CN110781788B - Method and system for field robot ground classification based on small amount of labels - Google Patents

Method and system for field robot ground classification based on small amount of labels Download PDF

Info

Publication number
CN110781788B
CN110781788B CN201910995967.0A CN201910995967A CN110781788B CN 110781788 B CN110781788 B CN 110781788B CN 201910995967 A CN201910995967 A CN 201910995967A CN 110781788 B CN110781788 B CN 110781788B
Authority
CN
China
Prior art keywords
vibration
ground
sample
module
state transition
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.)
Active
Application number
CN201910995967.0A
Other languages
Chinese (zh)
Other versions
CN110781788A (en
Inventor
吕文君
康宇
李泽瑞
昌吉
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.)
Hefei Zhongke Liheng Intelligent Technology Co ltd
Original Assignee
University of Science and Technology of China USTC
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 University of Science and Technology of China USTC filed Critical University of Science and Technology of China USTC
Priority to CN201910995967.0A priority Critical patent/CN110781788B/en
Publication of CN110781788A publication Critical patent/CN110781788A/en
Application granted granted Critical
Publication of CN110781788B publication Critical patent/CN110781788B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Signal Processing (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a field robot ground classification method and system based on a small amount of labels, which belong to the technical field of computers. The sample classification of the invention well reflects the actual situation of data classification, and the classified data is used for training the model, thereby improving the accuracy of semi-supervised classification.

