The characteristic image is a characteristic image after one layer of pooling, and p is a characteristic image after cavity convolution;
D. the vegetation coverage of the mining area is calculated by the following method:
d1, loading the high-precision digital geographic elevation model data in the step A or the input database in the step B into a trained improved Unet neural network model for classification and identification processing to obtain a classification result, and splicing the classification result images through an Envi platform to obtain a vegetation classification result of the research area;
d2, constructing a fishing net coverage research area with single grid dimension of m multiplied by n, and counting the area of vegetation coverage under the single grid through an Arcgis platform and recording the area as SvegeThen, the vegetation coverage of the study area is obtained by the following formula:
wherein FVC is the vegetation coverage of the study area, SvegeThe area of vegetation coverage under a single grid, k is the number of the grids covered in the study area, SallIs the total area of the ground in the area under study.
Step C of the preferred method of calculating FVC for a mine site of the present invention further comprises C2;
c2, adopting Precision model and/or Recall model and/or F1-score model to evaluate the Precision of the trained improved Unet neural network model by using the verification input data set:
wherein TP is the number of identifying the target sample as true, FP is the number of identifying the target sample as false, FN is the number of identifying the non-target sample as true, Recall is the output value of Recall evaluation model, Precision is the output value of Precision evaluation model, and finally the finally obtained F1-score value is used as the evaluation Precision of the improved Unet neural network model.
Preferably, in step a, the mining area FVC calculation method of the present invention performs flight planning for the unmanned aerial vehicle including takeoff, landing, and navigation routes according to the drawing accuracy of the orthophoto image set, with the minimum overlap ratio of aerial photography of 35%, and the unmanned aerial vehicle's navigation height formula is as follows:
h ═ f · m; h is the navigation height, f is the camera focal length, m is the mapping scale, and the course height and the navigation speed of the unmanned aerial vehicle are kept consistent as much as possible when data acquisition is needed.
Preferably, the mining area FVC calculation method of the present invention further comprises step a 1;
a1, arranging ground control points on the ground in a mining area research area, wherein the ground control points comprise a target and an image control point identifier, an orthographic image set overlapped by the unmanned aerial vehicle and orthographic aerial-photographed will form an identifier control point corresponding to the ground control point, and the orthographic image set depends on the identifier control point to perform image registration and geometric calibration when image splicing and/or space-three solution model SFM processing is performed.
Preferably, in step C1 of the mining area FVC calculation method of the present invention, the void convolution layer of the improved Unet neural network model is cross-multiplied by the convolution kernel of m × m and the pixel corresponding to the convolution kernel receptive field to obtain the feature image after void convolution.
A mining area FVC computing system for enhancing edge sampling and improving a Unet model comprises a ground vegetation parameter data extraction system, a sampling system for expanding a selected area cross overlapping method, an improved Unet neural network model and a mining area vegetation coverage statistical module, wherein the ground vegetation parameter data extraction system comprises an unmanned aerial vehicle aerial photography control unit, an image splicing module and an aerial three-dimensional calculation model SFM; the sampling system of the expanded selective area cross-overlapping method comprises a cross-overlapping sampling module, the sampling system of the expanded selective area cross-overlapping method is provided with a labelimg tool and a python platform, the sampling system of the expanded selective area cross-overlapping method carries out data labeling through the labelimg tool and obtains a labeled image file, the cross-overlapping sampling module carries out overall traversal cutting sampling according to a moving window by adopting a sample sampling model and a step length L and obtains a sample data image block, and the sampling system of the expanded selective area cross-overlapping method carries out data processing through the python platform and obtains an input database; the improved Unet neural network model comprises a down-sampling system and an up-sampling system, and is trained to obtain a trained improved Unet neural network model; the mining area vegetation coverage statistical module classifies, identifies and processes high-precision digital geographic elevation model data or an input database through a trained improved Unet neural network model to obtain a classification result, then splices classification result images through an Envi platform to obtain a vegetation classification result of a research area, and constructs a fishing net coverage research area with a single grid scale of mxn and counts the vegetation coverage of the research area through an Arcgis platform.
Compared with the prior art, the invention has the following advantages and beneficial effects:
(1) the method comprises the steps of firstly acquiring remote sensing data and ground soil vegetation coverage through low-altitude flight of an unmanned aerial vehicle, laying ground control points, constructing an air-ground integrated vegetation related data acquisition system under a mining area scene, providing a data base for vegetation coverage calculation, then designing and utilizing an extended selective area alternate overlapping sampling method, segmenting and extracting training sample data, strengthening image edge characteristics, constructing a vegetation coverage sample database, finally utilizing an improved Unet network model to perform model training and construct a vegetation coverage network relation model, and further accurately deducing centimeter-level information data of vegetation coverage in a research area to obtain distribution and growth conditions of vegetation of the mining area ecosystem, thereby providing powerful data support for mining area ecological environment monitoring management and mining area exploitation development plans.
(2) The method constructs a neural network for improving the Unet, and acquires the receptive field with a larger area than that of the common convolution under the condition that the sample characteristics are the same by combining a cavity convolution structure, so that more vegetation texture characteristic information of a mining area is extracted, the capture and the identification of small targets are accurately realized, compared with the traditional machine learning model, the model has higher inversion accuracy and higher calculation speed, and a method and a theoretical basis are provided for quickly extracting vegetation characteristics of the mining area so as to accurately calculate the vegetation coverage.
(3) According to the method for expanding the alternative overlapping sampling of the selected area, the sample set range is expanded, the sample slices are extracted, and the sample slices are in the area of alternative overlapping, so that the problem that the edge calculation result is not ideal is solved, the condition that data faults exist in the boundary between the samples is improved, the identification precision of a neural network is improved, and the challenge of difficulty in identifying vegetation in a mining area scene is completed.
