CN115546179A - Forest diameter at breast height volume accurate prediction method based on optimized fuzzy depth network - Google Patents

Forest diameter at breast height volume accurate prediction method based on optimized fuzzy depth network Download PDF

Info

Publication number
CN115546179A
CN115546179A CN202211316076.6A CN202211316076A CN115546179A CN 115546179 A CN115546179 A CN 115546179A CN 202211316076 A CN202211316076 A CN 202211316076A CN 115546179 A CN115546179 A CN 115546179A
Authority
CN
China
Prior art keywords
fuzzy
forest
tree
crown
volume
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.)
Pending
Application number
CN202211316076.6A
Other languages
Chinese (zh)
Inventor
云挺
姜维
张怀清
曹林
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing Forestry University
Original Assignee
Nanjing Forestry University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing Forestry University filed Critical Nanjing Forestry University
Priority to CN202211316076.6A priority Critical patent/CN115546179A/en
Publication of CN115546179A publication Critical patent/CN115546179A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/50—Depth or shape recovery
    • G06T7/521—Depth or shape recovery from laser ranging, e.g. using interferometry; from the projection of structured light
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10028—Range image; Depth image; 3D point clouds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30181—Earth observation
    • G06T2207/30188—Vegetation; Agriculture

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Multimedia (AREA)
  • Optics & Photonics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a forest diameter at breast height volume accurate prediction method based on an optimized fuzzy depth network, which comprises the following steps: preprocessing forest land point cloud data and segmenting single trees; acquiring forest parameters of a single tree; establishing a forest parameter prediction network, wherein the forest parameter prediction network comprises a fuzzy depth network and a pigeon group optimization module, training the forest parameter prediction network, outputting a predicted value after training is finished, and transmitting the predicted value to the pigeon group optimization module, updating parameters of the fuzzy depth network by the pigeon group optimization module and finishing optimal parameter searching, finishing self-adaptive training by the fuzzy depth network according to the optimal parameters, and establishing a forest parameter prediction model; the invention develops a forest parameter prediction model, provides a self-adaptive algorithm to enhance the generalization capability of the forest parameter prediction model to different forest varieties, embeds an attention mechanism module to enhance the robustness of a network, and adjusts the parameters of a fuzzy depth network in real time by fusing a pigeon swarm optimization algorithm, thereby further improving the prediction precision and the learning capability of the model.