Description

Method and system for field robot ground classification based on small amount of labels
Technical Field
The invention relates to the technical field of computers, in particular to a field robot ground classification method and system based on a small number of labels.
Background
How to utilize mass data is an important task faced by current machine learning, and a traditional support vector machine is a supervised learning method and needs a large number of marked samples for training. However, in practical applications, since most of the available sample data is unlabeled, there are fewer labeled sample points, and if only these fewer labeled samples are used, the information existing in a large number of position-labeled samples is lost. Therefore, the learner proposes a semi-supervised learning method, namely, the semi-supervised learning method utilizes unlabeled sample data knowledge and utilizes a small amount of labeled sample data knowledge. However, currently available semi-supervised learning methods do not make full use of spatial smoothness, especially in robot ground classification, resulting in a large number of misclassifications.
Disclosure of Invention
The invention aims to overcome the defects in the background technology and improve the accuracy of semi-supervised classification.
In order to achieve the above object, in one aspect, a field robot ground classification method based on a small number of labels is adopted, which includes the following steps:
acquiring a vibration signal acquired by a vibration sensor and a ground image signal acquired by image acquisition equipment, and acquiring a vibration frame set and a ground image signal set;
converting the vibration frame into a sample by adopting d-point Fourier transform to obtain a sample set X ═ X1,x2,…,xp,…,xPIn which xpIs a sample, specimen
Figure GDA0002701723550000011
Representing a real number set, d being a sample dimension;
let L ═ {1,2, …, c, …, L } denote the set of ground type number values, c ═ 1,2, …, L, establish the state transition matrix
Figure GDA0002701723550000025
m,n＝1,2,…,l，Tm,nRepresenting the transition probability from state m to state n;
setting a labeling interval delta according to the minimum element on the diagonal line in the state transition matrix, wherein delta is an integer;
read at mark interval Δ
Figure GDA0002701723550000021
Artificial identification
Figure GDA0002701723550000022
Corresponding real ground type and pair xpLabeling to obtain a labeled sample set { xi,yiAnd unlabeled sample set { x }j},i＝1,2,…,I，j＝1,2,…,J，I+J＝P，yiIs a sample xiMarking of (1);
training the constructed support vector machine model by using the labeled sample set and the unlabeled sample set to obtain a trained support vector machine;
and processing the currently acquired vibration signals by using a trained support vector machine to realize the classification of the ground types.
Further, the acquiring a vibration signal acquired by a vibration sensor and a ground image signal acquired by an image acquisition device, and obtaining a set of vibration frames and a set of ground image signals includes:
converting a vibration signal into a set of vibration frames v1,v2,…,vp,…,vPIn which v ispIs a vibration frame, P is 1,2, …, P is the number of all vibration frames;
corresponding each vibration frame with the ground image signal according to the time stamp to obtain a set of the ground image signals
Figure GDA0002701723550000023
The vibration signal and the image signal are both time-stamped.
Further, the setting the labeling interval Δ according to the smallest element on the diagonal in the state transition matrix includes:
computing satisfaction
Figure GDA0002701723550000024
Corresponding to w, to obtain wminWherein, TminRepresents the smallest element on the diagonal in the state transition matrix, τ ∈ (0,1) is the threshold, w ═ 0,1,2, …;
calculating the maximum annotation separation Wmax＝2wmin+1 and setting the marking interval Delta epsilon [1, W ∈ ]max]。
Further, before the training of the constructed support vector machine model by using the labeled sample set and the unlabeled sample set, the method further includes establishing a support vector machine model as follows:
Figure GDA0002701723550000031
wherein HKFor regenerating nuclear Hilbert space, V (x)i,yiF) is a loss function, | f |KIs the complexity metric norm, gamma, of f in the regenerative nuclear Hilbert spaceK,γA,γS>0；
Figure GDA0002701723550000032
θpqIs a characteristic similarity coefficient, xqFor the sample, q is 1,2, …, P,
Figure GDA0002701723550000033
Figure GDA0002701723550000034
is a spatial similarity coefficient.
Further, the thetapqThe calculation formula is as follows:
Figure GDA0002701723550000035
wherein,
Figure GDA0002701723550000036
is the distance x in the feature spaceqSet of most recent N samples, tθIs the gaussian kernel width;
the spatial similarity coefficient
Figure GDA0002701723550000037
The calculation formula is as follows:
Figure GDA0002701723550000038
wherein,
Figure GDA0002701723550000039
is the width of the gaussian kernel and is,
Figure GDA00027017235500000310
for the y th diagonal of the state transition matrixiElement, p (x)i,xj) Is xiAnd xjThe sampling spatial distance therebetween.
On the other hand, adopt a field robot ground classification system based on a small amount of marks, include: the robot comprises data acquisition equipment and data processing equipment, wherein the data acquisition equipment comprises a vibration sensor and image acquisition equipment which are installed on a robot body, and the data processing equipment comprises a data acquisition module, a sample construction module, a state transition matrix establishing module, a labeling interval determining module, a labeling module, a model training module and a classification module;