(3) The invention provides a vegetation coverage calculation model combining unmanned aerial vehicle images and an enhanced edge sampling method based on a neural network of improved Unet in a mining area scene, and a high-resolution mining area vegetation image sample data set is constructed by using the unmanned aerial vehicle images, so that the problems of strong surface heterogeneity and sparse vegetation distribution and difficult recognition in the mining area scene are solved; the sample image acquired by the low-altitude flight sensor is less influenced by atmosphere and cloud cover, so that the precision of the vegetation sample data set in the mining area is further improved; high-precision digital elevation model data are introduced, topographic feature information under a mining area scene is supplemented, multidimensional features of sample data are perfected, and a method and a theoretical basis are provided for constructing a mine vegetation sample data set.
Drawings
FIG. 1 is a schematic layout diagram of ground control points in an embodiment;
FIG. 2 is a diagram illustrating a moving window of a sample sampling model according to an embodiment;
FIG. 3 is a schematic diagram of a modified Unet neural network model in an embodiment;
FIG. 4 is a schematic diagram of the improved Unet neural network model hole convolution in the embodiment;
FIG. 5 is a schematic representation of the pooling of the improved Unet neural network model in an embodiment;
FIG. 6 is a schematic diagram comparing a conventional sample with a sample according to the present invention;
FIG. 7 is a schematic flow chart of the method of the present invention;
FIG. 8 is a diagram illustrating exemplary placement of image control points for ground control points in an embodiment;
FIG. 9 is an orthophoto map of FIG. 8;
fig. 10 is a visualization result diagram of vegetation coverage in the embodiment.
Detailed Description
The present invention will be described in further detail with reference to the following examples:
example one
As shown in fig. 1 to 10, a method for calculating a mining area FVC of an enhanced edge sampling and improved Unet model includes the following steps:
A. collecting ground vegetation parameter data in a mining area scene: determining a mining area research area, carrying out overlapped orthographic aerial photography on the mining area research area by adopting an unmanned aerial vehicle according to drawing precision to obtain an orthographic image set, wherein the lowest aerial photography overlap ratio is 32-38% (designing a flight plan, firstly carrying out on-site survey on the whole research area, determining the terrain of the research area, carrying out simulation planning on take-off, landing and navigation of the unmanned aerial vehicle, then dividing the whole research area into a plurality of flight measurement areas with overlap areas on equipment, determining the area overlap ratio according to the requirement on data precision, preferably 35%), then carrying out image splicing to obtain an orthographic image of the research area, then establishing an air-to-three resolving model SFM and generating dense point cloud data, and obtaining high-precision digital geographic elevation model data through conversion;
B. constructing a sample data set based on an enlarged selective area cross overlapping method, wherein the method comprises the following steps:
b1, dividing the orthographic image set of the research area into a training set and a verification set, respectively carrying out data annotation on the training set and the verification set by adopting a labelimg tool, marking a vegetation sample in the image, wherein the annotation result is a json file, and then converting the json file into an annotated image file with the same format as that of the orthographic image, so as to obtain an annotated training set and an annotated verification set;
b2, simultaneously, cross-overlapping and sampling the orthoimage in the training set and the labeled image file labeled with the training set, wherein the moving window of the sample sampling model is as follows:
W=2n(ii) a Wherein W is the image side length of a single sample data, and n is a positive integer;
establishment 2n×2nThe moving window carries out global traversal on the orthoimages in the training set and the marked image files marked with the training set according to the step length L, wherein the step length L is less than 2nCutting according to the size of the moving window in each traversal to obtain a sample data image block, and completing the traversal to obtain a training input data set;
b3, performing cross overlapping sampling and traversing cutting on the orthoimages in the verification set and the marked image files of the marked verification set according to the method B2 to obtain a verification input data set;
b4, performing data enhancement processing on the training input data set and the verification input data set in sequence by adopting rotation, mirror image, color transformation, contrast variation and Gaussian noise based on the python platform, wherein the training input data set and the verification input data set after the enhancement processing form an input database together;
C. constructing an improved Unet neural network model, wherein the improved Unet neural network model comprises a down-sampling system and an up-sampling system, the down-sampling system comprises a plurality of down-sampling modules, the up-sampling system comprises a plurality of up-sampling modules, and the down-sampling module is composed of a module structure of cavity convolution support constructed by two cavity convolution layers and a maximum pooling layer;
c1, taking the training input data set after the enhancement processing in the step B and the high-precision digital geographic elevation model data in the step A as an input layer of the improved Unet neural network model, training the improved Unet neural network model, extracting the characteristics of the hole convolution layer of the improved Unet neural network model by using hole convolution, and finally obtaining the characteristic images after the hole convolution as follows:
wherein p (i) represents the feature value extracted at position i, k (f) is the parameter value of the convolution kernel at position f, r is the hole convolution rate, and x (i + r f) is the image value of the corresponding receptive field position;
then, pooling the characteristic images according to the following formula and further refining the characteristic points: p is a radical ofm1Max-pool (p); wherein P ism1The characteristic image is a characteristic image after one layer of pooling, and p is a characteristic image after cavity convolution;
D. the vegetation coverage of the mining area is calculated by the following method:
d1, loading the high-precision digital geographic elevation model data in the step A or the input database in the step B into a trained improved Unet neural network model for classification and identification processing to obtain a classification result, and splicing the classification result images through an Envi platform to obtain a vegetation classification result of the research area;
d2, constructing a fishing net coverage research area with single grid dimension of m multiplied by n, and counting the area of vegetation coverage under the single grid through an Arcgis platform and recording the area as SvegeThen, the vegetation coverage of the study area is obtained by the following formula:
wherein FVC is the vegetation coverage of the study area, SvegeThe area of vegetation coverage under a single grid, k is the number of the grids covered in the study area, SallIs the total area of the ground in the area under study.
