CN106650791A - Improved particle swarm-based non-supervised remote sensing image classification method - Google Patents

Improved particle swarm-based non-supervised remote sensing image classification method Download PDF

Info

Publication number
CN106650791A
CN106650791A CN201611025653.0A CN201611025653A CN106650791A CN 106650791 A CN106650791 A CN 106650791A CN 201611025653 A CN201611025653 A CN 201611025653A CN 106650791 A CN106650791 A CN 106650791A
Authority
CN
China
Prior art keywords
particle
classification
suitability degree
remote sensing
suitability
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
CN201611025653.0A
Other languages
Chinese (zh)
Inventor
李华朋
张淑清
刘春悦
刘照
丁小辉
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.)
Northeast Institute of Geography and Agroecology of CAS
Original Assignee
Northeast Institute of Geography and Agroecology of CAS
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 Northeast Institute of Geography and Agroecology of CAS filed Critical Northeast Institute of Geography and Agroecology of CAS
Priority to CN201611025653.0A priority Critical patent/CN106650791A/en
Publication of CN106650791A publication Critical patent/CN106650791A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to a remote sensing image classification method, in particular, an improved particle swarm-based non-supervised remote sensing image classification method. When a particle swarm intelligent algorithm is adopted to cluster remote sensing images, the particle swarm intelligent algorithm easily falls into local optimum, and as a result, optimal remote sensing classification results cannot be obtained, and classification results are unstable, while, the invention aims to solve the above problems. The method of the invention comprises the following steps that: control parameters are set according to a reference particle swarm algorithm, and particle swarm expression and initialization are performed; after population suitability is calculated, particle suitability is evaluated, and then, the velocity of each particle is updated, and the positions of the particles are updated; the suitability of each particle is recalculated, the optimal suitability of a whole population is calculated, and new local optimal suitability is compared and selected, and the positions of the particles are recorded; Levy flight is adopted to update the positions of particles with the minimum suitability in each cycle; and cycling is stopped when the maximum number of cycles is reached, a global optimal solution is outputted, and therefore, a remote sensing image classification task is completed. The improved particle swarm-based non-supervised remote sensing image classification method of the present invention is applicable to the classification of remote sensing images.