the data acquisition module is used for acquiring vibration signals acquired by the vibration sensor and ground image signals acquired by the image acquisition equipment and acquiring a vibration frame set and a ground image signal set;
the sample construction module is used for converting the vibration frame into a sample by adopting d-point Fourier transform to obtain a sample set X ═ X1,x2,…,xp,…,xPIn which xpIs a sample, specimen
Figure GDA0002701723550000041
Representing a real number set, d being a sample dimension;
the state transition matrix establishing module is used for setting L to be {1,2, …, c, …, L } to represent a set of label values, c to be 1,2, …, L, and establishing the state transition matrix
Figure GDA0002701723550000042
m,n＝1,2,…,l，Tm,nRepresenting the transition probability from state m to state n;
the marking interval determining module is used for setting a marking interval delta according to the minimum element on the diagonal line in the state transition matrix, wherein the delta is an integer;
the marking module is used for reading according to the marking interval delta
Figure GDA0002701723550000043
Artificial identification
Figure GDA0002701723550000044
Corresponding real ground type and pair xpLabeling to obtain a labeled sample set { xi,yiAnd unlabeled sample set { x }j},i＝1,2,…,I，j＝1,2,…,J，I+J＝P，yiIs a sample xiMarking of (1);
the model training module is used for training the constructed support vector machine model by utilizing the marked sample set and the unmarked sample set to obtain a trained support vector machine;
the classification module is used for processing the currently acquired vibration signals by using a trained support vector machine to realize the classification of the ground types.
Further, the data acquisition module comprises a vibration frame set construction unit and a ground image signal set construction unit;
the vibration frame set construction unit is used for converting the vibration signal into a set of vibration frames { v1,v2,…,vp,…,vPIn which v ispIs a vibration frame, P is 1,2, …, P is the number of all vibration frames;
the ground image signal set construction unit is used for corresponding each vibration frame with the ground image signal according to the time stamp to obtain a set of the ground image signal
Figure GDA0002701723550000051
The vibration signal and the image signal are both time-stamped.
Further, the labeling interval determining module is specifically configured to:
computing satisfaction
Figure GDA0002701723550000052
Corresponding to w, to obtain wminWherein, TminRepresents the smallest element on the diagonal in the state transition matrix, τ ∈ (0,1) is the threshold, w ═ 0,1,2, …;
calculating the maximum annotation separation Wmax＝2wmin+1 and setting the marking interval Delta epsilon [1, W ∈ ]max]。
Further, the system also comprises a support vector machine model building module, wherein the support vector machine model building module is used for building a support vector machine model as follows:
Figure GDA0002701723550000053
wherein HKFor regenerating nuclear Hilbert space, V (x)i,yiF) is a loss function, | f |KIs the complexity metric norm, gamma, of f in the regenerative nuclear Hilbert spaceK,γA,γS>0；
Figure GDA0002701723550000054
θpqIs a characteristic similarity coefficient, xqFor the sample, q is 1,2, …, P,
Figure GDA0002701723550000055
is a spatial similarity coefficient.
Further, the thetapqThe calculation formula is as follows:
Figure GDA0002701723550000061
wherein,
Figure GDA0002701723550000062
is the distance x in the feature spaceqSet of most recent N samples, tθIs the gaussian kernel width;
the spatial similarity coefficient
Figure GDA0002701723550000063
The calculation formula is as follows:
Figure GDA0002701723550000064
wherein,
Figure GDA0002701723550000065
is the width of the gaussian kernel and is,
Figure GDA0002701723550000066
for the y th diagonal of the state transition matrixiElement, p (x)i,xj) Is xiAnd xjBetween the twoSample space distance.
Compared with the prior art, the invention has the following technical effects: the vibration signals acquired by the vibration sensor correspond to the ground image signals acquired by the image acquisition equipment, the real ground type corresponding to the ground image signals is manually identified, the vibration signals are labeled, labeled samples and unlabeled samples are obtained, and a support vector machine is trained, so that the lack of labeling of a certain sample is prevented, the assumption of spatial smoothness is better utilized, and the classification accuracy is improved.
Drawings
The following detailed description of embodiments of the invention refers to the accompanying drawings in which:
FIG. 1 is a flow chart diagram of a field robot ground classification method based on a small number of labels;
FIG. 2 is a schematic structural diagram of a field robot ground classification system based on a small number of labels.
Detailed Description
To further illustrate the features of the present invention, refer to the following detailed description of the invention and the accompanying drawings. The drawings are for reference and illustration purposes only and are not intended to limit the scope of the present disclosure.