Example two
As shown in fig. 1 to 10, a method for calculating a mining area FVC of an enhanced edge sampling and improved Unet model includes the following steps:
A. collecting ground vegetation parameter data in a mining area scene: determining a mining area research area, carrying out overlapped orthographic aerial photography on the mining area research area by adopting an unmanned aerial vehicle according to drawing precision to obtain an orthographic image set, wherein the drawing precision is centimeter-level, the lowest aerial photography overlapping proportion is 32-38%, then carrying out image splicing to obtain an orthographic image of the research area, then establishing an air-to-three calculation model (SFM) and generating dense point cloud data, and obtaining high-precision digital geographic elevation model data through conversion;
according to a preferred embodiment of the present embodiment, in step a, flight planning is performed for the unmanned aerial vehicle including takeoff, landing and navigation routes according to the drawing accuracy of the orthographic image set, the minimum overlap ratio of aerial photography is 35%, and the navigation height formula of the unmanned aerial vehicle is as follows:
h ═ f · m; h is the navigation height, f is the camera focal length, m is the mapping scale, and the course height and the navigation speed of the unmanned aerial vehicle are kept consistent as much as possible when data acquisition is needed.
In this embodiment, taking an ore district of the east of the elder-down-the-art of reduos as an example, firstly, the ore district is subjected to field reconnaissance, and the regional terrain is clearly researched, in this example, an unmanned aerial vehicle of the warrior 4pro is used for data acquisition, the model of the camera is Phantom 4pro v2.0, the flying side image overlapping degree is 60 °, the course overlapping degree is 30 °, and then the whole research area is divided into a plurality of flight survey areas with overlapping areas on *** earth equipment. The flying height is set to 50m in this example and the flying speed is 2.5m/s on average.
According to a preferred embodiment of this embodiment, step a further includes a 1;
a1, arranging ground control points on the ground in a mining area research area, wherein the ground control points comprise a target and an image control point identifier, an orthographic image set overlapped by the unmanned aerial vehicle and orthographic aerial-photographed will form an identifier control point corresponding to the ground control point, and the orthographic image set depends on the identifier control point to perform image registration and geometric calibration when image splicing and/or space-three solution model SFM processing is performed. As shown in fig. 1, ground control points are generally uniformly distributed in the whole survey area along a navigation route, the overall layout of point locations needs to be considered when the ground control points are distributed as well as a certain geometric strength and uniform distribution as possible in the survey area, as shown in fig. 1, wherein dots represent selected control points in the survey area (in this embodiment, an Ore district is taken as an example, and the ground control points are distributed as shown in fig. 8; in this embodiment, the ground control points are obtained, an unmanned aerial vehicle collects identification control points on an original remote sensing image, a context capture platform is used for image processing, the same name image point pair relationship is established for image registration and geometric calibration, then image splicing is performed to obtain an ortho image of a research area, as shown in fig. 9, then an SFM (structure-from-motion) is established to generate dense point cloud data, texture information is added through conversion, and finally, high-precision digital geographic elevation model Data (DEM) is obtained, the distribution density of image control points can be referred to as shown in table 1, and fixed, flat, clear and easily recognized, shadow-free and shelterless areas are selected as far as possible during point selection, such as points of intersection of fine linear ground objects with good intersection angles (30-150 degrees), corners of obvious ground objects, centers of point ground objects with no more than 3 x 3 pixels in original images, and places with small elevation fluctuation, relatively fixed throughout the year and easy and accurate positioning and measurement are selected, or image control point marks are made in a painting mode.
Spatial resolution | Image controlled dot density | Grade of measurement |
1.5cm | 100 to 200 | High-precision cadastral survey |
2cm | 200 to 300 | 1:500 topographical map measurements |
3cm | 300 to 500 | 1:1000 topographical map measurements |
5cm | 500 pieces of | General topography measurement |
TABLE 1 image-controlled Point layout Density reference
In the embodiment, a context capture platform is adopted for image processing, homonymous image points are established for image registration and geometric calibration, then image splicing is carried out to obtain an orthoimage of a research area, then a space-from-motion (SFM) is established to generate dense point cloud data, and high-precision digital geographic elevation model Data (DEM) is obtained through conversion.