Description

A kind of non-supervisory Remote Image Classification based on improvement population
Technical field
The present invention relates to a kind of Remote Image Classification.
Background technology
Land_use change/cover (LUCC) data can reflect basic nature and social processes, and act is occupied in geoscience The status of sufficient weight, this is because LUCC data are much to predict that the model of terrestrial ecosystem provides the foundation data.Additionally, LUCC data are most important with the complex interaction mechanism of whole world change for mankind's activity is understood.At present, remote sensing technology into To obtain the mainstream technology means of LUCC data, this is because it has broad perspectives, economy, repeatability and real-time etc. solely Special advantage.
With the continuous progress of remote sensing technology, the remote sensing image resource that people obtain is enriched and various further, day cumulative number Magnanimity rank is reached according to amount.Under this background, non-supervisory Classification in Remote Sensing Image method is subject to the more and more concerns of user (simple and high Effect characteristic).Wherein, k-means is the most commonly used method, can realize that Fast Convergent completes image cluster.However, its search energy Power is limited, is extremely difficult to globally optimal solution.
In fact, non-supervisory classification of remote-sensing images can be converted into an optimization problem, and intelligent algorithm is asked this The solution of topic has natural advantage.Genetic algorithm (genetic algorithm, GA) is the most frequently used artificial intelligence the most classical Energy algorithm, scholars have been developed based on the non-supervisory Classification in Remote Sensing Image method of GA, and prove that its performance is better than traditional k-means side Method.However, because remote sensing image general data amount is larger and data are complicated various, the exploitation energy of the search mechanisms of GA to solution space Power is limited, it is impossible to realize Fast Convergent.
In recent years, particle colony intelligence (PSO) has evolved into one of artificial intelligence field hot topic method.PSO is to imitate to fly Flock of birds look for food during behavior and formed.In PSO, each particle represents a flying bird.During particle evolution, Particle evolution is instructed using globally optimal solution (gbest) and locally optimal solution (pbest).Thus PSO convergence rates are compared to GA Algorithm is significantly improved.At present, there is scholar that PSO is introduced into remote sensing cluster field, and developed non-supervisory population Intelligent classification algorithm (UPSO).The method includes following basic step：
Each solution is considered as a particle by 1, and is at random each one position of particle distribution；
2 construction suitability degree functions, the evaluation for particle suitability degree provides basis；
3 particles complete its own evolution and position more by mutual learning and exchange mechanism (global with local optimization particle) Newly；
4 reach circulation stop condition, export optimal solution, realize remote sensing image unsupervised classification.
But UPSO possesses traditional intelligence algorithm advantage, i.e. self-organizing, self study and the ability from convergence.Also, disobey Bad data distribution situation, can obtain under many circumstances the classification of remote-sensing images knot more more excellent than the traditional intelligence such as GA method Really.However, UPSO yet suffers from some obvious shortcomings：
(1) global search cannot be realized.UPSO instructs particle evolution using optimal solution, although can reach the mesh of Fast Convergent , but cost is similarity more and more higher between particle, population loses diversity, so as to be absorbed in local optimum, it is impossible to obtain Optimized remote sensing classification results.
(2) performance relies on initial solution quality.Because UPSO instructs particle evolution, initial disaggregation to directly affect using optimal solution Evolutionary direction, thus its classification results extremely relies on initial disaggregation quality, classification results are unstable.
The content of the invention
The present invention in order to solve at present using particle swarm intelligence algorithm to remote sensing images carry out cluster presence be easily trapped into Local optimum and the unstable problem of problem and classification results of optimized remote sensing classification results cannot be obtained.
A kind of non-supervisory Remote Image Classification based on improvement population, comprises the steps：
Step 1, a width remote sensing image are made up of picture dot, and its wave band number and classification number to be sorted are respectively d and k；And set Determine clustering target M, clustering target M is each picture dot in remote sensing image to its affiliated cluster centre Euclidean distance sum；
With reference to particle cluster algorithm, five control parameters, including release number of particles, maximum cycle Max_ are set Iter, Dynamic Inertia factor w and two accelerated factors c1And c2；
Step 2, population expression and initialization：
Any type cluster centre ziWith one group of real number representation, real number number is equal to wave band number d；By by k classification Cluster centre connects to represent a particle, thus a particle length is d × k；Front d positional representation in one particle One categorical clusters center, by that analogy, per the d class cluster centre of positional representation one；
Each particle gives at random a position, generates particle initial position；
Step 3, calculating population suitability degree：