As shown in fig. 1, the present embodiment discloses a field robot ground classification method based on a small number of labels, wherein a vibration sensor and an image acquisition device are installed on a robot body, and the method includes the following steps S1 to S7:
s1, obtaining a vibration signal collected by a vibration sensor and a ground image signal collected by an image collecting device, and obtaining a vibration frame set and a ground image signal set;
it should be noted that the vibration sensor is used for detecting a vibration signal in a direction perpendicular to the ground, the image acquisition device may select a camera, a lens of the camera faces the ground and is used for shooting the ground where the robot is located, and the vibration sensor and the camera are both in an equal-time sampling working mode.
S2, converting the vibration frame into samples by adopting d-point Fourier transform, and obtaining a sample set X ═ X1,x2,…,xp,…,xPIn which xpIs a sample, specimen
Figure GDA0002701723550000071
Representing a real number set, d being a sample dimension;
s3, setting L ═ {1,2, …, c, …, L } to represent the set of ground type number values, c ═ 1,2, …, L, and establishing the state transition matrix
Figure GDA0002701723550000074
m,n＝1,2,…,l，Tm,nRepresenting the transition probability from state m to state n;
s4, setting a labeling interval delta according to the minimum element on the diagonal line in the state transition matrix, wherein delta is an integer;
s5, reading according to the marked interval delta
Figure GDA0002701723550000072
Artificial identification
Figure GDA0002701723550000073
Corresponding real ground type and pair xpLabeling to obtain a labeled sample set { xi,yiAnd unlabeled sample set { x }j},i＝1,2,…,I，j＝1,2,…,J，I+J＝P，yiIs a sample xiMarking of (1);
s6, training the constructed support vector machine model by using the labeled sample set and the unlabeled sample set to obtain a trained support vector machine;
and S7, processing the currently acquired vibration signals by using a trained support vector machine to realize the classification of the ground types.
It should be noted that, in this embodiment, the vibration signal acquired by the vibration sensor corresponds to the ground image signal acquired by the image acquisition device, and the real ground type corresponding to the ground image signal is manually identified, so as to label the vibration frame in the vibration frame set, thereby preventing a certain type of sample from lacking in labeling.
Further, the above step S1: the method comprises the following steps of obtaining vibration signals collected by a vibration sensor and ground image signals collected by image collection equipment, and obtaining a vibration frame set and a ground image signal set, wherein the method comprises the following steps:
converting a vibration signal into a set of vibration frames v1,v2,…,vp,…,vPIn which v ispIs a vibration frame, P is 1,2, …, P is the number of all vibration frames;
corresponding each vibration frame with the ground image signal according to the time stamp to obtain a set of the ground image signals
Figure GDA0002701723550000081
The vibration signal and the image signal are both time-stamped.
The robot is enabled to randomly walk on the ground expected to be identified, vibration signals and image signals are collected from the vibration sensor and the camera, the vibration signals and the image signals are all provided with time stamps, every S continuous vibration signals serve as a vibration frame, and the vibration signals are converted into a set of vibration frames according to { v & ltv & gt1,v2,…,vp,…,vP}。
Further, the above step S4: the marking interval Δ is set according to the smallest element on the diagonal in the state transition matrix, including the following subdivision steps S41-S42:
s41, calculating to satisfy
Figure GDA0002701723550000082
Corresponding to w, to obtain wminWherein, TminRepresents the smallest element on the diagonal in the state transition matrix, τ ∈ (0,1) is the threshold, w ═ 0,1,2, …;
s42, calculating the maximum labeling interval Wmax＝2wmin+1 and setting the marking interval Delta epsilon [1, W ∈ ]max]。
Further, in the above step S6: before training the constructed support vector machine model by using the labeled sample set and the unlabeled sample set, the method further comprises the following steps of:
Figure GDA0002701723550000091
wherein HKFor regenerating nuclear Hilbert space, V (x)i,yiF) is a loss function, | f |KIs the complexity metric norm, gamma, of f in the regenerative nuclear Hilbert spaceK,γA,γS>0；
Figure GDA0002701723550000092
θpqIs a characteristic similarity coefficient, xqFor the sample, q is 1,2, …, P,
Figure GDA0002701723550000093
is a spatial similarity coefficient.
Further, the thetapqThe calculation formula is as follows:
Figure GDA0002701723550000094
wherein,
Figure GDA0002701723550000095
is the distance x in the feature spaceqSet of most recent N samples, tθIs gaussian kernel width.
The spatial similarity coefficient
Figure GDA0002701723550000096
The calculation formula is as follows:
Figure GDA0002701723550000097
wherein,
Figure GDA0002701723550000098
is the width of the gaussian kernel and is,
Figure GDA0002701723550000099
for the y th diagonal of the state transition matrixiElement, p (x)i,xj) Is xiAnd xjThe sampling spatial distance therebetween.
It should be noted that the support vector machine model constructed in the embodiment better utilizes the assumption of spatial smoothness, thereby improving the classification accuracy.