B. Constructing a sample data set based on an enlarged selective area cross overlapping method, wherein the method comprises the following steps:
b1, dividing the orthographic image set of the research area into a training set and a verification set, respectively carrying out data annotation on the training set and the verification set by adopting a labelimg tool, marking a vegetation sample in the image, wherein the annotation result is a json file, and then converting the json file into an annotated image file with the same format as that of the orthographic image, so as to obtain an annotated training set and an annotated verification set;
b2, performing cross-over sampling on the ortho-image in the training set and the labeled image file of the labeled training set (in this embodiment, a python program is used to generate a vector file without an overlap region, an Arcpy second-order development program script is used to cut the ortho-image and the labeled image from the vector file, and then the ortho-image and the labeled image in the training region are subjected to cross-over sampling at the same time), wherein a moving window of the sample sampling model is as follows:
W=2n(ii) a Wherein W is the image side length of a single sample data, and n is 9;
establishing a 512 x 512 moving window to perform global traversal on the orthoimage in the training set and the labeled image file labeled with the training set according to a step length L, wherein the step length L is 212, cutting is performed on each traversal according to the size of the moving window to obtain a sample data image block A, the slice size of each sample data image block A is 512 x 512, and after the traversal is completed, data in a 300 x 300 pixel range are selected from the central area of the sample data image block A to form a training input data set;
b3, cross-overlapping and sampling the ortho-image in the verification set and the labeled image file in the labeled verification set, wherein the moving window of the sample sampling model is as follows:
W=2n(ii) a Wherein W is the image side length of a single sample data, and n is 9;
establishing a 512 x 512 moving window to perform global traversal on the orthoimage in the verification set and the labeled image file in the labeled verification set according to a step length L, wherein the step length L is 212, cutting is performed on each traversal according to the size of the moving window to obtain a sample data image block B, the slice size of each sample data image block B is 512 x 512, and after the traversal is completed, data in a 300 x 300 pixel range are selected from the central area of the sample data image block B to form a verification input data set; through repeated testing and adjustment optimization of the inventor, when n is 9, the model has the highest operation efficiency and precision, namely a 512 x 512 moving window is established to carry out global traversal on the marked image and the orthoimage under the same position, the orthoimage is uniformly cut into sample data image blocks with the size of 512 x 512, then the moving window is moved in a traversing manner, the step length of each traversing movement is 212, next sampling is carried out until the whole image is traversed, and the sampling process refers to fig. 2 to form a training set and a verification set; since the size of a single sample slice is 512 x 512, the overlap of 212 pixels between adjacent slices is obtained, and then data of 300 x 300 pixels range in the central area of each slice is taken from the training set and the verification set to form an input data set.
B4, performing data enhancement processing on the training input data set and the verification input data set in sequence by adopting rotation, mirror image, color transformation, contrast variation and Gaussian noise based on the python platform to enhance the generalization capability of the model, wherein the training input data set and the verification input data set after enhancement processing jointly form an input database. According to the method, the sample set range is enlarged, the sample slices are extracted, and the sample slices are overlapped alternately, so that the problem that the edge calculation result is not ideal is solved, the condition that data faults exist in the margins among the samples is improved, the identification precision of a neural network is improved, and the challenge of difficulty in identifying vegetation in a mining area scene is completed, and the method is shown in fig. 6.
C. Constructing an improved Unet neural network model, wherein the improved Unet neural network model comprises a down-sampling system and an up-sampling system, the down-sampling system comprises four down-sampling modules, the up-sampling system comprises four up-sampling modules, and the up-sampling module is formed by convolution with a corresponding transposition of the down-sampling modules; the down-sampling module consists of two cavity convolution layers and a module structure of cavity convolution support constructed by a maximum pooling layer; referring to fig. 3, a dotted line frame in the figure is an innovative structure of the model, and through repeated tests and optimization of experiments, the left side of the neural network is replaced by a down-sampling module supported by cavity convolution constructed by 2 cavity convolution layers and 1 maximum pooling layer, the whole left side is composed of 4 down-sampling modules supported by the cavity convolution, after down-sampling operation is completed, the right side is an up-sampling module composed of transposed convolution, after the feature image is reduced in size through 4 times of down-sampling, the up-sampling size is expanded through the transposed convolution, and simultaneously a feature map transmitted from a left side symmetrical position is superimposed, so that the rightmost classification output is finally obtained.
And C1, taking the training input data set after the enhancement processing in the step B and the high-precision digital geographic elevation model data in the step A as an input layer of the improved Unet neural network model, and training the improved Unet neural network model. The network parameter settings of the improved Unet neural network model of the present embodiment are shown in table 3, and the server configuration of the improved Unet neural network model of the present embodiment is shown in table 4.
Table 3 network parameter settings
TABLE 4 Server configuration
The hole convolution layer of the improved Unet neural network model adopts hole convolution (the feature extraction of the hole convolution layer is shown in figure 4) to carry out feature extraction, and finally obtains a feature image after the hole convolution as follows:wherein p (i) represents the feature value extracted at position i, k (f) is the parameter value of the convolution kernel at position f, r is the hole convolution rate, and x (i + r f) is the image value of the corresponding receptive field position;
referring to fig. 5, the feature images are then pooled and feature points are further refined according to the following formula: p is a radical ofm1Max-pool (p); wherein P ism1The characteristic image is a characteristic image after one layer of pooling, and p is a characteristic image after cavity convolution;
according to a preferred embodiment of this embodiment, in the step C1, the hole convolution layer of the improved Unet neural network model uses a convolution kernel of m × m (preferably, a convolution kernel of 3 × 3; in this embodiment, the convolution rate of r holes is 1) to cross-multiply with the pixel corresponding to the convolution kernel receptive field to obtain the feature image after hole convolution.
According to a preferred embodiment of this embodiment, step C further comprises C2;
c2, performing Precision evaluation on the trained improved Unet neural network model (the improved Unet neural network model is subjected to repeated iterative training for multiple times and the highest-Precision training model is selected for vegetation coverage calculation of a test area) by using a Precision model and/or a Recall model and/or an F1-score model through a verification input data set:
wherein TP is the number of identifying the target sample as true (i.e., true positive number), FP is the number of identifying the target sample as false (i.e., false positive number), FN is the number of identifying the non-target sample as true (i.e., false negative number), Recall is the output value of the Recall evaluation model, Precision is the output value of the Precision evaluation model, and finally the finally obtained F1-score value is taken as the evaluation accuracy of the improved Unet neural network model. In the embodiment, the training set and the verification set are synchronously brought into a common Unet network, a Deeplabv3+ network, a pixel binary model, a linear model and a nonlinear model for model training and precision testing, and the test results are shown as
Table 5 shows the results of the inversion performed at a test speed of 500 samples under the same operating environment.