The size of clustering target M values is inversely proportional to clustering result quality；Define suitability degree function f as follows：
F=1/ (M+1) (1)
After the suitability degree for calculating each particle, global and local optimum particle position is recorded；
Step 4, the search of particle optimal location：
After assessment particle suitability degree, each particle rapidity is updated using formula (2)：
Wherein, ViAnd V (t+1)iT () is respectively i-th particle in t+1 moment and the speed of t；And Xi(t) I-th particle history optimal location and the position in t, X are represented respectivelygbestFor the history optimal location of whole population； r1、r2Respectively two random numbers changed between 0-1；
Particle rapidity value is updated again after updating to particle position；
Step 5, optimum particle position update：
After particle completes location updating, using formula (1) suitability degree of each particle is recalculated；Calculate whole kind The optimum suitability degree of group, and be compared with the suitability degree of the global optimum position for recording before, update and record new optimum position Put and its suitability degree；For each particle, the particle history optimum suitability degree of current particle suitability degree and record is contrasted, will The higher new local optimum suitability degree for being recorded as the particle of suitable angle value, and while record particle position；
Step 6, search new position of being flown according to row dimension：
Suitability degree smallest particles position in circulation each time is identified, and the position is updated using row dimension flight, To reach global search purpose；The new position computing formula of suitability degree smallest particles (worst particle) is as follows：
Wherein,The new position of i-th particle is represented, original position is set to Xi；After location updating, the suitability degree of the particle Update immediately；S represents step-length；
Step 7, stopping circulation completing classification of remote-sensing images：
If circulation is not up to maximum cycle Max_Iter, returns to step 4 and proceed particle position search；Work as arrival Maximum cycle Max_Iter, stopping circulates and exports globally optimal solution, and globally optimal solution is exactly Optimal cluster centers, now It is considered as clustering target M values and reaches minimum, so as to completes classification of remote-sensing images task.
Preferably, the computing formula of the clustering target M described in step 1 is as follows：
Wherein, Ci′Represent classification i ', i '=1,2 ..., k；J ' is the picture dot number in classification i ', xj′For in classification i ' Any one picture dot, zi′As the cluster centre of classification i '.
Preferably, described in step 2 each particle gives at random the process that a position generates particle initial position Comprise the following steps：
Particle initial position is generated using below equation：
Wherein,I-th particle is represented in the position of j-th attribute,WithJ-th attribute is represented respectively most Little value and maximum, r is a random number changed between 0-1.
Preferably, the particle rapidity value described in step 4 is updated again to particle position after updating and comprises the following steps：
After particle rapidity value updates, subsequently particle position is updated with below equation：
Xi(t+1)=Xi(t)+Vi(t+1)
Wherein, Xi(t+1) represent i-th particle in the position at t+1 moment.
Preferably, the computing formula of step-length s described in step 6 is as follows：
μ, ν and λ are calculated respectively from normal distribution：
Wherein, Γ is gamma functions, and β is the constant changed between 1 to 2.
The present invention has the effect that：
When existing utilization UPSO carries out classification of remote-sensing images, during evolution by reference to globally optimal solution and local Optimal solution, instructs population Evolutionary direction, and the result for causing is, as particle similarity more and more higher is goed deep in evolution, thus to be absorbed in Local optimum.And the non-supervisory Remote Image Classification (UIPSO) based on improvement population of the present invention is in optimum cluster In heart search procedure, identification circulates each time worst particle, and using row dimension flight particle position is updated, and substantially increases population complete Office's search capability, it is relative to carry out Remote Image Classification using UPSO so as to improve remote sensing image accuracy of identification, it is of the invention Remote sensing image accuracy of identification improves more than 8%.
Also, existing utilization UPSO carries out classification of remote-sensing images and evolves to depend on initial disaggregation, and the present invention based on The non-supervisory Remote Image Classification (UIPSO) of population is improved by using row dimension flight, the overall situation of solution space is realized Roaming, thus greatly reduce the dependence to initial disaggregation；Even if giving low quality initial disaggregation, UIPSO still can be obtained Stable Classification in Remote Sensing Image result.
Description of the drawings
Fig. 1 is the schematic flow sheet of the present invention.
Specific embodiment
Specific embodiment one：Present embodiment is illustrated with reference to Fig. 1,
A kind of non-supervisory Remote Image Classification based on improvement population, comprises the steps：
Step 1, assume that a width remote sensing image is made up of N number of picture dot, its wave band number and classification number to be sorted be respectively d and k；And set clustering target M, clustering target M be in remote sensing image each picture dot to its affiliated cluster centre Euclidean distance it With；