As shown in fig. 2, the present embodiment discloses a field robot ground classification system based on a small number of labels, which includes: the robot comprises a data acquisition device 10 and a data processing device 20, wherein the data acquisition device 10 comprises a vibration sensor 11 and an image acquisition device 12 which are installed on a robot body, and the data processing device 20 comprises a data acquisition module 21, a sample construction module 22, a state transition matrix establishing module 23, a labeling interval determining module 24, a labeling module 25, a model training module 26 and a classification module 27;
the data acquisition module 21 is configured to acquire a vibration signal acquired by the vibration sensor and a ground image signal acquired by the image acquisition device, and obtain a vibration frame set and a ground image signal set;
the sample construction module 22 is configured to convert the vibration frame into a sample by using d-point fourier transform, and obtain a sample set X ═ X1,x2,…,xp,…,xPIn which xpIs a sample, specimen
Figure GDA0002701723550000101
Representing a real number set, d being a sample dimension;
the state transition matrix establishing module 23 is configured to set L ═ {1,2, …, c, …, L } to represent a set of tag values, and c ═ 1,2, …, L to establish the state transition matrix
Figure GDA0002701723550000102
m,n＝1,2,…,l，Tm,nRepresenting the transition probability from state m to state n;
the labeling interval determining module 24 is configured to set a labeling interval Δ according to a minimum element on a diagonal line in the state transition matrix, where Δ is an integer;
the marking module 25 is used for reading according to the marking interval delta
Figure GDA0002701723550000103
Artificial identification
Figure GDA0002701723550000104
Corresponding real ground type and pair xpLabeling to obtain a labeled sample set { xi,yiAnd unlabeled sample set { x }j},i＝1,2,…,I，j＝1,2,…,J，I+J＝P，yiIs a sample xiMarking of (1);
the model training module 26 is configured to train the constructed support vector machine model by using the labeled sample set and the unlabeled sample set to obtain a trained support vector machine;
the classification module 27 is configured to process the currently acquired vibration signal by using a trained support vector machine, so as to implement classification of the ground type.
Specifically, vibration sensor 11 is used for detecting the vibration signal of perpendicular to ground direction, and image acquisition equipment 12 chooses for use the camera, and the camera lens is towards ground for shoot the current ground of locating of robot, and vibration sensor and camera are the equal time sampling mode. The robot is enabled to randomly walk on the ground expected to be identified, vibration signals and image signals are collected from the vibration sensor and the camera, the vibration signals and the image signals are provided with time stamps, every S continuous vibration signals serve as a vibration frame, and the vibration signals are converted into a set of vibration frames according to { v }1,v2,…,vp,…,vP}。
Further, the labeling interval determining module 24 is specifically configured to:
computing satisfaction
Figure GDA0002701723550000111
Corresponding to w, to obtain wminWherein, TminRepresents the smallest element on the diagonal in the state transition matrix, τ ∈ (0,1) is the threshold, w ═ 0,1,2, …;
calculating the maximum annotation separation Wmax＝2wmin+1 and setting the marking interval Delta epsilon [1, W ∈ ]max]。
Further, the system also comprises a support vector machine model building module, wherein the support vector machine model building module is used for building a support vector machine model as follows:
Figure GDA0002701723550000112
wherein HKFor regenerating nuclear Hilbert space, V (x)i,yiF) is a loss function, | f |KIs the complexity metric norm, gamma, of f in the regenerative nuclear Hilbert spaceK,γA,γS>0；
Figure GDA0002701723550000113
θpqIs a characteristic similarity coefficient, xqFor the sample, q is 1,2, …, P,
Figure GDA0002701723550000114
is a spatial similarity coefficient.
Theta is describedpqThe calculation formula is as follows:
Figure GDA0002701723550000121
wherein,
Figure GDA0002701723550000122
is the distance x in the feature spaceqSet of most recent N samples, tθIs gaussian kernel width.
The spatial similarity coefficient
Figure GDA0002701723550000123
The calculation formula is as follows:
Figure GDA0002701723550000124
wherein,
Figure GDA0002701723550000125
is the width of the gaussian kernel and is,
Figure GDA0002701723550000126
for the y th diagonal of the state transition matrixiElement, p (x)i,xj) Is xiAnd xjThe sampling spatial distance therebetween.
It should be noted that, in this embodiment, the vibration sensor installed on the robot body is used to collect vibration signals, and the ground type identified by the image is used to label the vibration frame, so as to obtain a labeled sample and a label-free sample, and train the support vector machine, so that the assumption of spatial smoothness is better utilized, the classification accuracy is improved, meanwhile, the actual condition of data classification is well reflected by sample classification, and the influence of lack of labeling of a certain type of sample on the classification result is avoided. It should be understood that the scheme of the embodiment can also be applied to underground lithology recognition, satellite hyperspectral image classification and the like to improve the classification accuracy.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (8)