Model (model) | Precision | Recall | F1score | Speed of measurement |
The invention | 0.972 | 0.963 | 0.967 | 19s |
Unet | 0.902 | 0.932 | 0.916 | 23s |
Deeplabv3+ | 0.892 | 0.872 | 0.881 | 25s |
Pixel binary model | 0.753 | 0.721 | 0.736 | --------- |
Linear model | 0.358 | 0.388 | 0.372 | --------- |
Non-linear model | 0.419 | 0.401 | 0.409 | --------- |
TABLE 5 comparison of the test results of the models
D. The vegetation coverage of the mining area is calculated by the following method:
d1, loading the high-precision digital geographic elevation model data in the step A or the input database in the step B into a trained improved Unet neural network model for classification and identification processing to obtain a classification result, and splicing the classification result images through an Envi platform to obtain a vegetation classification result of the research area;
d2, constructing a fishing net coverage research area with single grid dimension of m multiplied by n, and counting the area of vegetation coverage under the single grid through an Arcgis platform and recording the area as SvegeThen, the vegetation coverage of the study area is obtained by the following formula:
wherein FVC is the vegetation coverage of the study area, SvegeThe area of vegetation coverage under a single grid, k is the number of the grids covered in the study area, SallIs the total area of the ground in the area under study. In the embodiment, the vegetation coverage is calculated by adopting the scale of 1m, and the range scale of the target vegetation coverage is set to be 1m2Recording its area as Sall. The pixel resolution ratio obtained by the orthographic image of the unmanned aerial vehicle is 0.01m2Constructing a single grid with a dimension of 1m2The fishing net covers all result areas, and the area of vegetation coverage under a single grid is counted through the arcgis platform, namely the area of vegetation coverage in each 10 x 10 pixel areas is counted. In the embodiment, all grids in the fishing net are traversed, and 1m of the whole test area can be obtained2Results of vegetation coverage at scale anda visualization result graph shown in fig. 10 is obtained.
A mining area FVC computing system for enhancing edge sampling and improving a Unet model comprises a ground vegetation parameter data extraction system, a sampling system for expanding a selected area cross overlapping method, an improved Unet neural network model and a mining area vegetation coverage statistical module, wherein the ground vegetation parameter data extraction system comprises an unmanned aerial vehicle aerial photography control unit, an image splicing module and an aerial three-dimensional calculation model SFM; the sampling system of the expanded selective area cross-overlapping method comprises a cross-overlapping sampling module, the sampling system of the expanded selective area cross-overlapping method is provided with a labelimg tool and a python platform, the sampling system of the expanded selective area cross-overlapping method carries out data labeling through the labelimg tool and obtains a labeled image file, the cross-overlapping sampling module carries out overall traversal cutting sampling according to a moving window by adopting a sample sampling model and a step length L and obtains a sample data image block, and the sampling system of the expanded selective area cross-overlapping method carries out data processing through the python platform and obtains an input database; the improved Unet neural network model comprises a down-sampling system and an up-sampling system, and is trained to obtain a trained improved Unet neural network model; the mining area vegetation coverage statistical module classifies, identifies and processes high-precision digital geographic elevation model data or an input database through a trained improved Unet neural network model to obtain a classification result, then splices classification result images through an Envi platform to obtain a vegetation classification result of a research area, and constructs a fishing net coverage research area with a single grid scale of mxn and counts the vegetation coverage of the research area through an Arcgis platform.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (7)
1. A mining area FVC calculation method for enhancing edge sampling and improving a Unet model is characterized in that: the method comprises the following steps:
A. collecting ground vegetation parameter data in a mining area scene: determining a mining area research area, carrying out overlapped orthographic aerial photography on the mining area research area by adopting an unmanned aerial vehicle according to drawing precision to obtain an orthographic image set, wherein the lowest aerial photography overlapping proportion is 32-38%, then carrying out image splicing to obtain an orthographic image of the research area, then establishing an air-to-three calculation model (SFM) and generating dense point cloud data, and obtaining high-precision digital geographic elevation model data through conversion;
B. constructing a sample data set based on an enlarged selective area cross overlapping method, wherein the method comprises the following steps:
b1, dividing the orthographic image set of the research area into a training set and a verification set, respectively carrying out data annotation on the training set and the verification set by adopting a labelimg tool, marking a vegetation sample in the image, wherein the annotation result is a json file, and then converting the json file into an annotated image file with the same format as that of the orthographic image, so as to obtain an annotated training set and an annotated verification set;
b2, simultaneously, cross-overlapping and sampling the orthoimage in the training set and the labeled image file labeled with the training set, wherein the moving window of the sample sampling model is as follows:
W=2n(ii) a Wherein W is the image side length of a single sample data, and n is a positive integer;
establishment 2n×2nThe moving window carries out global traversal on the orthoimages in the training set and the marked image files marked with the training set according to the step length L, wherein the step length L is less than 2nCutting according to the size of the moving window in each traversal to obtain a sample data image block, and completing the traversal to obtain a training input data set;
b3, performing cross overlapping sampling and traversing cutting on the orthoimages in the verification set and the marked image files of the marked verification set according to the method B2 to obtain a verification input data set;
b4, performing data enhancement processing on the training input data set and the verification input data set in sequence by adopting rotation, mirror image, color transformation, contrast variation and Gaussian noise based on the python platform, wherein the training input data set and the verification input data set after the enhancement processing form an input database together;
C. constructing an improved Unet neural network model, wherein the improved Unet neural network model comprises a down-sampling system and an up-sampling system, the down-sampling system comprises a plurality of down-sampling modules, the up-sampling system comprises a plurality of up-sampling modules, and the down-sampling module is composed of a module structure of cavity convolution support constructed by two cavity convolution layers and a maximum pooling layer;
c1, taking the training input data set after the enhancement processing in the step B and the high-precision digital geographic elevation model data in the step A as an input layer of the improved Unet neural network model, training the improved Unet neural network model, extracting the characteristics of the hole convolution layer of the improved Unet neural network model by using hole convolution, and finally obtaining the characteristic images after the hole convolution as follows:wherein p (i) represents the feature value extracted at position i, k (f) is the parameter value of the convolution kernel at position f, r is the hole convolution rate, and x (i + r f) is the image value of the corresponding receptive field position;