Description

Forest diameter at breast height volume accurate prediction method based on optimized fuzzy depth network
Technical Field
The invention belongs to the technical field of forest parameter research, and particularly relates to a forest diameter-height timber volume accurate prediction method based on an optimized fuzzy depth network.
Background
The accurate prediction of the breast diameter and the volume of the forest plays an important role in forest resource investigation, national wood strategic storage and carbon sink assessment. In recent years, methods for predicting forest parameters are roughly divided into two categories, namely, methods for predicting forest parameters based on manual measurement and laser point cloud. The two methods have difference in data acquisition, but the methods for establishing the prediction model have certain similarity.
A forest parameter prediction method based on each check ruler generally establishes a candidate model to estimate forest parameter values, establishes a tree height-breast diameter prediction model by combining a tree height residual error method, or establishes four candidate models to estimate the forest accumulation amount by utilizing a stepwise regression method and a partial least square method, or predicts the canopy volume, the canopy surface area and the biomass of an artificial forest based on a nonlinear mixed effect model. Many commonly used growth models or related extension models are selected as candidate models for the research, and are screened through a series of evaluation criteria, however, the types and the number of the candidate models are different according to research targets.
With the high-speed development of laser measurement technology and the application of the laser measurement technology to forestry informatization, the mode of acquiring forest parameters is generated by converting the traditional check scale of each tree into three-dimensional laser point cloud. Laser scanning has very high distance detection capability and stability, and meanwhile, the processing and analysis of tree three-dimensional point clouds are developed by combining the theories of computer graphics, machine learning and the like, such as: the method comprises the steps of utilizing a watershed algorithm to carry out crown segmentation on forest point cloud, adopting a support vector machine to establish a tree species identification frame, or estimating the breast diameter of a single tree by using random Hough transform and octree segmentation, and extracting the tree height according to the growth direction of the tree, thereby achieving certain progress.
Nowadays, as artificial intelligence technology continuously makes new breakthroughs and laser point cloud scanning technology matures, the machine learning technology leaves open the head in forest point cloud processing, and some forest parameter prediction models are generated. For example, a forest parameter prediction model proposed in 2020 proposes a hierarchical estimation method, which combines a Synthetic Aperture Radar (SAR), a LiDAR and a passive optical system forward model to generate a geometric model and an electromagnetic model of a real forest stand, and combines the geometric model and the electromagnetic model to estimate forest parameters; the breast diameter prediction model proposed in 2021 is based on a generalized nonlinear mixed effect method, random effects at a site level are added to improve the generalization capability of the model, and variables at a single tree and laser point cloud level are used as prediction factors, so that higher prediction accuracy is realized. Some of the works directly acquire information from forest point clouds, and a prediction model is established based on forest stand scale parameters, such as establishing a k-nearest neighbor prediction model, a different-speed growth model, a point cloud feature fusion prediction model, limited region growth and the like, so as to carry out the swedish large-area forest aboveground biomass prediction, research on the relation between the breast diameter and the forest biomass in tropical forests, calculation of the tree crown coverage rate of ginkgo artificial forests, and automatic segmentation of the forest stands of the qilian large-wild forest. Still other researches are carried out to perform individual segmentation on forest point clouds to obtain individual forest parameters, and then a single tree breast diameter prediction model of larch-Yun Lengsha mixed forest is established, a large hillock artificial forest stand cross-section area growth model is established, a tree height prediction model of Lexian forest fir is established and the like by adopting a random forest, a support vector machine and a BP neural network.
Although machine learning has achieved some success in forest parameter prediction, the following problems still exist: 1) In actual forest lands, part of individuals are subjected to natural disasters or forest competition, necrosis, compression and the like occur and become abnormal samples, most forest parameter prediction models lack a mechanism for judging the abnormal samples, the models are easily interfered by the samples, and the prediction accuracy of the models needs to be improved; 2) The model parameters are determined by adopting methods such as a trial and error method and an empirical formula based on the machine learning prediction model, the number of algorithm parameters cannot be further selected and the prior experience is relied on, and the related prediction model has an improved space in structure and modules; 3) The real-time acquisition of forest growth parameters is beneficial to the planting and cultivation of large-scale artificial forests, and the mutual combination of artificial intelligence and an airborne laser radar and the related work applied to forest parameter prediction are not enough at present.
Disclosure of Invention
The invention aims to solve the technical problem of providing an accurate prediction method of the diameter at breast height of a forest based on an optimized fuzzy depth network aiming at the defects of the prior art, developing a forest parameter prediction model by the accurate prediction method of the diameter at breast height of the forest based on the optimized fuzzy depth network, establishing a nonlinear relation between forest parameters, providing an adaptive algorithm to enhance the generalization capability of the forest parameter prediction model to different forest varieties, embedding an attention mechanism module to enhance the robustness of the network, and fusing a pigeon swarm optimization algorithm to adjust the parameters of the fuzzy depth network in real time so as to further improve the prediction accuracy and the learning capability of the model.
In order to achieve the technical purpose, the technical scheme adopted by the invention is as follows:
the method for accurately predicting the diameter at breast height volume of the forest based on the optimized fuzzy depth network comprises the following steps:
step 1: acquiring forest land point cloud data through an airborne laser radar;
step 2: denoising the point cloud data, filtering the point cloud data by adopting a point cloud ground point filtering method, and performing single-tree segmentation on the filtered point cloud data;
and step 3: acquiring forest parameters of the single tree, including east-west crown width, south-north crown width, tree height, point cloud density and crown product, according to the point cloud of the single tree after the division;
and 4, step 4: acquiring the diameter at breast height and volume of the corresponding tree by adopting an artificial mapping method;
and 5: taking east-west crown breadth, south-north crown breadth, tree height, point cloud density, crown product, breast diameter and volume of timber of a plurality of trees as a training sample data set;
step 6: establishing a forest parameter prediction network, wherein the forest parameter prediction network comprises a fuzzy depth network and a pigeon swarm optimization module, training the forest parameter prediction network by adopting a training sample data set, the input of the forest parameter prediction network is east-west crown width, south-north crown width, tree height, point cloud density and crown product of a tree, the output is breast diameter and volume of the tree, a predicted value is output after the training of the fuzzy depth network is finished, the predicted value is output to the pigeon swarm optimization module, the parameters of the fuzzy depth network are updated by the pigeon swarm optimization module, after the searching of the optimal parameters of the fuzzy depth network is finished by the pigeon swarm optimization module, the fuzzy depth network completes self-adaptive training according to the optimal parameters, and a final forest parameter prediction model is established;
and 7: acquiring point cloud data of the forest land to be detected, acquiring east-west crown width, north-south crown width, tree height, point cloud density and crown product of each tree in the forest land to be detected according to the methods in the step 2 and the step 3, and inputting the east-west crown width, the south-north crown width, the tree height, the point cloud density and the crown product of the tree to a final forest parameter prediction model to obtain predicted values of the breast diameter and the volume of the corresponding tree.
As a further improved technical scheme of the present invention, the method for obtaining the forest parameters of the single tree in step 3 comprises:
selecting the maximum distance from the east-west crown width in the east-west direction of the single tree crown point cloud; selecting the maximum distance as the south-north crown breadth in the south-north direction of the single tree crown point cloud; the vertical distance between the highest point of the single tree crown point cloud and the horizontal plane is the tree height; dividing the total number of the individual plant point clouds by the projection area of the crown to obtain point cloud density; and calculating the convex hull volume of the single tree crown point cloud, wherein the convex hull volume is the crown volume.
As a further improved technical scheme of the invention, the method for acquiring the breast diameter and the volume of the tree in the step 4 comprises the following steps:
acquiring the perimeter of a trunk at the trunk part of the tree 1.3m away from the ground through a measuring tape, wherein the perimeter is the breast-height diameter of the tree;
selecting the diameter of the upper part of the trunk of the tree as
Figure BDA0003909354190000021
The point is used as a shape point, D is the diameter at breast height of the tree, and the length from the shape point to the treetop is measured as H t Measuring the cross-sectional area of the breast diameter to be S D Measuring the height of the tree as H, the volume of the tree is:
Figure BDA0003909354190000022
in the formula: r is the dry form index; s vl Is the volume of the tree.
As a further improved technical solution of the present invention, the fuzzy depth network in step 6 sequentially includes, from input to output, an adaptive fuzzy layer, a fuzzy inference layer, and a weight update layer based on attention.
As a further improved technical scheme of the invention, the calculation process of the adaptive fuzzy layer comprises the following steps:
input attribute x u At the adaptive fuzzy level, there are k fuzzy subsets, u ∈ {1,2,3,4,5}, where x 1 、x 2 、x 3 、x 4 、x 5 The method comprises the steps that input attributes of a self-adaptive fuzzy layer are respectively east-west crown breadth, south-north crown breadth, tree height, point cloud density and crown product in sequence, and each input attribute has k fuzzy subsets; the adaptive fuzzy layer has 5 x k fuzzy subsets; training sample
Figure BDA0003909354190000031
Input attribute x in the incoming adaptive fuzzy layer u The jth fuzzy subset to which the system belongs outputs the degree of membership
Figure BDA0003909354190000032
As shown in equation (2):
Figure BDA0003909354190000033
in the formula: x is the number of u Centered on the jth fuzzy subset of the adaptive fuzzy layer
Figure BDA0003909354190000034
x u The variance of the jth fuzzy subset at the adaptive fuzzy layer is
Figure BDA0003909354190000035
j =1, …, k, the total number of fuzzy subsets of the input attributes is k, and k is updated by a pigeon flock optimization module, i =1, …, n, n is the number of training samples, namely the total number of trees of the training samples;
according to x u Solving for n training samples
Figure BDA0003909354190000036
Local density of
Figure BDA0003909354190000037
And distance
Figure BDA0003909354190000038
i =1, …, n, u ∈ {1,2,3,4,5}, local density
Figure BDA0003909354190000039
The calculation formula is as follows:
Figure BDA00039093541900000310
in the formula:
Figure BDA00039093541900000311
is that
Figure BDA00039093541900000312
Local density of d u Is x u U e {1,2,3,4,5}, i =1,2, …, n, m =1,2, …, n;
respectively calculate the local density is greater than
Figure BDA00039093541900000313
All training samples of (1) and
Figure BDA00039093541900000314
with the minimum value as
Figure BDA00039093541900000315
Is a distance of
Figure BDA00039093541900000316
As shown in equation (4):
Figure BDA00039093541900000317
in the formula:
Figure BDA00039093541900000318
is that
Figure BDA00039093541900000319
Local density of
Figure BDA00039093541900000320
m＝1,2,…n；
Figure BDA00039093541900000321
Is composed of
Figure BDA00039093541900000322
And with
Figure BDA00039093541900000323
U ∈ {1,2,3,4,5}; if it is
Figure BDA00039093541900000324
Has the highest local density of
Figure BDA00039093541900000325
And m is not equal to i;
computing
Figure BDA00039093541900000326
As an input attribute x u Performing clustering of initial centersPerformance value
Figure BDA00039093541900000327
As shown in equation (5):
Figure BDA0003909354190000041
in the formula: max (rho) u ) Is rho u Maximum of n local densities; min (rho) u ) Is rho u Minimum of n local densities; max (delta) u ) Is delta u Maximum of the n distances; min (delta) u ) Is delta u The minimum of the n distances; rho u For n training samples as input attribute x u The local density of (a), including n local densities; delta u For n training samples as input attribute x u Including n distances;
step (a), inputting attribute x u N training samples
Figure BDA0003909354190000042
According to the following
Figure BDA0003909354190000043
Sorting the values in a descending order, and taking the first k sample points as x u Carrying out DPKM clustering on the n training samples to obtain initial central points; step (b), calculating input attribute x u According to the Euclidean distance between the n training samples and the k category centers, the training samples are distributed to the category where the nearest center is located according to the minimum distance distribution principle; after n training samples are distributed for one time, calculating the mean value of each category, taking the mean value as a new category center and updating the category center; repeating the step (b) and the step (c) until the variation of the category center is smaller than the set error; completing DPKM clustering, wherein the center of k clusters is x u Centering on jth fuzzy subset of adaptive fuzzy layer
Figure BDA0003909354190000044
j＝1,…,k；
When inputting attribute x u Centering on the jth fuzzy subset of the adaptive fuzzy layer