1. A field robot ground classification method based on a small amount of labels is characterized in that a vibration sensor and image acquisition equipment are installed on a robot body, and the method comprises the following steps:
acquiring a vibration signal acquired by a vibration sensor and a ground image signal acquired by image acquisition equipment, and acquiring a vibration frame set and a ground image signal set;
converting the vibration frame into a sample by adopting d-point Fourier transform to obtain a sample set X ═ X1，x2，…，xp，…，xPIn which xpIs a sample, specimen
Figure FDA0002701723540000011
Figure FDA0002701723540000012
Representing a real number set, d being a sample dimension;
let L ═ {1,2, …, c, …, L } denote the set of ground type number values, c ═ 1,2, …, L, establish the state transition matrix
Figure FDA0002701723540000018
m，n＝1，2，…，l，Tm，nRepresenting the transition probability from state m to state n;
setting a labeling interval delta according to the smallest element on the diagonal in the state transition matrix, wherein delta is an integer, and the method comprises the following steps:
computing satisfaction
Figure FDA0002701723540000013
Corresponding to w, to obtain wminWherein, TminDenotes the smallest diagonal element in the state transition matrix, τ ∈ (0,1) is the threshold, w ∈ 0,1,2, …, wminRepresents the smallest element in the w set;
Figure FDA0002701723540000014
the smallest element on the diagonal in the state transition matrix at the w-th time is represented;
calculating the maximum annotation separation Wmax＝2wmin+1 and setting the marking interval Delta epsilon [1, W ∈ ]max]；
Read at mark interval Δ
Figure FDA0002701723540000015
Artificial identification
Figure FDA0002701723540000016
Corresponding real ground type and pair xpLabeling to obtain a labeled sample set { xi，yiAnd unlabeled sample set { x }j}，i＝1，2，…，I，j＝1，2，…，J，I+J＝P，yiIs a sample xiThe labeling of (a), wherein,
Figure FDA0002701723540000017
elements in a set of ground image signals;
training the constructed support vector machine model by using the labeled sample set and the unlabeled sample set to obtain a trained support vector machine;
and processing the currently acquired vibration signals by using a trained support vector machine to realize the classification of the ground types.
2. The field robot ground classification method based on small amount of labels as claimed in claim 1, wherein said obtaining vibration signals collected by a vibration sensor and ground image signals collected by an image collecting device and obtaining a set of vibration frames and a set of ground image signals comprises:
converting a vibration signal into a set of vibration frames v1，v2，…，vp，…，vPIn which v ispIs a vibration frame, P is 1,2, …, P is the number of all vibration frames;
corresponding each vibration frame with the ground image signal according to the time stamp to obtain a set of the ground image signals
Figure FDA0002701723540000021
The vibration signal and the image signal are both time-stamped.
3. The ground classification method for a field robot based on small amount of labels of claim 1, wherein before the training of the constructed support vector machine model by using the labeled sample set and the unlabeled sample set, the ground classification method further comprises the following steps:
Figure FDA0002701723540000022
wherein HKFor regenerating nuclear Hilbert space, V (x)i，yiF) is a loss function, | f | | non-woven phosphorKIs the complexity metric norm, gamma, of f in the regenerative nuclear Hilbert spaceK，γA，γS＞0；
Figure FDA0002701723540000023
θpqIs a characteristic similarity coefficient, xqFor the sample, q is 1,2, …, P,
Figure FDA0002701723540000024
Figure FDA0002701723540000025
is a spatial similarity coefficient.
4. The field robot ground classification method based on small number of labels of claim 3, characterized in that θ ispqThe calculation formula is as follows:
Figure FDA0002701723540000031
wherein,
Figure FDA0002701723540000032
is the distance x in the feature spaceqSet of most recent N samples, tθIs the gaussian kernel width; the spatial similarity coefficient
Figure FDA0002701723540000033
The calculation formula is as follows:
Figure FDA0002701723540000034
wherein,
Figure FDA0002701723540000035
is the width of the gaussian kernel and is,
Figure FDA0002701723540000039
for the y th diagonal of the state transition matrixiElement, p (x)i，xj) Is xiAnd xjThe sampling spatial distance therebetween.
5. A field robot ground classification system based on a small amount of labels, comprising: the robot comprises data acquisition equipment and data processing equipment, wherein the data acquisition equipment comprises a vibration sensor and image acquisition equipment which are installed on a robot body, and the data processing equipment comprises a data acquisition module, a sample construction module, a state transition matrix establishing module, a labeling interval determining module, a labeling module, a model training module and a classification module;