then, pooling the characteristic images according to the following formula and further refining the characteristic points: p is a radical ofm1Max-pool (p); wherein P ism1The characteristic image is a characteristic image after one layer of pooling, and p is a characteristic image after cavity convolution;
D. the vegetation coverage of the mining area is calculated by the following method:
d1, loading the high-precision digital geographic elevation model data in the step A or the input database in the step B into a trained improved Unet neural network model for classification and identification processing to obtain a classification result, and splicing the classification result images through an Envi platform to obtain a vegetation classification result of the research area;
d2, constructing a fishing net coverage research area with single grid dimension of m multiplied by n, and counting the area of vegetation coverage under the single grid through an Arcgis platform and recording the area as SvegeThen, the vegetation coverage of the study area is obtained by the following formula:
2. A mining area FVC calculation method for enhancing edge sampling and improving a Unet model comprises the following steps:
A. collecting ground vegetation parameter data in a mining area scene: determining a mining area research area, carrying out overlapped orthographic aerial photography on the mining area research area by adopting an unmanned aerial vehicle according to drawing precision to obtain an orthographic image set, wherein the drawing precision is centimeter-level, the lowest aerial photography overlapping proportion is 32-38%, then carrying out image splicing to obtain an orthographic image of the research area, then establishing an air-to-three calculation model (SFM) and generating dense point cloud data, and obtaining high-precision digital geographic elevation model data through conversion;
B. constructing a sample data set based on an enlarged selective area cross overlapping method, wherein the method comprises the following steps:
b1, dividing the orthographic image set of the research area into a training set and a verification set, respectively carrying out data annotation on the training set and the verification set by adopting a labelimg tool, marking a vegetation sample in the image, wherein the annotation result is a json file, and then converting the json file into an annotated image file with the same format as that of the orthographic image, so as to obtain an annotated training set and an annotated verification set;
b2, simultaneously, cross-overlapping and sampling the orthoimage in the training set and the labeled image file labeled with the training set, wherein the moving window of the sample sampling model is as follows:
W=2n(ii) a Wherein W isThe side length of an image of a single sample data is 9;
establishing a 512 x 512 moving window to perform global traversal on the orthoimage in the training set and the labeled image file labeled with the training set according to a step length L, wherein the step length L is 212, cutting is performed on each traversal according to the size of the moving window to obtain a sample data image block A, the slice size of each sample data image block A is 512 x 512, and after the traversal is completed, data in a 300 x 300 pixel range are selected from the central area of the sample data image block A to form a training input data set;
b3, cross-overlapping and sampling the ortho-image in the verification set and the labeled image file in the labeled verification set, wherein the moving window of the sample sampling model is as follows:
W=2n(ii) a Wherein W is the image side length of a single sample data, and n is 9;
establishing a 512 x 512 moving window to perform global traversal on the orthoimage in the verification set and the labeled image file in the labeled verification set according to a step length L, wherein the step length L is 212, cutting is performed on each traversal according to the size of the moving window to obtain a sample data image block B, the slice size of each sample data image block B is 512 x 512, and after the traversal is completed, data in a 300 x 300 pixel range are selected from the central area of the sample data image block B to form a verification input data set;
b4, performing data enhancement processing on the training input data set and the verification input data set in sequence by adopting rotation, mirror image, color transformation, contrast variation and Gaussian noise based on the python platform, wherein the training input data set and the verification input data set after the enhancement processing form an input database together;
C. constructing an improved Unet neural network model, wherein the improved Unet neural network model comprises a down-sampling system and an up-sampling system, the down-sampling system comprises four down-sampling modules, the up-sampling system comprises four up-sampling modules, and the up-sampling module is formed by convolution with a corresponding transposition of the down-sampling modules; the down-sampling module consists of two cavity convolution layers and a module structure of cavity convolution support constructed by a maximum pooling layer;
c1, enhancing the processed training input data set in step B, step AThe high-precision digital geographic elevation model data is used as an input layer of the improved Unet neural network model and is used for training the improved Unet neural network model, the cavity convolution layer of the improved Unet neural network model adopts cavity convolution to extract the characteristics, and finally the characteristic images after the cavity convolution are obtained as follows:wherein p (i) represents the feature value extracted at position i, k (f) is the parameter value of the convolution kernel at position f, r is the hole convolution rate, and x (i + r f) is the image value of the corresponding receptive field position;
then, pooling the characteristic images according to the following formula and further refining the characteristic points: p is a radical ofm1Max-pool (p); wherein P ism1The characteristic image is a characteristic image after one layer of pooling, and p is a characteristic image after cavity convolution;
D. the vegetation coverage of the mining area is calculated by the following method:
d1, loading the high-precision digital geographic elevation model data in the step A or the input database in the step B into a trained improved Unet neural network model for classification and identification processing to obtain a classification result, and splicing the classification result images through an Envi platform to obtain a vegetation classification result of the research area;
d2, constructing a fishing net coverage research area with single grid dimension of m multiplied by n, and counting the area of vegetation coverage under the single grid through an Arcgis platform and recording the area as SvegeThen, the vegetation coverage of the study area is obtained by the following formula:
3. A method for calculating the FVC of a mine site with enhanced edge sampling and improved Unet model according to claim 1 or 2, wherein: step C also includes C2;
c2, adopting Precision model and/or Recall model and/or F1-score model to evaluate the Precision of the trained improved Unet neural network model by using the verification input data set:
wherein TP is the number of identifying the target sample as true, FP is the number of identifying the target sample as false, FN is the number of identifying the non-target sample as true, Recall is the output value of Recall evaluation model, Precision is the output value of Precision evaluation model, and finally the finally obtained F1-score value is used as the evaluation Precision of the improved Unet neural network model.