Figure BDA0003909354190000045
And
Figure BDA0003909354190000046
belonging DPKM cluster
Figure BDA0003909354190000047
After establishment, x is calculated according to the following adaptive algorithm u Variance of jth fuzzy subset in adaptive fuzzy layer
Figure BDA0003909354190000048
First calculate x u Center of the k fuzzy subsets center
Figure BDA0003909354190000049
As shown in equation (6):
Figure BDA00039093541900000410
then solve for clustering
Figure BDA00039093541900000411
Inner sample of
Figure BDA00039093541900000412
And with
Figure BDA00039093541900000413
Average euclidean distance of
Figure BDA00039093541900000414
As shown in equation (7):
Figure BDA00039093541900000415
in the formula:
Figure BDA00039093541900000416
represent
Figure BDA00039093541900000417
Inner sample of
Figure BDA00039093541900000418
And
Figure BDA00039093541900000419
the Euclidean distance;
Figure BDA00039093541900000420
is composed of
Figure BDA00039093541900000421
Inner sample of
Figure BDA00039093541900000422
The number of (2);
Figure BDA0003909354190000051
in the formula:
Figure BDA0003909354190000052
is composed of
Figure BDA0003909354190000053
A width factor of (d); alpha is a variance scaling coefficient and is updated by a pigeon flock optimization module;
Figure BDA0003909354190000054
as an input attribute x u The maximum distance between every two of the k fuzzy subsets; u ∈ {1,2,3,4,5}, j =1,2, …, k.
As a further improved technical scheme of the invention, the calculation process of the fuzzy inference layer is as follows:
membership degree is input to a fuzzy inference layer and adoptsThe product reasoning method comprises establishing fuzzy unit of fuzzy reasoning layer, and calculating jth unit output value of ith training sample
Figure BDA0003909354190000055
As a blur unit output, as shown in equation (9):
Figure BDA0003909354190000056
in the formula:
Figure BDA0003909354190000057
the product of the membership degrees of the jth fuzzy subset of the ith training sample in different input attributes;
Figure BDA0003909354190000058
after normalization processing is
Figure BDA0003909354190000059
u belongs to {1,2,3,4,5}, i =1, …, n, j =1,2, …, k, n is the total number of trees in the training sample, and the total number of fuzzy units is k.
As a further improved technical solution of the present invention, the calculation process of the attention-based weight update layer is as follows:
the loss function is shown in equation (10):
Figure BDA00039093541900000510
in the formula: y is 1 Is the predicted value of breast diameter, y 2 The predicted value of the volume is taken as the volume;
Figure BDA00039093541900000511
is the measured value of the diameter of the breast,
Figure BDA00039093541900000512
measured value of volume of timber; q i Training the attention weight of the ith training sample, i =1,2, …, n;
output from fuzzy inference layer
Figure BDA00039093541900000513
The connection weight between the breast diameter and the breast diameter is
Figure BDA00039093541900000514
Figure BDA00039093541900000515
And the connection weight between the volume is
Figure BDA00039093541900000516
Then for all training samples there are:
Figure BDA00039093541900000517
in the formula:
Figure BDA0003909354190000061
represents the j (th) training sample output from the fuzzy inference layer
Figure BDA0003909354190000062
A value;
Figure BDA0003909354190000063
shows the predicted value of the chest diameter of the ith training sample,
Figure BDA0003909354190000064
representing the predicted value of the volume of the ith training sample; j =1,2, …, k, i =1,2, …, n, n is the total strain number of the training samples, and the total number of fuzzy units is k;
carrying out normalization pretreatment on the training sample, and then, training the parameter matrix of the ith forest tree in the sample
Figure BDA0003909354190000065
Contains 7 actually measured forest parameters and forest parameter matrix z i The middle part is sequentially provided with east-west crown breadth from left to rightSouth-north crown breadth, tree height, point cloud density, crown volume, volume and breast diameter, wherein the average value of east-west crown breadth of all training samples is avg 1 The average value of the north-south crown widths of all the training samples is avg 2 The mean tree height of all training samples is avg 3 The average value of the point cloud density of all the training samples is avg 4 Mean value of the crown product of all training samples is avg 5 Volume average of all training samples is avg 6 The mean chest diameter of all training samples is avg 7 The average matrix of each forest parameter in the training sample is avg = [ avg = [ [ avg ] 1 ,…,avg 7 ]，Q i Is calculated as shown in equation (12):
Figure BDA0003909354190000066
in the formula:
Figure BDA0003909354190000067
representing forest parameters matrix z i Middle p-th forest parameter, avg, from left to right p Represents the p-th average value from left to right of the average value matrix avg, and belongs to p ∈ {1,2,3,4,5,6,7}; cosine similarity Sim (z) i Avg) and Euclidean distance Dist (z) i Avg) all represent the internal relation between the forest parameters of the ith tree and the average value avg; tau is an attention weight scaling coefficient and is updated by a pigeon flock optimization module; i =1,2, …, n;
the attention-based connection weight is continuously updated in a back propagation manner, and the updating manner is shown in formula (13):
Figure BDA0003909354190000071
in the formula: the value range of eta is more than 0 and less than 1, which represents the learning efficiency; t is the current iteration number;
the initial value of the attention-based connection weight is randomly given, the connection weight is continuously updated in a back propagation mode, and when the iteration times reach the maximum, the updating is terminated.
As a further improved technical scheme of the invention, the calculation process of the pigeon group optimization module is as follows:
the pigeon flock optimization module is used for optimizing parameters k, alpha and tau in the fuzzy depth network, and the value combination of k, alpha and tau is called as model parameter combination delta [ k, alpha, tau)]In the parameter search space range, k is 0-100 and must be an integer, alpha is 0-30, and tau is 0-10; pigeon swarm optimization module is initially L-set model parameter delta l [k l ,α l ,τ l ](L =1,2, … L), i.e. the pigeon flock optimization module has L sets of model parameters to develop optimization and t max Performing secondary iteration;
the first group of model parameter combination values of the t iteration are
Figure BDA0003909354190000072
The fitness is shown in formula (14):
Figure BDA0003909354190000073
in the formula: t is the current iteration number of the pigeon group; l =1,2, … L; i =1,2, …, n, n is the total number of strains of the training sample;
Figure BDA0003909354190000074
is the chest diameter predicted value of the fuzzy depth network,
Figure BDA0003909354190000075
to blur the volume prediction value of the depth network,
Figure BDA0003909354190000076
is the measured value of the diameter of the breast,
Figure BDA0003909354190000077
measured value of volume of timber;
the search strategy of the model parameter combination is divided into two stages according to the iteration times, the first stage is started, and when the iteration times reach the maximum iteration times t max 80% of the total amount of the catalyst is introduced intoIn the two phases, the specific search strategy is as follows:
the calculation process of the first stage of the search strategy of the model parameter combination is shown as formula (15):
Figure BDA0003909354190000078
in the formula:
Figure BDA0003909354190000079
is a cosine iteration weight term; t is t max The maximum iteration times of the pigeon groups are obtained; ε is a minimal constant; rand (0,1) is [0,1]A random number in between; array of elements
Figure BDA00039093541900000710
Combining model parameters of the l-th group for the t-th iteration
Figure BDA00039093541900000711
Figure BDA00039093541900000712
Is an array of
Figure BDA00039093541900000713
The speed of (d); l =1,2, … L;
Figure BDA0003909354190000081
when the fitness of the optimal parameter combination of the L groups of model parameter combinations does not change for a long time, the L groups of model parameter combinations are sorted in a descending order according to the fitness, and the fitness is higher
Figure BDA0003909354190000082
Group parameters were subject to population variation in the manner shown in equation (16):
Figure BDA0003909354190000083
in the formula:
Figure BDA0003909354190000084
is composed of
Figure BDA0003909354190000085
The updated value of the population variation is,
Figure BDA0003909354190000086
rounding; if it is not
Figure BDA0003909354190000087
In the range of 0 to 100, the content of the organic solvent,
Figure BDA0003909354190000088
otherwise
Figure BDA0003909354190000089
Figure BDA00039093541900000810
Updating when the value is in the range of 0-30, otherwise not updating;
Figure BDA00039093541900000811
updating when the value is in the range of 0-10, otherwise not updating;
Figure BDA00039093541900000812
rand (-1,1) is [ -1,1]A random number within the range;
the calculation process of the second stage of the model parameter combination search strategy is shown in formula (17):
Figure BDA00039093541900000813
in the formula:
Figure BDA00039093541900000814
Figure BDA00039093541900000815
the tentative value of the model parameter of the first group at the time t is obtained; if it is
Figure BDA00039093541900000816
The ith set of model parameters is updated, i.e.
Figure BDA00039093541900000817
Otherwise it is not updated, i.e.
Figure BDA00039093541900000818
l＝1,2,…L；
After each iteration, abandoning part of model parameter combinations with higher fitness values, updating the number L of model parameter groups, and when the iteration number reaches the maximum iteration number t max Or finishing iteration when only one group of model parameter combination is left, outputting the parameter combination with the lowest fitness in the L groups of model parameter combinations at the moment, namely giving the optimal values of the parameters k, alpha and tau, and transmitting the model parameter combination into a fuzzy depth network to finish training.
The invention has the beneficial effects that:
the invention provides an optimized fuzzy depth network to develop a rubber tree parameter prediction model, establish a nonlinear relation between tree parameters, provide a self-adaptive algorithm to enhance the generalization capability of the prediction model to different rubber tree varieties, embed an attention mechanism module to enhance the robustness of the network, and adjust the parameters of the fuzzy depth network in real time by fusing a Pigeon group Optimization algorithm (Pigeon-interpolated Optimization) so as to further improve the prediction precision and learning capability of the model. The rubber forest parameter prediction model designed by the invention can accurately invert complex forest parameters and provides quantitative decision and data support for afforestation and growth fostering of different varieties of rubber trees.
The method is based on the fuzzy depth network, can be used for accurately predicting a complex model, provides a self-adaptive learning algorithm to determine a network structure, searches for optimal parameters by combining a pigeon group optimization algorithm, improves the effect of the self-adaptive algorithm, and adds an attention mechanism to judge abnormal data of a training sample. The forest parameter prediction model further improves the prediction result precision of the key parameters of the rubber forest.
According to the method, the breast diameter and the volume of the single rubber tree are predicted by the partial forest parameters automatically acquired by the airborne laser point cloud through establishing a forest parameter prediction model. The forest parameter prediction model combines the advantages of a fuzzy depth network, an attention mechanism and a pigeon group optimization algorithm, can realize self-adaptive model establishment according to nonlinear relations among forest parameters, weight distribution of abnormal samples and model parameter optimization strategies, is suitable for establishment of most complex forest relations, and has good universality and robustness. Based on the combination of the fuzzy depth network and various artificial intelligence algorithms, the training of automatic model completion of different varieties of rubber trees is completed according to the forest parameter prediction model, and the method is suitable for the prediction of key parameters of the same variety of rubber trees in the same forest land, and is one of the landing applications of artificial intelligence technology in the forest neighborhood.
Drawings
Fig. 1 (a) is a schematic view of the position of the rubber forest 1.
FIG. 1 (b) is a schematic view showing the position of the rubber forest land 2.
FIG. 1 (c) is a schematic view of the position of the rubber forest land 3.
FIG. 1 (d) is a schematic diagram of hot reclamation 628.
Fig. 1 (e) is a schematic diagram of hot reclamation 525.
FIG. 1 (f) is a schematic drawing of heat grinding 72059.
FIG. 1 (g) is a PR107 schematic.
Fig. 2 (a) is a graph showing the result of splitting a single tree in the rubber forest 1.
Fig. 2 (b) is a graph showing the result of splitting a single tree in the rubber forest plot 2.
Fig. 2 (c) is a graph showing the result of splitting a single tree in the rubber forest plot 3.
FIG. 3 is the rubber tree point cloud data for Hot reclamation 628, hot reclamation 525, hot research 72059 and PR107.
Fig. 4 is a general frame diagram of a forest parameter prediction network.
FIG. 5 is a diagram of a process of attention-based weight update.
Fig. 6 (a) is a diagram of the result of iterative optimization of the parameters of the rubber tree network of the pigeon flock optimization module facing hot grinding 72059.
Fig. 6 (b) is a diagram of the result of iterative optimization of network parameters of rubber tree with hot reclamation 525 in the pigeon flock optimization module.
Fig. 6 (c) is a diagram of the result of iterative optimization of the parameters of the rubber tree network for hot reclamation 628 in the pigeon flock optimization module.
Fig. 6 (d) is a diagram of the result of iterative optimization of PR107 rubber tree network parameters by the pigeon flock optimization module.
Fig. 7 is a transformation curve diagram of the fitness of the optimal parameter group of the pigeon flock optimization module group with the increase of the network training times.
Fig. 8 (a) is an iterative graph of the loss values of the PR107 rubber tree during back propagation.
Fig. 8 (b) is an iterative graph of the loss value of the thermal grinding 72059 rubber tree during back propagation.
Fig. 8 (c) is an iterative graph of the loss values of rubber trees in the hot reclamation 525 during back propagation.
Fig. 8 (d) is an iterative graph of the loss values of the rubber tree during back propagation in hot reclamation 628.
FIG. 9 (a) is a comparison analysis chart of the predicted value and the measured value of the chest diameter
Fig. 9 (b) is a graph of comparing the predicted volume value with the measured volume value.
Detailed Description
The following further description of embodiments of the invention is made with reference to the accompanying drawings:
in recent years, airborne laser radars have wide application in forest resource investigation and parameter inversion, but complex forest parameters which are difficult to measure, such as breast diameter, volume and the like, are difficult to obtain due to view angle shielding. In order to solve the problem, the embodiment provides an optimized fuzzy depth network-based accurate prediction method for diameter at breast height of a forest, and firstly, an optimized fuzzy learning network integrating an attention mechanism module is constructed, and a multi-parameter autonomous optimization module based on a pigeon swarm optimization algorithm is added. Secondly, extracting four points from airborne point clouds of three forest lands by combining single plant separation algorithm with artificial forest toneA plurality of growth parameters of rubber trees (hot reclamation 628, hot reclamation 525, hot grinding 72059 and PR 107) of each variety are taken as training sets to be introduced into the deep learning network to optimize the training parameters. Finally, test sets of the four varieties are respectively brought into the trained network to predict the forest key parameters and are compared and analyzed with the actual values, and the results show that the comparison results of the predicted values and the actual measured values of the breast height diameters of the four rubber trees are as follows: RMSE<1.75cm，R 2 >91.42 percent; the comparison results of the predicted values and the measured values of the volume of the four rubber trees satisfy that: RMSE<0.052m 3 ，R 2 >90.14 percent. Compared with the traditional back propagation and radial basis function neural network, the correlation of the forest parameter inversion result obtained by the deep learning network is higher than 4-9%. The embodiment applies the latest artificial intelligence technology to forest land airborne laser point cloud to realize accurate prediction of forest breast diameter and accumulation amount, and can meet large-range rubber forest parameter inversion and operation investigation. The specific steps are set forth below.
1. Materials and data:
1.1, research area and data acquisition:
the study area is located in the rubber tree plantation in delirium city, northwest of Hainan island, from which three multi-variety rubber forest plots are selected in Google map as shown in (a), (b) and (c) of FIG. 1. The terrain of the region is a typical hilly plateau, belongs to tropical monsoon climate, annual average precipitation is 1815 mm, rainy season (5-10 months) accounts for more than 89% of total annual precipitation, annual average temperature is about 23 ℃, and the growth requirement of rubber trees can be met. The rubber tree varieties of hot reclamation 628, hot reclamation 525, hot grinding 72059 and PR107 in the three sample plots are excellent varieties with the characteristics of high and stable yield, strong stress resistance, high tree storage rate and the like, and are planted in large scale in the Hainan area. Wherein, the cold resistance and the wind resistance of the hot reclamation 628 are stronger, and the variety is a good variety with relatively stable yield and wide adaptability; the hot reclamation 525 and 523 has fast growth speed, early maturity and high yield, and is excellent bakelite as well as superior variety. The PR107 has low initial rubber tapping yield, but has high dry rubber content, irritation resistance and high frequency resistance, and the later dry rubber yield is continuously increased, so the strain is an excellent high-yield variety. Therefore, the four rubber trees (with different ages) are selected from the artificial plantation of the rubber trees of various varieties, and as shown in (d), (e), (f) and (g) of fig. 1, the hot reclamation 628, the hot reclamation 525, the hot grinding 72059 and the PR107 are respectively adopted.
The airborne laser radar carrying the Velodyne HDL-32E laser radar sensor can realize a vertical field of view (FOV) from-30.67 degrees to +10.67 degrees, provides a horizontal field of view of 360 degrees, has the working frequency of 10HZ, has the measuring range of 70m and has the measuring precision of +/-2cm. The shooting mode of the airborne laser radar is set as continuous shooting, the flying line route is a pre-programmed 'back-and-forth rectangular parallel' route (as a dotted line in (b) in fig. 1), the flying speed, the flying height and the laser scanning overlap are respectively set to be 10m/s, 30m (higher than a takeoff position) and 30%, the purpose of ensuring the complete and clear rubber tree vertical structure of the acquisition of the branch parameters is achieved, and finally the extracted point cloud is stored in an LAS 1.2 format.
1.2, training sample and test sample:
after point cloud data of the rubber tree forest land is obtained through an airborne laser radar, denoising is carried out through Gaussian filtering, and adverse factors of terrain are eliminated through point cloud ground point filtering (CSF). Then, the present embodiment adopts the existing double-gaussian filter and the single-wood segmentation method with minimized energy function, which has universality in subtropical forests in china. Experiments prove that the method is suitable for rubber woodland, has a good segmentation effect at the junction of the tree crown, and the segmentation results of three rubber tree plots are represented by different colors, which is shown in fig. 2. Fig. 2 (a) is a graph showing the result of splitting a single tree in the rubber forest 1. Fig. 2 (b) is a graph showing the result of splitting a single wood 2 in a rubber forest. Fig. 2 (c) is a graph showing the result of splitting a single tree in the rubber forest plot 3.
1364 rubber trees are shared by three polyclonal multi-variety rubber tree sample plots, 813 trees comprising four varieties (about 200 varieties) are selected from the sample plots by manually and visually inspecting point cloud data of the single trees according to the principle that single branches are as complete as possible, and the specific single morphological characteristics of different varieties are shown in figure 3. FIG. 3 is the point cloud data of rubber trees of different clone varieties. In FIG. 3, from top to bottom; the individual rubber trees in the first and second rows are reclaimed hot 628; the third and fourth rows of individual rubber trees are reclaimed 525 with heat; the single-plant rubber trees in the fifth row and the sixth row are heat ground 72059; the individual rubber tree in the seventh and eighth rows is PR107.
Hot reclamation 628 has trunk, stem, nearly no branch, small crown, broom shape, oval leaf, thick leaf, luster, and three separated leaves. The hot reclamation 525 trees have less deflection, branches with lower height and larger angle, more branches and larger crowns, and are in a multi-head shape. The hot ground 72059 has a softer tree body, is easy to bend, has more drooping branches, has more branches and larger branch angles, and has a larger crown and a fan shape. The PR107 has a straighter tree body and strong wind resistance, the branches have higher height, less branches, more bifurcations and smaller tree crowns, and the tree is broom-shaped, long and elliptical leaves and regular small waves at the leaf edges.
According to the point cloud of the single rubber tree after segmentation, east-west crown breadth, south-north crown breadth, tree height, point cloud density and crown product in forest parameters are automatically obtained according to the following method, and the method is concretely shown as follows. Selecting the maximum distance as an east-west crown width parameter in the east-west direction of each crown point cloud; the parameters of the south-north crown frames are similar to those of the north-south crown frames; the vertical distance between the highest point of the single plant point cloud and the horizontal plane is a tree height parameter; the total number of the individual plant point clouds is divided by the projection area of the crown, namely the point cloud density; and (3) calculating the convex hull volume of the single plant crown point cloud by using an Alphashape method, namely the crown volume parameter.
Because forest parameters such as the breast diameter and the volume of timber of the rubber tree are not easy to directly acquire from the point cloud of the single plant on the airplane, the breast diameter and the volume of timber of the rubber tree are acquired by adopting an artificial surveying and mapping mode, and the specific method is as follows. And (3) acquiring the perimeter of the trunk at the trunk part of the rubber tree 1.3m away from the ground through a measuring tape, and further obtaining the breast diameter parameter. Using a tree measurement method to obtain the volume parameters of the stumpage, and selecting the diameter of the upper part of a rubber trunk as
Figure BDA0003909354190000111
The position is taken as a shape point, D is the diameter at breast height of the rubber tree, and the length from the shape point to the treetop is measured to be H t Measuring the cross-sectional area S of the breast diameter D Measuring the height H of the rubber tree, using the following equationAnd (4) obtaining the volume of the rubber tree.
Figure BDA0003909354190000112
In the formula: r is the dry form index; s. the vl Is the volume of the rubber tree. The parameters and the manual mapping are automatically obtained from the point cloud of the single plant, the forest parameters of about 200 plants of each variety are obtained, a training set and a testing set are divided, and the forest parameters and the training samples of four rubber trees, namely hot reclamation 628, hot reclamation 525, hot grinding 72059 and PR107, are shown in Table 1.
Table 1 shows the parameters and training samples for different varieties of rubber trees in the study plot:
Figure BDA0003909354190000113
2. forest parameter prediction model:
2.1, designing a model overall architecture:
the forest parameters such as breast diameter and volume are not easy to obtain from the point cloud of the single plant on the airplane, and the establishment of a forest parameter prediction model to obtain the parameters such as breast diameter and volume is very important by finding the corresponding relation between the forest parameters. The artificial forests of the rubber trees of multiple varieties have the same conditions such as soil and climate, the forest parameters of the same rubber tree variety are normally distributed near the average value of the forest parameters of the variety, but the forest parameters of part of individuals have larger difference with other groups due to factors such as rubber tree necrosis and intraspecific competition, and therefore a prediction model is required to be capable of automatically identifying the different individuals. In the face of the existence of a plurality of varieties in the artificial rubber tree plantation, the growth forms of different varieties are different, so self-adaptive learning is needed during parameter prediction; in addition, the parameter optimization of the prediction model also greatly influences the prediction effect.
Comprehensive research on a common neural network prediction model shows that the attention mechanism effectively enhances the anti-interference capability of the prediction model, and the forest parameter attention mechanism is added to improve the robustness of the forest parameter prediction model and improve the accuracy of the prediction model.
The membership function of the fuzzy subset in the fuzzy depth network usually adopts a Gaussian function, the center and the variance of the fuzzy subset are determined in a self-adaptive mode, and the learning capability of the fuzzy depth network is embodied. Therefore, the present embodiment proposes a DPKM algorithm combining Density Peaks Clustering (DPC) and K-Means algorithm to determine the centers of the membership functions, and proposes an algorithm for adaptively determining the variance according to the euclidean distance between the centers. The Pigeon swarm Optimization algorithm (PIO) can effectively solve the parameter Optimization problem of the fuzzy depth network, so that a forest parameter prediction model fusing a Pigeon swarm Optimization module and the fuzzy depth network is designed.
The overall framework of the forest parameter prediction network provided by the embodiment is shown in fig. 4 and is divided into a fuzzy depth network and a pigeon swarm optimization module. And the pigeon group optimization module updates parameters of the fuzzy depth network, outputs a predicted value after the network finishes training and returns the predicted value to the pigeon group module to serve as a precondition for solving the parameter fitness value given by the pigeon group module. After the pigeon group module finishes the optimal parameter search of the fuzzy depth network, the fuzzy depth network finishes self-adaptive training according to the optimal parameters, and a final forest parameter prediction model is established. The fuzzy depth network comprises an adaptive fuzzy layer, a fuzzy inference layer and a weight updating layer based on attention from input to output. x is the number of 1 、x 2 、x 3 、x 4 、x 5 The input attributes of the self-adaptive fuzzy layer are east-west crown width, south-north crown width, tree height, point cloud density and crown product, each input attribute has the same fuzzy subset quantity, and the center and variance of the membership function of the fuzzy subset are determined by a self-adaptive algorithm. Sample introduction of input attribute into self-fuzzy subset output membership value
Figure BDA0003909354190000121
Figure BDA0003909354190000122
x u Corresponding degree of membership of
Figure BDA0003909354190000123
u ∈ {1,2,3,4,5}, j =1, …, k, and the total number of fuzzy subsets is k. The product reasoning determines fuzzy units of a fuzzy reasoning layer, the membership degree h is transmitted into the fuzzy units, one unit corresponds to one output, and the output value is
Figure BDA0003909354190000124
j =1, …, k, and the total number of blur units is k.
Figure BDA0003909354190000125
Transmitting into attention-based weight updating layer, continuously updating connection weight by back propagation, and finally outputting predicted value y by weighting operation 1 、y 2 Wherein y is 1 Is a predicted value of breast diameter, y 2 The volume prediction value is obtained.
2.2, adaptive fuzzy layer:
input attribute x u In the adaptive fuzzy layer, there are k fuzzy subsets, u is ∈ {1,2,3,4,5}, and the adaptive fuzzy layer has 5 × k fuzzy subsets. Training sample
Figure BDA0003909354190000126
Input attribute x in the incoming adaptive fuzzy layer u The jth fuzzy subset to which the system belongs outputs the degree of membership
Figure BDA0003909354190000127
As shown in equation (2).
Figure BDA0003909354190000128
In the formula: x is the number of u The center and variance of the j fuzzy subset in the adaptive fuzzy layer are respectively
Figure BDA0003909354190000129
x u Training samples according to the present embodiment proposes the DPKM algorithm to perform clustering,
Figure BDA0003909354190000131
is a clustering center; according to
Figure BDA0003909354190000132
Euclidean distance therebetween, and
Figure BDA0003909354190000133
sample density and self-adaptive variance determination inside DPKM cluster
Figure BDA0003909354190000134