the data acquisition module is used for acquiring vibration signals acquired by the vibration sensor and ground image signals acquired by the image acquisition equipment and acquiring a vibration frame set and a ground image signal set;
the sample construction module is used for converting the vibration frame into a sample by adopting d-point Fourier transform to obtain a sample set X ═ X1，x2，…，xp，…，xPIn which xpIs a sample, specimen
Figure FDA0002701723540000036
Figure FDA0002701723540000037
Representing a real number set, d being a sample dimension;
the state transition matrix establishing module is used for setting L to be {1,2, …, c, …, L } to represent a set of label values, c to be 1,2, …, L, and establishing the state transition matrix
Figure FDA0002701723540000038
m，n＝1，2，…，l，Tm，nRepresenting the transition probability from state m to state n;
the labeling interval determining module is configured to set a labeling interval Δ according to a minimum element on a diagonal line in the state transition matrix, where Δ is an integer, and includes:
computing satisfaction
Figure FDA0002701723540000041
Corresponding to w, to obtain wminWherein, TminDenotes the smallest diagonal element in the state transition matrix, τ ∈ (0,1) is the threshold, w ∈ 0,1,2, …, wminRepresents the smallest element in the w set;
Figure FDA0002701723540000042
the smallest element on the diagonal in the state transition matrix at the w-th time is represented;
calculating the maximum annotation separation Wmax＝2wmin+1 and setting the marking interval Delta epsilon [1, W ∈ ]max]；
The marking module is used for reading according to the marking interval delta
Figure FDA0002701723540000043
Artificial identification
Figure FDA0002701723540000044
Corresponding real ground type and pair xpLabeling to obtain a labeled sample set { xi，yiAnd unlabeled sample set { x }j}，i＝1，2，…，I，j＝1，2，…，J，I+J＝P，yiIs a sample xiThe labeling of (a), wherein,
Figure FDA0002701723540000045
elements in a set of ground image signals;
the model training module is used for training the constructed support vector machine model by utilizing the marked sample set and the unmarked sample set to obtain a trained support vector machine;
the classification module is used for processing the currently acquired vibration signals by using a trained support vector machine to realize the classification of the ground types.
6. The field robot ground classification system based on small number of labels of claim 5, characterized in that, the data acquisition module comprises a vibration frame set construction unit and a ground image signal set construction unit;
the vibration frame set construction unit is used for converting the vibration signal into a set of vibration frames { v1，v2，…，vp，…，vPIn which v ispIs a vibration frame, P is 1,2, …, P is the number of all vibration frames;
the ground image signal set construction unit is used for corresponding each vibration frame with the ground image signal according to the time stamp to obtain a set of the ground image signal
Figure FDA0002701723540000051
The vibration signal and the image signal are both time-stamped.
7. The field robot ground classification system based on small number of labels of claim 5, further comprising a support vector machine model construction module for constructing a support vector machine model as follows:
Figure FDA0002701723540000052
wherein HKFor regenerating nuclear Hilbert space, V (x)i，yiF) is a loss function, | f | | non-woven phosphorKIs the complexity metric norm, gamma, of f in the regenerative nuclear Hilbert spaceK，γA，γS＞0；
Figure FDA0002701723540000053
θpqIs a characteristic similarity coefficient, xqFor the sample, q is 1,2, …, P,
Figure FDA0002701723540000054
Figure FDA0002701723540000055
is a spatial similarity coefficient.
8. The field robotic ground classification system based on low number of labels of claim 7, wherein θ ispqThe calculation formula is as follows:
Figure FDA0002701723540000056
wherein,
Figure FDA0002701723540000057
is the distance x in the feature spaceqSet of most recent N samples, tθIs the gaussian kernel width;
the spatial similarity coefficient
Figure FDA0002701723540000058
The calculation formula is as follows:
Figure FDA0002701723540000061
wherein,
Figure FDA0002701723540000062
is the width of the gaussian kernel and is,
Figure FDA0002701723540000063
for the y th diagonal of the state transition matrixiElement, p (x)i，xj) Is xiAnd xjThe sampling spatial distance therebetween.
CN201910995967.0A 2019-10-18 2019-10-18 Method and system for field robot ground classification based on small amount of labels Active CN110781788B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910995967.0A CN110781788B (en) 2019-10-18 2019-10-18 Method and system for field robot ground classification based on small amount of labels