4. A method for calculating the FVC of a mine site with enhanced edge sampling and improved Unet model according to claim 1 or 2, wherein: in the step A, flight planning is carried out on the unmanned aerial vehicle including takeoff, landing and navigation routes according to the drawing precision of the orthographic image set, the minimum overlap proportion of aerial photography is 35%, and the navigation height formula of the unmanned aerial vehicle is as follows:
h ═ f · m; h is the navigation height, f is the camera focal length, m is the mapping scale, and the course height and the navigation speed of the unmanned aerial vehicle are kept consistent as much as possible when data acquisition is needed.
5. A method for calculating the FVC of a mine site with enhanced edge sampling and improved Unet model according to claim 1 or 2, wherein: step A also comprises A1;
a1, arranging ground control points on the ground in a mining area research area, wherein the ground control points comprise a target and an image control point identifier, an orthographic image set overlapped by the unmanned aerial vehicle and orthographic aerial-photographed will form an identifier control point corresponding to the ground control point, and the orthographic image set depends on the identifier control point to perform image registration and geometric calibration when image splicing and/or space-three solution model SFM processing is performed.
6. A method for calculating the FVC of a mine site with enhanced edge sampling and improved Unet model according to claim 1 or 2, wherein: in the step C1, the hole convolution layer of the improved Unet neural network model is cross-multiplied by the convolution kernel of m × m and the pixel corresponding to the convolution kernel receptive field to obtain the feature image after hole convolution.
7. A mining area FVC computing system for enhancing edge sampling and improving Unet model is characterized in that: the system comprises a ground vegetation parameter data extraction system, an extended-area cross-overlapping sampling system, an improved Unet neural network model and a mining area vegetation coverage statistical module, wherein the ground vegetation parameter data extraction system comprises an unmanned aerial vehicle aerial photography control unit, an image splicing module and an aerial three-resolution model SFM; the sampling system of the expanded selective area cross-overlapping method comprises a cross-overlapping sampling module, the sampling system of the expanded selective area cross-overlapping method is provided with a labelimg tool and a python platform, the sampling system of the expanded selective area cross-overlapping method carries out data labeling through the labelimg tool and obtains a labeled image file, the cross-overlapping sampling module carries out overall traversal cutting sampling according to a moving window by adopting a sample sampling model and a step length L and obtains a sample data image block, and the sampling system of the expanded selective area cross-overlapping method carries out data processing through the python platform and obtains an input database; the improved Unet neural network model comprises a down-sampling system and an up-sampling system, and is trained to obtain a trained improved Unet neural network model; the mining area vegetation coverage statistical module classifies, identifies and processes high-precision digital geographic elevation model data or an input database through a trained improved Unet neural network model to obtain a classification result, then splices classification result images through an Envi platform to obtain a vegetation classification result of a research area, and constructs a fishing net coverage research area with a single grid scale of mxn and counts the vegetation coverage of the research area through an Arcgis platform.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111201530.9A CN113920262B (en) | 2021-10-15 | 2021-10-15 | Mining area FVC calculation method and system for enhancing edge sampling and improving Unet model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111201530.9A CN113920262B (en) | 2021-10-15 | 2021-10-15 | Mining area FVC calculation method and system for enhancing edge sampling and improving Unet model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113920262A true CN113920262A (en) | 2022-01-11 |
CN113920262B CN113920262B (en) | 2022-04-12 |
Family
ID=79240607
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111201530.9A Expired - Fee Related CN113920262B (en) | 2021-10-15 | 2021-10-15 | Mining area FVC calculation method and system for enhancing edge sampling and improving Unet model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113920262B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114782801A (en) * | 2022-03-01 | 2022-07-22 | 泰瑞数创科技(北京)有限公司 | Machine learning-based automatic control point layout method and device |
CN116363502A (en) * | 2023-01-31 | 2023-06-30 | 中国科学院地理科学与资源研究所 | Mining land multidimensional information acquisition method and device integrating multisource geographic big data |
CN117274499A (en) * | 2023-10-17 | 2023-12-22 | 山东泰山普惠建工有限公司 | Unmanned aerial vehicle oblique photography-based steel structure processing and mounting method |
CN117315146A (en) * | 2023-09-22 | 2023-12-29 | 武汉大学 | Reconstruction method and storage method of three-dimensional model based on trans-scale multi-source data |
CN117671437A (en) * | 2023-10-19 | 2024-03-08 | 中国矿业大学(北京) | Open stope identification and change detection method based on multitasking convolutional neural network |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103824223A (en) * | 2014-02-21 | 2014-05-28 | 浙江大学 | Crop yield remote sensing estimation method based on MapReduce and neural network |
CN113128405A (en) * | 2021-04-20 | 2021-07-16 | 北京航空航天大学 | Plant identification and model construction method combining semantic segmentation and point cloud processing |