j =1, …, k, the total number of fuzzy subsets of the input attribute is k, and k is updated by the pigeon flock optimization module, i =1, …, n, n is the total number of rubber tree plants of the training sample.
DPKM algorithm according to x u Solving for n training samples
Figure BDA0003909354190000135
Local density of
Figure BDA0003909354190000136
And distance
Figure BDA0003909354190000137
And calculating according to the above two
Figure BDA0003909354190000138
As a likelihood value of the initial center of clustering
Figure BDA0003909354190000139
Then, clustering is carried out, wherein the specific method is shown as follows, i =1, …, n, u epsilon {1,2,3,4,5}.
Figure BDA00039093541900001310
In the formula:
Figure BDA00039093541900001311
is that
Figure BDA00039093541900001312
Local density of d u Is x u U e {1,2,3,4,5}, i =1,2, …, n, m =1,2, …, n. Computing
Figure BDA00039093541900001313
Need to consider x u N training samples and
Figure BDA00039093541900001314
by a cutoff distance d u Determining
Figure BDA00039093541900001315
In the truncated distance amplification neighborhood of equation (3), training sample pairs
Figure BDA00039093541900001316
The influence of local density of the neighborhood is reduced, and the effect of sample points outside the neighborhood is reduced; d u Let x be selected u In training samples
Figure BDA00039093541900001317
The ratio of the number in the neighborhood to the total number n of plants is 1 to 2 percent. x is the number of u The n training samples are sorted in descending order according to the local density, and the local density is greater than
Figure BDA00039093541900001318
The minimum Euclidean distance between any two samples in the training samples is taken as
Figure BDA00039093541900001319
Is a distance of
Figure BDA00039093541900001320
As shown in the following formula.
Figure BDA00039093541900001321
In the formula:
Figure BDA00039093541900001322
is that
Figure BDA00039093541900001323
Local density of
Figure BDA00039093541900001324
m＝1,2,…n；
Figure BDA00039093541900001325
Is composed of
Figure BDA00039093541900001326
And with
Figure BDA00039093541900001327
U e {1,2,3,4,5}. In particular, if
Figure BDA00039093541900001328
The local density of (a) is the greatest,
Figure BDA00039093541900001329
and m ≠ i.
When in use
Figure BDA00039093541900001330
Local density of
Figure BDA00039093541900001331
And distance
Figure BDA00039093541900001332
After determination, calculating
Figure BDA00039093541900001333
As an input attribute x u Likelihood value of initial center when performing DPKM clustering
Figure BDA00039093541900001334
As shown in equation (5).
Figure BDA00039093541900001335
In the formula: max (rho) u )、min(ρ u ) Are respectively rho u Maximum and minimum values of the middle n local densities; max (delta) u )、min(δ u ) Are respectively delta u The maximum value and the minimum value of the n distances. The purpose of equation (5) is to map the local densities ρ of different scales u And a distance delta u Normalized to the product of the same scale. Rho u Refers to the local density of the u-th input attribute, which has n local densities. Delta u Refers to the distance of the u-th input attribute, which has n distances.
To this end, x u N training samples
Figure BDA0003909354190000141
According to
Figure BDA0003909354190000142
Sorting the values in descending order, and taking the first k sample points as x u Carrying out DPKM clustering on the n training samples to obtain initial central points; second, calculate input attribute x u The Euclidean distance between the n training samples and the k category centers is distributed to the category where the nearest center is located according to the minimum distance distribution principle; thirdly, after the n samples are distributed for one time, calculating the mean value of each category, updating the category center, and repeating the two steps (namely the second step and the second step) until the change of the category center is smaller than the set error; completing DPKM clustering, wherein the center of k clusters is x u Fuzzy subset center of
Figure BDA0003909354190000143
j＝1,…,k。
When inputting attribute x u Fuzzy subset center of
Figure BDA0003909354190000144
And
Figure BDA0003909354190000145
belonging DPKM cluster
Figure BDA0003909354190000146
After establishment, x is calculated according to the following adaptive algorithm u Fuzzy subset variance of
Figure BDA0003909354190000147
First calculate x u Center of the k fuzzy subsets center
Figure BDA0003909354190000148
As shown in equation (6).
Figure BDA0003909354190000149
Then solve for clustering
Figure BDA00039093541900001410
Inner sample of
Figure BDA00039093541900001411
And
Figure BDA00039093541900001412
average euclidean distance of
Figure BDA00039093541900001413
Is to calculate the variance
Figure BDA00039093541900001414
The precondition of (2) is as shown in equation (7).
Figure BDA00039093541900001415
In the formula:
Figure BDA00039093541900001416
to represent
Figure BDA00039093541900001417
Inner sample of (2)
Figure BDA00039093541900001418
And
Figure BDA00039093541900001419
the Euclidean distance;
Figure BDA00039093541900001420
is composed of
Figure BDA00039093541900001421
Inner sample of
Figure BDA00039093541900001422
The number of the cells.
Figure BDA00039093541900001423
In the formula:
Figure BDA00039093541900001424
is composed of
Figure BDA00039093541900001425
A width factor of (d); alpha is a variance scaling coefficient and is updated by a pigeon group optimization module;
Figure BDA00039093541900001426
as an input attribute x u The maximum distance between every two of the k fuzzy subsets; u ∈ {1,2,3,4,5}, j =1,2, …, k.
The initial fuzzy subset number k and the variance scaling coefficient alpha are given by the layer, and after the pigeon group optimization module transmits new values of k and alpha, the model structure of the self-adaptive fuzzy layer is changed.
2.3, fuzzy inference layer:
the fuzzy depth network carries out fuzzification processing on different input attributes in sequence so as to carry out fuzzy reasoning on complex nonlinear relations among forest parameters, and the fuzzy depth network is very effective in processing complex models which are difficult to be precise, so that the defect of the traditional neural network is overcome.
The membership degree is input into the fuzzy inference layer, a product inference method is adopted to establish fuzzy units of the fuzzy inference layer, and the output value of the jth unit of the ith training sample is calculated
Figure BDA0003909354190000151
As a fuzzy unit output.
Figure BDA0003909354190000152
In the formula:
Figure BDA0003909354190000153
the product of the membership degrees of the jth fuzzy subset of the ith training sample in different input attributes;
Figure BDA0003909354190000154
after normalization processing is
Figure BDA0003909354190000155
u belongs to {1,2,3,4,5}, i =1, …, n, j =1,2, …, k, n is the total number of rubber tree plants in the training sample, and the total number of fuzzy units is k.
2.4, weight updating layer based on attention:
the initial value of the attention-based connection weight w is randomly given, and the updating is continuously performed in a back propagation manner as shown in fig. 5, and the updating is terminated when the iteration number reaches the maximum. Of a fuzzy inference layer
Figure BDA0003909354190000156
The value is used as an input to the present layer,
Figure BDA0003909354190000157
and carrying out weighted operation on the connection weight w to output a predicted value y, thereby realizing defuzzification calculation of the fuzzy depth network.
The loss function integrates the predicted value y and the measured value
Figure BDA0003909354190000158
The attention weighting value Q is shown in equation (10).
Figure BDA0003909354190000159
In the formula: y is 1 Is the predicted value of breast diameter, y 2 The predicted value of the volume is taken as the volume;
Figure BDA00039093541900001510
is composed of
Figure BDA00039093541900001511
The corresponding measured value; q i For the attention weight of the i-th training sample, i =1,2, …, n. Fitness value
Figure BDA00039093541900001512
The connection weight between the breast diameter and the breast diameter is
Figure BDA00039093541900001513
Figure BDA00039093541900001514
The connection weight between the volume and the volume is
Figure BDA00039093541900001515
Then for all training samples there are:
Figure BDA0003909354190000161
in the formula:
Figure BDA0003909354190000162
represents the j (th) training sample output from the fuzzy inference layer
Figure BDA0003909354190000163
A value;
Figure BDA0003909354190000164
shows the predicted value of the chest diameter of the ith training sample,
Figure BDA0003909354190000165
representing the predicted value of the volume of the ith training sample; j =1,2, …, k, i =1,2, …, n, n is the total strain number of the training samples, and the total number of fuzzy units is k.
The attention mechanism is essentially weight distribution, wherein the weight Q is distributed to the rubber tree sample in the loss function, and the rubber tree training sample in the abnormal growth state has the anti-interference capability. In order to eliminate the influence of dimensions among different forest parameters, the training samples are subjected to normalization preprocessing. Then, training parameter matrix of ith forest tree in sample
Figure BDA0003909354190000166
The total 7 actually measured forest parameters are east-west crown width, south-north crown width, tree height, point cloud density, crown volume, volume and breast diameter, and the average matrix of each forest parameter in the training sample is avg = [ ] 1 ,…,avg 7 ]，Q i From z in the training sample i The intrinsic connection with avg is obtained.
Figure BDA0003909354190000167
In the formula:
Figure BDA0003909354190000168
representing forest parameters matrix z i Middle p-th forest parameter, avg, from left to right p Represents the pth average value from left to right of the average value matrix avg, and belongs to {1,2,3,4,5,6,7}; cosine similarity Sim (z) i Avg) and Euclidean distance Dist (z) i Avg) represents the intrinsic relationship between the parameter of the ith rubber tree and the average value avg; tau is an attention weight scaling coefficient and is updated by a pigeon group optimization module; i =1,2, …, n.
The connection weight w is continuously updated in a back propagation manner, and the updating manner is shown in formula (13).
Figure BDA0003909354190000171
In the formula: the value range of eta is more than 0 and less than 1, which represents the learning efficiency; and t is the current iteration number.
Meanwhile, given an initial attention weight scaling coefficient tau, the forest parameter prediction effect of the fuzzy depth network is changed when the pigeon group optimization module updates the value tau every time.
2.5, a pigeon group optimizing module:
the module is used for optimizing key parameters k, alpha and tau in the fuzzy deep learning network, wherein k is the total fuzzy subset number of the self-adaptive fuzzy layer, alpha is a variance scaling coefficient of a formula (8), and tau is an attention weight scaling coefficient of a formula (12). The value combination of k, alpha and tau is called model parameter combination delta [ k, alpha and tau ]]In the parameter search space range, k is 0-100 and must be an integer, α is 0-30, and τ is 0-10. Pigeon swarm optimization module is initially L-set model parameter delta l [k l ,α l ,τ l ](L =1,2, … L), i.e., the module has L sets of model parameters to develop optimization and t max A sub-iteration, a tth iteration according to the search strategy pair of the text
Figure BDA0003909354190000172
Updating the values of the L groups of model parameters, and searching the optimal parameter combination delta in the L groups of model parameters after the iteration is terminated l And the data is transmitted into the fuzzy deep network to finish the training of the fuzzy deep network.
First, updating the parameters k, α of the adaptive fuzzy layer results in the output value of the layer
Figure BDA0003909354190000173
Updating of (3); secondly, updating the attention mechanism coefficient τ affects the calculation of the attention weight Q, and further affects the iteration of the connection weight w by the loss function, i.e. formula (10)(ii) a Finally, the predicted value of the fuzzy deep learning network is followed
Figure BDA0003909354190000174
And the update of the connection weight w. In this embodiment, the predicted value and the measured value of the completed fuzzy deep training model are used to calculate the fitness of the model parameter combination, and as shown in formula (14), the lower the fitness value is, the closer the group of parameters is to the optimal model parameter combination.
Figure BDA0003909354190000175
In the formula: the first group of model parameter combination values of the t iteration are
Figure BDA0003909354190000176
t is the current iteration number of the pigeon group; l =1,2, … L; i =1,2, …, n, n is the total strain number of the training sample;
Figure BDA0003909354190000177
the predicted values of the breast diameter and the volume of the fuzzy depth network,
Figure BDA0003909354190000178
is prepared by reacting with
Figure BDA0003909354190000179
The corresponding measured value.
The search strategy of the model parameter combination is divided into two stages according to the iteration times, the first stage is started, and the second stage is started when the iteration times reach 80% of the maximum iteration times, and the specific search strategy is shown as follows.
In the first stage of the model parameter combination search strategy, the present embodiment provides a cosine iteration weight term and a population variation idea to help an individual jump out of a local optimal solution, so as to give an optimal model parameter combination, and thus, the fuzzy deep network completes training. The cosine iteration weight term is added, so that the global search capability is emphasized more in the initial iteration stage, the local search capability is stronger in the later iteration stage, and the actual iteration requirement is met, as shown in the following.
Figure BDA0003909354190000181
In the formula:
Figure BDA0003909354190000182
is a cosine iteration weight term; t is t max The maximum iteration times of the pigeon groups are obtained; ε is a minimal constant; rand (0,1) is [0,1]A random number in between; the value of the l group of model parameter combination of the t iteration is
Figure BDA0003909354190000183
Figure BDA0003909354190000184
Rounding is needed when the value is updated;
Figure BDA0003909354190000185
is an array of
Figure BDA0003909354190000186
The speed of (d); l =1,2, … L;
Figure BDA0003909354190000187
in order to further enhance the capability of the PIO algorithm to jump out of the local optimal solution, when the fitness of the optimal parameter combination of the L sets of model parameter combinations does not change for a long time, it is described that the forest parameter prediction model of the embodiment falls into a local extremum. At the moment, the L groups of model parameter combinations are sorted in a descending order according to the fitness,
Figure BDA0003909354190000188
l =1,2, … L, with higher adaptability
Figure BDA0003909354190000189
Group parameters were subject to population variation in the manner shown below.
Figure BDA00039093541900001810
In the formula:
Figure BDA00039093541900001811
is composed of
Figure BDA00039093541900001812
The updated value of the population variation is,
Figure BDA00039093541900001813