With reference to particle cluster algorithm, five control parameters, including release number of particles Num_Particle, largest loop are set Number of times Max_Iter, Dynamic Inertia factor w and two accelerated factors c1And c2；Num_Particle is used for controlling the number of population Amount, Max_Iter provides circulation stop condition, and w defines the weight of velocity inertia, c1And c2Define local and global optimum The impact of solution；
Step 2, population expression and initialization：
In the present invention, any type cluster centre ziWith one group of real number representation, real number number is equal to wave band number d；Pass through The cluster centre connection of k classification represents a particle (i.e. the solution of clustering problem), thus a particle length For d × k；Front d positional representation first category cluster centre in one particle, by that analogy, per the d class of positional representation one cluster Center；
Each particle gives at random a position, generates particle initial position；
Step 3, calculating population suitability degree：
The size of clustering target M values is inversely proportional to clustering result quality, i.e. clustering target value is less, and clustering result quality is higher (same Similarity is higher in cluster cluster)；Define suitability degree function f as follows：
F=1/ (M+1) (1)
After the suitability degree for calculating each particle, global and local optimum particle position is recorded；
Step 4, the search of particle optimal location：
After assessment particle suitability degree, each particle rapidity is updated using formula (2)：
Wherein, ViAnd V (t+1)iT () is respectively i-th particle in t+1 moment and the speed of t；And Xi(t) I-th particle history optimal location and the position in t, X are represented respectivelygbestFor the history optimal location of whole population；w For Dynamic Inertia coefficient, impact of the upper moment speed to this moment speed is represented；r1、r2Respectively two change between 0-1 Random number；c1、c2Respectively accelerated factor；
Particle rapidity value is updated again after updating to particle position；
Step 5, optimum particle position update：
After particle completes location updating, using formula (1) suitability degree of each particle is recalculated；Calculate whole kind The optimum suitability degree of group, and be compared with the suitability degree of the global optimum position for recording before, update and record new optimum position Put (suitable angle value is higher) and its suitability degree；For each particle, contrast current particle suitability degree and go through with the particle of record History optimum suitability degree, by the higher new local optimum suitability degree for being recorded as the particle of suitable angle value, and while records particle position Put；Local optimum suitability degree for each particle, i.e. the optimum suitability degree of each particle；Global optimum's suitability degree is pin Optimum suitability degree for whole population；
Step 6, search new position of being flown according to row dimension：
Suitability degree smallest particles position in circulation each time is identified, and the position is updated using row dimension flight, To reach global search purpose；Row dimension flight is a kind of food source way of search extensively utilized by animal, its step-size in search Combination of long drives and drop shots, does not meet Gaussian Profile rule, thus can realize global roaming and search；Suitability degree smallest particles (worst particle) New position computing formula it is as follows：
Wherein,The new position of i-th particle (worst particle) is represented, original position is set to Xi；After location updating, the particle Suitability degree also update immediately；S represents step-length；
Step 7, stopping circulation completing classification of remote-sensing images：
If circulation is not up to maximum cycle Max_Iter, returns to step 4 and proceed particle position search；Work as arrival Maximum cycle Max_Iter, stopping circulates and exports globally optimal solution, and globally optimal solution is exactly Optimal cluster centers, now It is considered as clustering target M values and reaches minimum, so as to completes classification of remote-sensing images task.
Specific embodiment two：
The computing formula of the clustering target M described in the step of present embodiment 1 is as follows：
Wherein, Ci′Represent classification i ', i '=1,2 ..., k；J ' is the picture dot number in classification i ', xj′For in classification i ' Any one picture dot, zi′As the cluster centre of classification i '.
Other steps and parameter are identical with specific embodiment one.
Specific embodiment three：
Each particle described in the step of present embodiment 2 gives at random a position and generates particle initial position Process is comprised the following steps：
Particle initial position is generated using below equation：
Wherein,I-th particle is represented in the position of j-th attribute,WithJ-th attribute is represented respectively most Little value and maximum, r is a random number changed between 0-1.
Other steps and parameter are identical with specific embodiment one or two.
Specific embodiment four：
Particle rapidity value described in the step of present embodiment 4 is updated including following again after updating to particle position Step：
After particle rapidity value updates, subsequently particle position is updated with below equation：
Xi(t+1)=Xi(t)+Vi(t+1)
Wherein, Xi(t+1) represent i-th particle in the position at t+1 moment.
Other steps and parameter are identical with one of specific embodiment one to three.
Specific embodiment five：
The computing formula of step-length s described in the step of present embodiment 6 is as follows：
μ, ν and λ are calculated respectively from normal distribution：
Wherein, Γ is gamma functions, and β is the constant changed between 1 to 2.
Other steps and parameter are identical with one of specific embodiment one to four.