CN113435411A (en) * | 2021-07-26 | 2021-09-24 | 中国矿业大学(北京) | Improved DeepLabV3+ based open pit land utilization identification method |
CN113469094A (en) * | 2021-07-13 | 2021-10-01 | 上海中科辰新卫星技术有限公司 | Multi-mode remote sensing data depth fusion-based earth surface coverage classification method |
-
2021
- 2021-10-15 CN CN202111201530.9A patent/CN113920262B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103824223A (en) * | 2014-02-21 | 2014-05-28 | 浙江大学 | Crop yield remote sensing estimation method based on MapReduce and neural network |
CN113128405A (en) * | 2021-04-20 | 2021-07-16 | 北京航空航天大学 | Plant identification and model construction method combining semantic segmentation and point cloud processing |
CN113469094A (en) * | 2021-07-13 | 2021-10-01 | 上海中科辰新卫星技术有限公司 | Multi-mode remote sensing data depth fusion-based earth surface coverage classification method |
CN113435411A (en) * | 2021-07-26 | 2021-09-24 | 中国矿业大学(北京) | Improved DeepLabV3+ based open pit land utilization identification method |
Non-Patent Citations (5)
Title |
---|
HONGYAN ZHANG等: "Investigating the accuracy of vegetation index-based models for estimating the fractional vegetation cover and the effects of varying soil backgrounds using in situ measurements and the PROSAIL model", 《INTERNATIONAL JOURNAL OF REMOTE SENSING》 * |
吴志峰等: "时空协同的精准农业遥感研究", 《地球信息科学学报》 * |
李全生等: "东部草原区大型煤电基地开发的生态影响与修复技术", 《煤炭学报》 * |
闫利等: "基于注意力网络的遥感影像植被提取方法", 《测绘地理信息》 * |
陈昂等: "基于面向对象与深度学习的榆树疏林识别方法研究", 《地球信息科学学报》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114782801A (en) * | 2022-03-01 | 2022-07-22 | 泰瑞数创科技(北京)有限公司 | Machine learning-based automatic control point layout method and device |
CN116363502A (en) * | 2023-01-31 | 2023-06-30 | 中国科学院地理科学与资源研究所 | Mining land multidimensional information acquisition method and device integrating multisource geographic big data |
CN116363502B (en) * | 2023-01-31 | 2023-10-20 | 中国科学院地理科学与资源研究所 | Mining land multidimensional information acquisition method and device integrating multisource geographic big data |
CN117315146A (en) * | 2023-09-22 | 2023-12-29 | 武汉大学 | Reconstruction method and storage method of three-dimensional model based on trans-scale multi-source data |
CN117315146B (en) * | 2023-09-22 | 2024-04-05 | 武汉大学 | Reconstruction method and storage method of three-dimensional model based on trans-scale multi-source data |
CN117274499A (en) * | 2023-10-17 | 2023-12-22 | 山东泰山普惠建工有限公司 | Unmanned aerial vehicle oblique photography-based steel structure processing and mounting method |
CN117274499B (en) * | 2023-10-17 | 2024-05-07 | 山东泰山普惠建工有限公司 | Unmanned aerial vehicle oblique photography-based steel structure processing and mounting method |
CN117671437A (en) * | 2023-10-19 | 2024-03-08 | 中国矿业大学(北京) | Open stope identification and change detection method based on multitasking convolutional neural network |
Also Published As
Publication number | Publication date |
---|---|
CN113920262B (en) | 2022-04-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113920262B (en) | Mining area FVC calculation method and system for enhancing edge sampling and improving Unet model | |
CN108020211B (en) | Method for estimating biomass of invasive plants through aerial photography by unmanned aerial vehicle | |
CN114037911B (en) | Large-scale forest height remote sensing inversion method considering ecological zoning | |
CN110245709B (en) | 3D point cloud data semantic segmentation method based on deep learning and self-attention | |
Lu et al. | Cultivated land information extraction in UAV imagery based on deep convolutional neural network and transfer learning | |
CN110263717B (en) | Method for determining land utilization category of street view image | |
Mourato et al. | Improving a DSM obtained by unmanned aerial vehicles for flood modelling | |
CN104680167A (en) | Aurora oval position determining method based on deep learning | |
CN115561181A (en) | Water quality inversion method based on multispectral data of unmanned aerial vehicle | |
CN115546551A (en) | Deep learning-based geographic information extraction method and system | |
CN103777196B (en) | Based on terrain object distance single station measuring method and the measuring system thereof of geography information | |
Rusli et al. | Accuracy assessment of DEM from UAV and TanDEM-X imagery | |
CN111414867A (en) | Method for measuring and calculating aboveground biomass of plants | |
Nwilag et al. | Digital topographic mapping and modelling using low altitude unmanned aerial vehicle | |
KR102587445B1 (en) | 3d mapping method with time series information using drone | |
CN112686988B (en) | Three-dimensional modeling method, three-dimensional modeling device, electronic equipment and storage medium | |
Ćatić et al. | Using the semi-professional UAV system in surveying the medium size area of complex urban surface | |
CN113418510A (en) | High-standard farmland acceptance method based on multi-rotor unmanned aerial vehicle | |
Hron et al. | Use of aerial images for regular updates of buildings in the fundamental base of geographic data of the Czech Republic | |
Loghin et al. | Supervised classification and its repeatability for point clouds from dense VHR tri-stereo satellite image matching using machine learning | |
Albattah et al. | The potential of space geomatics engineering applications in transportation analysis and planning | |
Dansena et al. | Cadastral resurvey and integration of geospatial technique using high-resolution satellite image (HRSI): a case study in Indian district | |
Šiljeg et al. | Quality Assessment of Worldview-3 Stereo Imagery Derived Models Over Millennial Olive Groves | |
Idris et al. | LiDAR Assessments and Mapping for Klang Valley: a Case Study at Jinjang District, Selangor | |
Prakash et al. | Assessment of Urban Built-Up Volume Using Geospatial Methods: A Case Study of Bangalore |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20220412 |
|
CF01 | Termination of patent right due to non-payment of annual fee |