rounding; if it is not
Figure BDA00039093541900001814
In the range of 0 to 100, the content of the organic solvent,
Figure BDA00039093541900001815
otherwise
Figure BDA00039093541900001816
According to
Figure BDA00039093541900001817
In the manner of the update of (2),
Figure BDA00039093541900001818
updating when the value is in the range of 0-30 and 0-10 respectively, or not updating;
Figure BDA00039093541900001819
rand (-1,1) is [ -1,1]Random numbers in the range.
In the second stage of the model parameter combination search strategy, the update strategy of the model parameters is adjusted, and delta is assumed c (t) is the center position of all model parameter combinations at time t, towards which the parameter set flies.
Figure BDA0003909354190000191
In the formula:
Figure BDA0003909354190000192
Figure BDA0003909354190000193
the tentative value of the model parameter of the first group at the time t is obtained; if it is
Figure BDA0003909354190000194
The first set of model parameters is updated
Figure BDA0003909354190000195
Otherwise, it is not updated
Figure BDA0003909354190000196
L =1,2, … L. After each iteration, part of model parameter combinations with higher fitness values are abandoned, and the number L of model parameter groups is updated, so that the better model parameter combinations are reserved, and the convergence of the algorithm is ensured. When the number of iterations reaches the maximum number of iterations t max Or ending iteration when only one group of model parameters is left, outputting the parameter combination with the lowest fitness in the L groups of model parameter combinations at the moment, namely giving the optimal values of the parameters k, alpha and tau, and transmitting the model parameter combination into the fuzzy depth network to finish training.
3. Results and discussion:
3.1, training and testing results of the pigeon flock optimization module:
the training and testing of the forest parameter prediction model are carried out on a Windows 10-64-bit server which is loaded with an AMD Ryzen 7 4800H CPU@2.9GHZ processor and a 16 GB-RAM. In the forest parameter prediction model constructed in this embodiment, the maximum iteration number of the weight update layer is set to 200, and the learning efficiency η is set to 8; the total number L of model parameter combinations of the pigeon group optimization module is set to be 32, and the maximum iteration turns t max Set to 50.
The 32 sets of model parameters in the initial round are subjected to random value taking and are uniformly distributed in a parameter search space, the values of k, alpha and tau in the model parameter combination are continuously updated along with the continuous iteration of the pigeon group module, the optimal model parameter combination results of different varieties of training sets in different stages of iteration are shown in table 2, and the 32 sets of parameters of each variety gradually converge towards the optimal array, which shows that the pigeon group module can adaptively learn the optimal model parameter combination of rubber trees of different varieties, as shown in fig. 6. Fig. 6 (a) is a diagram of the result of iterative optimization of the parameters of the rubber tree network of the pigeon flock optimization module facing hot grinding 72059. Fig. 6 (b) is a diagram of the result of iterative optimization of the parameters of the rubber tree network with hot reclamation 525 in the pigeon flock optimization module. Fig. 6 (c) is a diagram illustrating the result of iterative optimization of the parameters of the rubber tree network with respect to hot reclamation 628 by the pigeon flock optimization module. Fig. 6 (d) is a diagram of the result of iterative optimization of PR107 rubber tree network parameters by the pigeon flock optimization module.
Table 2 shows the optimal results of the model parameter combinations of the pigeon flock optimization module at different stages:
Figure BDA0003909354190000197
each iteration process of the pigeon swarm module for searching the optimal model parameter combination of the training set has 32 sets of model parameters, wherein the lowest fitness is the optimal model parameter combination of the current iteration process, and a fitness curve formed by the optimal model parameter combinations of different iteration stages is shown in fig. 7. The fitness of the optimal model parameter combination shows a downward trend, which shows that the forest parameter prediction model of the embodiment is a global optimization process. The fitness curves of different varieties of training sets are obviously reduced in the first 30 epochs, which shows that the parameters of the forest parameter prediction model are close to the optimal array rapidly. Model parameters are sequentially transmitted into corresponding modules of the fuzzy depth network, and correlation coefficients in the neural network are adjusted on the basis that the fuzzy depth network adaptively constructs a network structure according to training samples, so that the optimal parameter set in the initial round can also reach a better fitness value. After 50 epochs, fitness values of the hot grinding 72059, the hot reclamation 525, the hot reclamation 628 and the PR107 training samples respectively converge to 0.025, 0.022, 0.016 and 0.015, which indicates that the forest parameter prediction model constructed in the embodiment has the capability of accurate parameter prediction.
And after the pigeon group optimization module confirms the optimal model parameter combination of different varieties, the optimal model parameter combination is transmitted into a fuzzy depth network to complete the training of the forest parameter prediction models of all varieties. At this time, in the attention-based weight value updating layer, the loss value E in the training process is as shown in fig. 8. Fig. 8 (a) is an iterative graph of the loss value of the PR107 rubber tree during the back propagation. Fig. 8 (b) is an iterative graph of the loss value of the thermal grinding 72059 rubber tree in the back propagation process. Fig. 8 (c) is an iterative graph of the loss values of rubber trees in the hot reclamation 525 during back propagation. Fig. 8 (d) is an iterative graph of the loss values of the rubber tree during back propagation in hot reclamation 628. In order to improve the training efficiency of the prediction model in this embodiment, a small-Batch Gradient (Mini-Batch Gradient) method is adopted in the attention-based weight update layer for back propagation, which results in local oscillation of the regression loss value. However, as the learning process is continuously iterated, the loss value E generally tends to decrease, which indicates that the fuzzy deep network of the embodiment has better convergence. After 100 iterations, the loss values E of PR107, thermal research 72059, thermal reclamation 525 and thermal reclamation 628 converge to 0.00176, 0.00349, 0.00345 and 0.00072 respectively, which indicates that the fuzzy depth network constructed by the embodiment has better forest parameter prediction capability.
3.2, comparing with the prior method:
based on the prediction model and the conventional method of the embodiment, the prediction results of the parameters of the breast diameter and the volume of the rubber tree are shown in table 3. The BP (Back Propagation) neural network is a method based on a multi-layer feedforward neural network, and the determination of the network structure depends on experience and trial and error, and excitation functions are global and interfere with each other, so that the problem of local minimum is easily caused. The RBF (Radial Basis Function) neural network and the BP are both suitable for establishing a nonlinear model, but the local excitation Function of the RBF overcomes the problem of mutual interference of the BP global excitation Function, and for a new training set, only the number of hidden layer neuron nodes and the connection weight are required to be changed, so that the learning speed is greatly improved compared with that of a BP algorithm, the convergence is easier to ensure, and the RBF is easy to obtain a better result. GRNN (General Regression Neural Network) is a kind of radial basis function Neural Network, and compared with the traditional radial basis function Network, a summation layer is added between an implicit layer and an output layer, and the method is more advantageous than RBF in the aspects of less sample data, unstable data and the like. However, the RBF and GRNN neural networks often determine the network structure through trial and error and empirical formulas, relying on prior experience; meanwhile, the method lacks a mechanism for judging abnormal data in the training sample, and the robustness of the neural network is reduced. The method is based on the fuzzy depth network, can be used for accurately predicting a complex model, provides a self-adaptive learning algorithm to determine a network structure, searches for optimal parameters by combining a pigeon group optimization algorithm, improves the effect of the self-adaptive algorithm, and adds an attention mechanism to judge abnormal data of a training sample. Table 3 lists the comparison results of the four methods for predicting the breast diameter and the volume, and it is shown in the table that the method of the present embodiment obtains better quantitative results on three indexes, namely, the coefficient of determination (R2), the Root Mean Square Error (RMSE), and the Mean Absolute Percentage Error (MAPE), and thus, the forest parameter prediction model of the present embodiment further improves the accuracy of the prediction result of the key parameters of the rubber forest.
Table 3 shows the forest parameter prediction results for different methods:
Figure BDA0003909354190000201
3.3, forest parameter prediction result analysis:
after the fuzzy depth network establishes the optimal model parameters through the pigeon flock optimization module, the fuzzy depth network adaptively establishes forest parameter prediction models of various varieties according to different rubber tree variety training sets. Table 4 shows the actual measurements of the breast diameters and volume sizes of four rubber trees including hot reclamation 628, hot reclamation 525, hot grinding 72059 and PR107 and the predicted values of this example. At the same time, by comparing the index R 2 RMSE and MAPE quantification analysis the effectiveness of the method of this example, and fig. 9 is a comparison of specific parameters.
Table 4 shows the comparison between the forest growth parameters obtained by the method of the present embodiment and the actual measured values:
Figure BDA0003909354190000211
note that: (F): actual measured values; (O): the method herein.
Fig. 9 shows the results of parameter prediction of breast diameter and volume, respectively, and the predicted values and measured values of the parameters of the four rubber trees form experimental points which are uniformly distributed near the 45 ° regression line and are in a linear relationship.
Fig. 9 (a) shows the comparison result of the predicted value and the measured value of the chest diameter of the four rubber trees obtained by the method of the present embodiment. Wherein the comparison results of hot reclamation 525 and hot grinding 72059 are respectively (R) 2 =92.24%, RMSE =1.70cm, mape = 5.08%) and (R) 2 =91.42%, RMSE =1.75cm, mape = 5.10%). The study model of this example predicted better results in terms of breast diameters for hot reclamation 628 and PR107, relative to the first two breeds, which were (R), respectively 2 =94.31%, RMSE =1.44cm, mape = 4.87%) and (R 2 =93.87%, RMSE =1.48cm, mape = 5.03%). The method is mainly characterized in that due to the fact that the hot grinding 72059 rubber garden trees have a wind damage inclination phenomenon, adjacent rubber trees are shielded mutually to cause incomplete point cloud data acquisition, and further the final parameter prediction result is influenced. The growth form of hot reclamation 525 is more complex, the parts generating branches are lower, the number of branches is more and dense, and the breast diameter parameters of different rubber trees are more different. And the hot reclamation 628 and PR107 have strong wind resistance, are not easy to fall down, have simple growth form and small branch number, and have complete acquired branch data, higher point cloud quality and better prediction precision.
Fig. 9 (b) shows the comparison of the predicted volume values and the actual measured volume values of the four rubber trees. Wherein heat grinding 72059 (R) 2 ＝91.25％，RMSE＝0.050m 3 MAPE = 6.06%) and hot reclamation 525 (R) 2 ＝90.14％，RMSE＝0.052m 3 MAPE = 8.19%) relative hot reclamation of RMSE 628 (R) 2 ＝93.88％，RMSE＝0.027m 3 MAPE = 5.02%) and PR107 (R) 2 ＝93.73％，RMSE＝0.028m 3 MAPE = 5.33%) is significantly higher. This phenomenon can be explained by the fact that the hot reclamation 628 and PR107 have a small number of branches, and different trees are in the same forest land and in the same atmosphereThe difference of the volume parameters is smaller under the same variety; the crowns of hot reclamation 628 and PR107 are small, the influence of the shielding of adjacent trees of the same variety is small, and the obtained tree parameters are more accurate.
4. And (3) ending:
by establishing a forest parameter prediction model, the breast diameter and the volume of the single rubber tree are predicted by partial forest parameters automatically acquired by airborne laser point cloud. The prediction model combines the advantages of a fuzzy depth network, an attention mechanism and a pigeon group optimization algorithm, can realize self-adaptive model establishment according to nonlinear relations among forest parameters, weight distribution of abnormal samples and model parameter optimization strategies, is suitable for establishment of most complex forest relations, and has good universality and robustness. Compared with the three prediction methods of BP, RBF and GRNN, the prediction model of the embodiment obtains better results in predicting the chest diameter and the volume, and the RMSE is respectively 1.59 +/-0.15 cm and 0.040 +/-0.013 m 3 . The experiment result shows that the forest parameter prediction model of the embodiment can obtain better prediction effect on various rubber trees, the average MAPE of the predicted breast diameter of the four rubber trees is 5.02% of disturbance, the average MAPE of the predicted volume is 6.15% of disturbance, the single forest parameter of the artificial forest can be effectively obtained, and the prediction model is superior to the traditional prediction model in the experiment sample. In the prediction research of forest parameters, the differences of different rubber tree varieties can find out the rules from the influences of natural environment on rubber tree sample plots, incomplete acquisition of airborne point clouds caused by mutual shielding among forests and growth morphological characteristics of different varieties. Based on the combination of the fuzzy depth network and various artificial intelligence algorithms, the training of the automatic model completion of the rubber trees of different varieties is completed according to the wood parameter prediction model, and the method is suitable for the prediction of key parameters of the rubber trees of the same variety in the same forest land, and is one of the landing applications of the artificial intelligence technology in the forest neighborhood.