Claims (5)

1. it is a kind of based on the non-supervisory Remote Image Classification for improving population, it is characterised in that to comprise the steps：
Step 1, a width remote sensing image are made up of picture dot, and its wave band number and classification number to be sorted are respectively d and k；And set poly- Class index M, clustering target M is each picture dot in remote sensing image to its affiliated cluster centre Euclidean distance sum；
With reference to particle cluster algorithm, five control parameters are set, including are discharged number of particles, maximum cycle Max_Iter, moved State inertial factor w and two accelerated factors c1And c2；
Step 2, population expression and initialization：
Any type cluster centre ziWith one group of real number representation, real number number is equal to wave band number d；By by the cluster of k classification The heart connects to represent a particle, thus a particle length is d × k；Front d positional representation first category in one particle Cluster centre, by that analogy, per the d class cluster centre of positional representation one；
Each particle gives at random a position, generates particle initial position；
Step 3, calculating population suitability degree：
The size of clustering target M values is inversely proportional to clustering result quality；Define suitability degree function f as follows：
F=1/ (M+1) (1)
After the suitability degree for calculating each particle, global and local optimum particle position is recorded；
Step 4, the search of particle optimal location：
After assessment particle suitability degree, each particle rapidity is updated using formula (2)：
V i ( t + 1 ) = wV i ( t ) + c 1 r 1 ( X i p b e s t - X i ( t ) ) + c 2 r 2 ( X g b e s t - X i ( t ) ) - - - ( 2 )
Wherein, ViAnd V (t+1)iT () is respectively i-th particle in t+1 moment and the speed of t；And Xi(t) difference table Show i-th particle history optimal location and the position in t, XgbestFor the history optimal location of whole population；r1、r2Point Wei not two random numbers changed between 0-1；
Particle rapidity value is updated again after updating to particle position；
Step 5, optimum particle position update：
After particle completes location updating, using formula (1) suitability degree of each particle is recalculated；Calculate whole population most Excellent suitability degree, and being compared with the suitability degree of the global optimum position for recording before, update and record new optimal location and Its suitability degree；For each particle, the particle history optimum suitability degree of current particle suitability degree and record is contrasted, will be suitable The higher new local optimum suitability degree for being recorded as the particle of angle value, and while record particle position；
Step 6, search new position of being flown according to row dimension：
Suitability degree smallest particles position in circulation each time is identified, and the position is updated using row dimension flight, to reach To global search purpose；The new position computing formula of suitability degree smallest particles is as follows：
X i n e w = X i + s - - - ( 3 )
Wherein,The new position of i-th particle is represented, original position is set to Xi；After location updating, the suitability degree of the particle is also immediately Update；S represents step-length；
Step 7, stopping circulation completing classification of remote-sensing images：
If circulation is not up to maximum cycle Max_Iter, returns to step 4 and proceed particle position search；It is maximum when reaching Cycle-index Max_Iter, stopping circulates and exports globally optimal solution, and globally optimal solution is exactly Optimal cluster centers, is now considered as Clustering target M values reach minimum, so as to complete classification of remote-sensing images task.
2. according to claim 1 a kind of based on the non-supervisory Remote Image Classification for improving population, its feature exists In the computing formula of the clustering target M described in step 1 is as follows：
M = Σ i ′ = 1 k Σ x j ′ ∈ C i ′ | | x j ′ - z i ′ | |
Wherein, Ci′Represent classification i ', i '=1,2 ..., k；J ' is the picture dot number in classification i ', xj′It is any in for classification i ' One picture dot, zi′As the cluster centre of classification i '.
3. according to claim 1 and 2 a kind of based on the non-supervisory Remote Image Classification for improving population, its feature It is that each particle described in step 2 gives at random the process of a position generation particle initial position includes following step Suddenly：
Particle initial position is generated using below equation：
X i j = X m i n j + r ( X m a x j - X m i n j )
Wherein,I-th particle is represented in the position of j-th attribute,WithThe minimum of a value of j-th attribute is represented respectively And maximum, r is a random number changed between 0-1.
4. according to claim 3 a kind of based on the non-supervisory Remote Image Classification for improving population, its feature exists In the particle rapidity value described in step 4 is updated again to particle position after updating and comprises the following steps：
After particle rapidity value updates, subsequently particle position is updated with below equation：
Xi(t+1)=Xi(t)+Vi(t+1)
Wherein, Xi(t+1) represent i-th particle in the position at t+1 moment.
5. according to claim 4 a kind of based on the non-supervisory Remote Image Classification for improving population, its feature exists In the computing formula of step-length s described in step 6 is as follows：
s = 0.01 μ | ν | 1 / β λ
μ, ν and λ are calculated respectively from normal distribution：
λ～N (0,1)σv=1 wherein, and Γ is Gamma functions, β is the constant changed between 1 to 2.
CN201611025653.0A 2016-11-21 2016-11-21 Improved particle swarm-based non-supervised remote sensing image classification method Pending CN106650791A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611025653.0A CN106650791A (en) 2016-11-21 2016-11-21 Improved particle swarm-based non-supervised remote sensing image classification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611025653.0A CN106650791A (en) 2016-11-21 2016-11-21 Improved particle swarm-based non-supervised remote sensing image classification method

Publications (1)

Publication Number Publication Date
CN106650791A true CN106650791A (en) 2017-05-10

Family

ID=58808490

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611025653.0A Pending CN106650791A (en) 2016-11-21 2016-11-21 Improved particle swarm-based non-supervised remote sensing image classification method

Country Status (1)

Country Link
CN (1) CN106650791A (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102880872A (en) * 2012-08-28 2013-01-16 中国科学院东北地理与农业生态研究所 Classification and construction method for semi-supervised support vector machine (SVM) remote sensing image
CN103886335A (en) * 2014-04-09 2014-06-25 西安电子科技大学 Polarized SAR image classifying method based on fuzzy particle swarms and scattering entropy
US8768868B1 (en) * 2012-04-05 2014-07-01 Hrl Laboratories, Llc Optimal multi-class classifier threshold-offset estimation with particle swarm optimization for visual object recognition

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8768868B1 (en) * 2012-04-05 2014-07-01 Hrl Laboratories, Llc Optimal multi-class classifier threshold-offset estimation with particle swarm optimization for visual object recognition
CN102880872A (en) * 2012-08-28 2013-01-16 中国科学院东北地理与农业生态研究所 Classification and construction method for semi-supervised support vector machine (SVM) remote sensing image
CN103886335A (en) * 2014-04-09 2014-06-25 西安电子科技大学 Polarized SAR image classifying method based on fuzzy particle swarms and scattering entropy

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
UJJWAL MAULIK 等: "Genetic algorithm-based clustering technique", 《THE JOURNAL OF THE PATTERN RECOGNITION SOCIETY》 *
王庆喜 等: "基于莱维飞行的粒子群优化算法", 《计算机应用研究》 *
蒋卫国 等: "基于HPSO和FCM的多光谱遥感图像湿地分类", 《光谱学与光谱分析》 *

Similar Documents

Publication Publication Date Title
Wang et al. Stct: Sequentially training convolutional networks for visual tracking
Cui et al. A fast particle swarm optimization
CN103971160B (en) particle swarm optimization method based on complex network
CN107358293A (en) A kind of neural network training method and device
Ouyang et al. Adaptive spiral flying sparrow search algorithm
CN108038507A (en) Local receptor field extreme learning machine image classification method based on particle group optimizing
CN111988329B (en) Network intrusion detection method based on deep learning
Lin et al. Machine learning templates for QCD factorization in the search for physics beyond the standard model
CN103679132A (en) A sensitive image identification method and a system
CN109063719A (en) A kind of image classification method of co-ordinative construction similitude and category information
CN108985440A (en) Two type intuitionistic fuzzy random vector function Connection Neural Network design method of section
CN107578056A (en) A kind of manifold learning system integrated classical model and be used for sample dimensionality reduction
Yang et al. Adaptive chaotic spherical evolution algorithm
CN110197307A (en) A kind of domain type sea-surface temperature prediction technique of combination attention mechanism
CN114021738A (en) Distributed generation countermeasure model-based federal learning method
CN116451737A (en) PG-W-PSO method for improving particle swarm based on reinforcement learning strategy gradient
CN115912502A (en) Intelligent power grid operation optimization method and device
Chen et al. A feature selection method for intrusion detection based on parallel sparrow search algorithm
Strickland et al. A reinforcement learning approach to multi-robot planar construction
CN106650791A (en) Improved particle swarm-based non-supervised remote sensing image classification method
CN110110690A (en) A kind of target pedestrian method for tracing, device, equipment and storage medium
Ou et al. The Multidimensional Motion Features of Spatial Depth Feature Maps: An Effective Motion Information Representation Method for Video‐Based Action Recognition
Alkhraisat et al. Dynamic inertia weight particle swarm optimization for solving nonogram puzzles
Wang et al. Improved Particle Swarm Optimization Algorithm Based on Gaussian-Grid Search Method.
CN107203805A (en) A kind of many group's bi-directional drive collaborative searching algorithms under big data environment

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
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20170510