Claims (8)

1. The method for accurately predicting the diameter at breast height volume of the forest based on the optimized fuzzy depth network is characterized by comprising the following steps of:
step 1: acquiring forest land point cloud data through an airborne laser radar;
step 2: denoising the point cloud data, filtering the point cloud data by adopting a point cloud ground point filtering method, and performing single-tree segmentation on the filtered point cloud data;
and 3, step 3: acquiring forest parameters of the single tree, including east-west crown width, south-north crown width, tree height, point cloud density and crown product, according to the point cloud of the single tree after the division;
and 4, step 4: acquiring the diameter at breast height and volume of the corresponding tree by adopting an artificial mapping method;
and 5: taking east-west crown breadth, south-north crown breadth, tree height, point cloud density, crown product, breast diameter and volume of timber of a plurality of trees as a training sample data set;
step 6: establishing a forest parameter prediction network, wherein the forest parameter prediction network comprises a fuzzy depth network and a pigeon swarm optimization module, training the forest parameter prediction network by adopting a training sample data set, the input of the forest parameter prediction network is east-west crown width, south-north crown width, tree height, point cloud density and crown product of a tree, the output is breast diameter and volume of the tree, a predicted value is output after the training of the fuzzy depth network is finished, the predicted value is output to the pigeon swarm optimization module, the parameters of the fuzzy depth network are updated by the pigeon swarm optimization module, after the searching of the optimal parameters of the fuzzy depth network is finished by the pigeon swarm optimization module, the fuzzy depth network completes self-adaptive training according to the optimal parameters, and a final forest parameter prediction model is established;
and 7: acquiring point cloud data of the forest land to be measured, acquiring east-west crown width, south-north crown width, tree height, point cloud density and crown product of each tree in the forest land to be measured according to the methods in the step 2 and the step 3, and inputting the east-west crown width, the south-north crown width, the tree height, the point cloud density and the crown product of the tree into a final forest parameter prediction model to obtain predicted values of the breast diameter and the volume of the corresponding tree.
2. The method for accurately predicting the diameter at breast height volume of the tree based on the optimized fuzzy depth network as claimed in claim 1, wherein the method for obtaining the tree parameters of the single tree in the step 3 comprises:
selecting the maximum distance from the east-west crown width in the east-west direction of the single tree crown point cloud; selecting the maximum distance as the south-north crown breadth in the south-north direction of the single tree crown point cloud; the vertical distance between the highest point of the single tree crown point cloud and the horizontal plane is the tree height; dividing the total number of the individual point clouds by the projection area of the crown to obtain point cloud density; and calculating the convex hull volume of the single tree crown point cloud, wherein the convex hull volume is the crown volume.
3. The method for accurately predicting the diameter at breast height and volume of the forest based on the optimized fuzzy depth network as claimed in claim 2, wherein the method for acquiring the diameter at breast height and volume of the tree in the step 4 comprises the following steps:
acquiring the perimeter of a trunk at the trunk part of the tree 1.3m away from the ground through a measuring tape, wherein the perimeter is the breast-height diameter of the tree;
selecting the diameter of the upper part of the trunk of the tree as
Figure FDA0003909354180000011
Taking the point as a shape point, D is the diameter at breast height of the tree, and the length from the shape point to the treetop is measured to be H t Measuring the cross-sectional area of the breast diameter to be S D Measuring the height of the tree as H, and then the volume of the tree is as follows:
Figure FDA0003909354180000021
in the formula: r is the dry form index; s vl Is the volume of the tree.
4. The method for accurately predicting the diameter volume of the breast of the forest based on the optimized fuzzy depth network as claimed in claim 1, wherein the fuzzy depth network in the step 6 comprises an adaptive fuzzy layer, a fuzzy inference layer and an attention-based weight value updating layer in sequence from input to output.
5. The optimized fuzzy depth network-based forest diameter at breast height volume accurate prediction method according to claim 4, wherein the calculation process of the adaptive fuzzy layer is as follows:
input attribute x u At the adaptive fuzzy level, there are k fuzzy subsets, u ∈ {1,2,3,4,5}, where x 1 、x 2 、x 3 、x 4 、x 5 The method comprises the steps that input attributes of a self-adaptive fuzzy layer are respectively east-west crown breadth, south-north crown breadth, tree height, point cloud density and crown product in sequence, and each input attribute has k fuzzy subsets; the adaptive fuzzy layer has 5 x k fuzzy subsets; training sample
Figure FDA0003909354180000022
Input attribute x in the incoming adaptive fuzzy layer u The jth fuzzy subset to which the system belongs outputs the degree of membership
Figure FDA0003909354180000023
As shown in equation (2):
Figure FDA0003909354180000024
in the formula: x is the number of u Centered on the jth fuzzy subset of the adaptive fuzzy layer
Figure FDA0003909354180000025
x u The variance of the jth fuzzy subset at the adaptive fuzzy level is
Figure FDA0003909354180000026
The total number of fuzzy subsets of the input attributes is k, the k is updated by a pigeon group optimization module, i =1, …, n and n are the number of training samples, namely the total number of trees of the training samples;
according to x u N training samples of
Figure FDA0003909354180000027
Local density of
Figure FDA0003909354180000028
And distance
Figure FDA0003909354180000029
Local density
Figure FDA00039093541800000210
The calculation formula is as follows:
Figure FDA00039093541800000211
in the formula:
Figure FDA00039093541800000212
is that
Figure FDA00039093541800000213
Local density of d u Is x u U e {1,2,3,4,5}, i =1,2, …, n, m =1,2, …, n;
respectively calculate the local density is greater than
Figure FDA0003909354180000031
All training samples of (2) and
Figure FDA0003909354180000032
with the minimum value as
Figure FDA0003909354180000033
Of (2) is
Figure FDA0003909354180000034
As shown in equation (4):
Figure FDA0003909354180000035
in the formula:
Figure FDA0003909354180000036
is that
Figure FDA0003909354180000037
Local density of
Figure FDA0003909354180000038
Figure FDA0003909354180000039
Is composed of
Figure FDA00039093541800000310
And
Figure FDA00039093541800000311
u ∈ {1,2,3,4,5}; if it is
Figure FDA00039093541800000312
Has the highest local density of
Figure FDA00039093541800000313
And m is not equal to i;
computing
Figure FDA00039093541800000314
As an input attribute x u Probability value for clustering initial centers
Figure FDA00039093541800000315
As shown in equation (5):
Figure FDA00039093541800000316
in the formula: max (rho) u ) Is ρ u Maximum of n local densities; min (rho) u ) Is rho u The minimum of the n local densities; max (delta) u ) Is delta u The maximum of the n distances; min(δ u ) Is delta u The minimum of the n distances; rho u For n training samples as input attribute x u The local density of (a), including n local densities; delta u For n training samples as input attribute x u Including n distances;
step (a), inputting attribute x u N training samples
Figure FDA00039093541800000317
According to
Figure FDA00039093541800000318
Sorting the values in descending order, and taking the first k sample points as x u Performing initial central point of DPKM clustering on the n training samples; step (b), calculating input attribute x u The Euclidean distance between the n training samples and the k category centers is distributed to the category where the nearest center is located according to the minimum distance distribution principle; after n training samples are distributed once, calculating the mean value of each category, taking the mean value as a new category center and updating the category center; repeating the step (b) and the step (c) until the variation of the category center is smaller than the set error; completing DPKM clustering, wherein the center of k clusters is x u Centering on the jth fuzzy subset of the adaptive fuzzy layer
Figure FDA00039093541800000319
When inputting attribute x u Centering on the jth fuzzy subset of the adaptive fuzzy layer
Figure FDA00039093541800000320
And
Figure FDA00039093541800000321
belonging DPKM cluster
Figure FDA00039093541800000322
After establishment, adaptation is as followsCalculating x by an algorithm u Variance of jth fuzzy subset in adaptive fuzzy layer
Figure FDA00039093541800000323
First calculate x u Center of the k fuzzy subsets center
Figure FDA00039093541800000324
As shown in equation (6):
Figure FDA00039093541800000325
then solve for clustering
Figure FDA0003909354180000041
Inner sample of
Figure FDA0003909354180000042
And
Figure FDA0003909354180000043
average euclidean distance of
Figure FDA0003909354180000044
As shown in equation (7):
Figure FDA0003909354180000045
in the formula:
Figure FDA0003909354180000046
to represent
Figure FDA0003909354180000047
Inner sample of
Figure FDA0003909354180000048
And with
Figure FDA0003909354180000049
The Euclidean distance;
Figure FDA00039093541800000410
is composed of
Figure FDA00039093541800000411
Inner sample of
Figure FDA00039093541800000412
The number of (2);
Figure FDA00039093541800000413
in the formula:
Figure FDA00039093541800000414
is composed of
Figure FDA00039093541800000415
A width factor of (d); alpha is a variance scaling coefficient and is updated by a pigeon group optimization module;
Figure FDA00039093541800000416
as an input attribute x u The maximum distance between every two of the k fuzzy subsets; u ∈ {1,2,3,4,5}, j =1,2, …, k.
6. The method for accurately predicting the diameter of the breast wood volume based on the optimized fuzzy depth network according to claim 5, wherein the fuzzy inference layer comprises the following calculation processes:
the membership degree is input into the fuzzy inference layer, a product inference method is adopted to establish fuzzy units of the fuzzy inference layer, and the output value of the jth unit of the ith training sample is calculated
Figure FDA00039093541800000417
As a blur unit output, as shown in equation (9):
Figure FDA00039093541800000418
in the formula:
Figure FDA00039093541800000419
the product of the membership degrees of the jth fuzzy subset of the ith training sample in different input attributes;
Figure FDA00039093541800000420
after normalization processing is
Figure FDA00039093541800000421
n is the total number of the trees in the training sample, and the total number of the fuzzy units is k.
7. The method for accurately predicting the diameter of the breast wood volume based on the optimized fuzzy depth network as claimed in claim 6, wherein the calculation process of the attention-based weight updating layer is as follows:
the loss function is shown in equation (10):
Figure FDA00039093541800000422
in the formula: y is 1 Is the predicted value of breast diameter, y 2 The predicted value of the volume is taken as the volume;
Figure FDA00039093541800000423
is the measured value of the diameter of the breast,
Figure FDA00039093541800000424
measured values of volume of the wood; q i Attention weight for the i-th training sample, i =1,2,…,n；
Output from fuzzy inference layer
Figure FDA0003909354180000051
The connection weight between the breast diameter and the breast diameter is
Figure FDA0003909354180000052
Figure FDA0003909354180000053
And the connection weight between the volume is
Figure FDA0003909354180000054
Then for all training samples there are:
Figure FDA0003909354180000055
in the formula:
Figure FDA0003909354180000056
represents the j (th) training sample output from the fuzzy inference layer
Figure FDA0003909354180000057
A value;
Figure FDA0003909354180000058
shows the predicted value of the breast diameter of the ith training sample,
Figure FDA0003909354180000059
representing the predicted value of the volume of the ith training sample; j =1,2, …, k, i =1,2, …, n, n is the total strain number of the training samples, and the total number of fuzzy units is k;
carrying out normalization pretreatment on the training sample, and then training the parameter matrix of the ith forest tree in the sample
Figure FDA00039093541800000510
Contains 7 actually measured forest parameters and forest parameter matrix z i The east-west crown breadth, the south-north crown breadth, the tree height, the point cloud density, the crown volume, the volume and the breast diameter are sequentially arranged from left to right in the training sample, and the average value of the east-west crown breadth of all the training samples is avg 1 The average value of the north-south crown widths of all the training samples is avg 2 The mean treeGage of all training samples is avg 3 The average value of the point cloud density of all the training samples is avg 4 The mean value of the crown product of all training samples is avg 5 Volume average of all training samples is avg 6 The mean chest diameter of all training samples is avg 7 The average value matrix of each forest parameter in the training sample is avg = [ avg ] 1 ,…,avg 7 ]，Q i Is calculated as shown in equation (12):
Figure FDA00039093541800000511
in the formula:
Figure FDA0003909354180000061
representing forest parameters matrix z i Middle p-th forest parameter, avg, from left to right p Represents the p-th average value from left to right of the average value matrix avg, and belongs to p ∈ {1,2,3,4,5,6,7}; cosine similarity Sim (z) i Avg) and Euclidean distance Dist (z) i Avg) all represent the intrinsic relationship between the forest parameters of the ith tree and the average value avg; tau is an attention weight scaling coefficient and is updated by a pigeon flock optimization module; i =1,2, …, n;
the attention-based connection weight is continuously updated in a back propagation manner, and the updating manner is shown in formula (13):
Figure FDA0003909354180000062
in the formula: the value range of eta is more than 0 and less than 1, which represents the learning efficiency; t is the current iteration number;
the initial value of the attention-based connection weight is randomly given, the connection weight is continuously updated in a back propagation mode, and when the iteration times reach the maximum, the updating is terminated.
8. The optimized fuzzy depth network-based accurate prediction method for diameter at breast height of forest as claimed in claim 7, wherein the calculation process of said pigeon swarm optimization module is:
the pigeon flock optimization module is used for optimizing parameters k, alpha and tau in the fuzzy depth network, and the value combination of k, alpha and tau is called as model parameter combination delta [ k, alpha, tau)]In the parameter search space range, k is 0-100 and must be an integer, alpha is 0-30, and tau is 0-10; pigeon swarm optimization module is initially L-set model parameter delta l [k l ,α l ,τ l ](L =1,2, … L), that is, the pigeon flock optimization module has L sets of model parameters to develop optimization and t max Performing secondary iteration;
the first group of model parameter combination values of the t iteration are
Figure FDA0003909354180000063
The fitness is shown in formula (14):
Figure FDA0003909354180000064
in the formula: t is the current iteration number of the pigeon group; l =1,2, … L; i =1,2, …, n, n is the total number of strains of the training sample;
Figure FDA0003909354180000065
is the chest diameter predicted value of the fuzzy depth network,
Figure FDA0003909354180000066
to blur the volume prediction value of the depth network,
Figure FDA0003909354180000067
for the diameter at breast heightThe value of the measured value is measured,
Figure FDA0003909354180000068
measured value of volume of timber;
the search strategy of the model parameter combination is divided into two stages according to the iteration times, the first stage is started, and when the iteration times reach the maximum iteration times t max The second stage is entered at 80%, and the specific search strategy is as follows:
the calculation process of the first stage of the search strategy of the model parameter combination is shown as formula (15):
Figure FDA0003909354180000071
in the formula:
Figure FDA0003909354180000072
is a cosine iteration weight term; t is t max The maximum iteration times of the pigeon groups are obtained; ε is a minimal constant; rand (0,1) is [0,1]A random number in between; array of elements
Figure FDA0003909354180000073
Combining model parameters of the l-th group for the t-th iteration
Figure FDA0003909354180000074
Figure FDA0003909354180000075
Is an array of
Figure FDA0003909354180000076
The speed of (d); l =1,2, … L;
Figure FDA0003909354180000077
when the optimal parameter combination adaptability of the L group of model parameter combinations does not change for a long time, the L group of model parameter combinations are adaptedThe degrees are sorted in descending order and have higher fitness
Figure FDA0003909354180000078
Group parameters were subject to population variation in the manner shown in equation (16):
Figure FDA0003909354180000079
in the formula:
Figure FDA00039093541800000710
is composed of
Figure FDA00039093541800000711
The updated value of the population variation is,
Figure FDA00039093541800000712
rounding; if it is used
Figure FDA00039093541800000713
In the range of 0 to 100, the content of the organic solvent,
Figure FDA00039093541800000714
otherwise
Figure FDA00039093541800000715
Figure FDA00039093541800000716
Updating when the value is in the range of 0-30, otherwise not updating;
Figure FDA00039093541800000717
updating when the value is in the range of 0-10, otherwise not updating;
Figure FDA00039093541800000718
rand (-1,1) is [ -1,1]A random number within the range;
the calculation process of the second stage of the model parameter combination search strategy is shown in formula (17):
Figure FDA00039093541800000719
in the formula:
Figure FDA00039093541800000720
Figure FDA00039093541800000721
the tentative value of the model parameter of the first group at the time t is obtained; if it is
Figure FDA00039093541800000722
The ith set of model parameters is updated, i.e.
Figure FDA00039093541800000723
Otherwise it is not updated, i.e.
Figure FDA00039093541800000724
l＝1,2,…L；
After each iteration, abandoning part of model parameter combinations with higher fitness values, updating the number L of model parameter groups, and when the iteration number reaches the maximum iteration number t max Or ending iteration when only one group of model parameters is left, outputting the parameter combination with the lowest fitness in the L groups of model parameter combinations at the moment, namely giving the optimal values of the parameters k, alpha and tau, and transmitting the model parameter combination into the fuzzy depth network to finish training.
CN202211316076.6A 2022-10-26 2022-10-26 Forest diameter at breast height volume accurate prediction method based on optimized fuzzy depth network Pending CN115546179A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211316076.6A CN115546179A (en) 2022-10-26 2022-10-26 Forest diameter at breast height volume accurate prediction method based on optimized fuzzy depth network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211316076.6A CN115546179A (en) 2022-10-26 2022-10-26 Forest diameter at breast height volume accurate prediction method based on optimized fuzzy depth network

Publications (1)

Publication Number Publication Date
CN115546179A true CN115546179A (en) 2022-12-30

Family

ID=84719054

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211316076.6A Pending CN115546179A (en) 2022-10-26 2022-10-26 Forest diameter at breast height volume accurate prediction method based on optimized fuzzy depth network

Country Status (1)

Country Link
CN (1) CN115546179A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115953607A (en) * 2023-01-04 2023-04-11 北京数字绿土科技股份有限公司 Trunk diameter at breast height extraction method and system using point cloud data

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115953607A (en) * 2023-01-04 2023-04-11 北京数字绿土科技股份有限公司 Trunk diameter at breast height extraction method and system using point cloud data
CN115953607B (en) * 2023-01-04 2024-02-13 北京数字绿土科技股份有限公司 Trunk breast diameter extraction method and system using point cloud data

Similar Documents

Publication Publication Date Title
Keerthana et al. An ensemble algorithm for crop yield prediction
CN112906298B (en) Blueberry yield prediction method based on machine learning
CN110222215B (en) Crop pest detection method based on F-SSD-IV3
CN113269825B (en) Forest breast diameter value extraction method based on foundation laser radar technology
CN109214591B (en) Method and system for predicting aboveground biomass of woody plant
Tanwar et al. Deep learning-based hybrid model for severity prediction of leaf smut sugarcane infection
Zhang et al. Robust image segmentation method for cotton leaf under natural conditions based on immune algorithm and PCNN algorithm
CN115546179A (en) Forest diameter at breast height volume accurate prediction method based on optimized fuzzy depth network
Wang et al. An improved crop disease identification method based on lightweight convolutional neural network
CN113139717B (en) Crop seedling condition grading remote sensing monitoring method and device
CN116822381B (en) Marine temperature salt structure inversion method based on artificial intelligence
CN115880487A (en) Forest laser point cloud branch and leaf separation method based on deep learning method
Radhika et al. Ensemble subspace discriminant classification of satellite images
Deng et al. Kiwifruit vine extraction based on low altitude UAV remote sensing and deep semantic segmentation
CN106056167A (en) Normalization possibilistic fuzzy entropy clustering method based on Gaussian kernel hybrid artificial bee colony algorithm
Gao et al. How can agricultural water production be promoted? a review on machine learning for irrigation
CN113221447A (en) Soil humidity prediction method for optimizing BP neural network based on improved genetic algorithm
Monica et al. Soil NPK prediction using enhanced genetic algorithm
CN117034116A (en) Machine learning-based traditional village space type identification method
Gao et al. Classification Method of Rape Root Swelling Disease Based on Convolution Neural Network
CN113052379A (en) Tomato water demand prediction method based on BAS-RBF
Rungyaem et al. Comparison of 3D Rice Organs Point Cloud Classification Techniques
Alajas et al. Grape pseudocercospora leaf specked area estimation using hybrid genetic algorithm and recurrent neural network
Jamil et al. Ensemble of Classifiers Based Soil Evaluation for Crop Cultivation
CN113298280A (en) Deep mining and intelligent decision algorithm for agricultural breeding big data

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