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910995967.0A CN110781788B (en) 2019-10-18 2019-10-18 Method and system for field robot ground classification based on small amount of labels

Publications (2)

Publication Number Publication Date
CN110781788A CN110781788A (en) 2020-02-11
CN110781788B true CN110781788B (en) 2020-12-25

Family

ID=69386074

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910995967.0A Active CN110781788B (en) 2019-10-18 2019-10-18 Method and system for field robot ground classification based on small amount of labels

Country Status (1)

Country Link
CN (1) CN110781788B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111814894B (en) * 2020-07-17 2022-09-09 福州大学 Multi-view semi-supervised classification method for rapid seed random walk
CN112101263B (en) * 2020-09-22 2022-09-06 中国科学技术大学 Active learning sample selection method and system and computer readable storage medium
CN113537240B (en) * 2021-07-09 2023-09-05 北方工业大学 Deformation zone intelligent extraction method and system based on radar sequence image

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108036789A (en) * 2017-11-29 2018-05-15 安徽省通信息科技有限公司 A kind of field robot reckoning method
WO2019113572A1 (en) * 2017-12-08 2019-06-13 Geomni, Inc. Computer vision systems and methods for geospatial property feature detection and extraction from digital images
CN110059385A (en) * 2019-04-09 2019-07-26 集美大学 A kind of urban cellular automata Scene Simulation method and terminal device

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105426858A (en) * 2015-11-26 2016-03-23 哈尔滨工业大学 Vision and vibration information fusion based ground type identification method
JP6941123B2 (en) * 2016-06-30 2021-09-29 コニカ ミノルタ ラボラトリー ユー．エス．エー．，インコーポレイテッド Cell annotation method and annotation system using adaptive additional learning
CN108564115A (en) * 2018-03-30 2018-09-21 西安电子科技大学 Semi-supervised polarization SAR terrain classification method based on full convolution GAN
CN108960294B (en) * 2018-06-12 2019-07-23 中国科学技术大学 A kind of mobile robot classification of landform method of view-based access control model

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108036789A (en) * 2017-11-29 2018-05-15 安徽省通信息科技有限公司 A kind of field robot reckoning method
WO2019113572A1 (en) * 2017-12-08 2019-06-13 Geomni, Inc. Computer vision systems and methods for geospatial property feature detection and extraction from digital images
CN110059385A (en) * 2019-04-09 2019-07-26 集美大学 A kind of urban cellular automata Scene Simulation method and terminal device

Also Published As

Publication number Publication date
CN110781788A (en) 2020-02-11

Similar Documents

Publication Publication Date Title
CN110781788B (en) Method and system for field robot ground classification based on small amount of labels
CN105404886B (en) Characteristic model generation method and characteristic model generating means
CN111783576A (en) Pedestrian re-identification method based on improved YOLOv3 network and feature fusion
CN105354548A (en) Surveillance video pedestrian re-recognition method based on ImageNet retrieval
CN111339975A (en) Target detection, identification and tracking method based on central scale prediction and twin neural network
WO2022218396A1 (en) Image processing method and apparatus, and computer readable storage medium
CN111639629B (en) Pig weight measurement method and device based on image processing and storage medium
CN108711172B (en) Unmanned aerial vehicle identification and positioning method based on fine-grained classification
CN113538585B (en) High-precision multi-target intelligent identification, positioning and tracking method and system based on unmanned aerial vehicle
CN103745238A (en) Pantograph identification method based on AdaBoost and active shape model
CN111768429A (en) Pedestrian target tracking method in tunnel environment based on Kalman filtering and pedestrian re-identification algorithm
CN114662594B (en) Target feature recognition analysis system
CN112598051B (en) Cross-domain SAR image terrain classification method based on polarization statistical scattering components
CN106971150B (en) Queuing abnormity detection method and device based on logistic regression
Aljelawy et al. License plate recognition in slow motion vehicles
CN116310263A (en) Pointer type aviation horizon instrument indication automatic reading implementation method
CN110781942B (en) Semi-supervised image classification method and system
CN116503358A (en) Method for detecting rust of power transmission tower based on attention mechanism fusion
CN110826478A (en) Aerial photography illegal building identification method based on countermeasure network
CN115565008A (en) Transferable image recognition detection system, method and computer readable storage medium
CN114782500A (en) Kart race behavior analysis method based on multi-target tracking
CN115063624A (en) Small sample classification learning method based on graph neural network
CN114494355A (en) Trajectory analysis method and device based on artificial intelligence, terminal equipment and medium
CN113963230A (en) Parking space detection method based on deep learning
CN112560969A (en) Image processing method for human weight recognition, model training method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20220711

Address after: 230026 Jinzhai Road, Baohe District, Hefei, Anhui Province, No. 96

Patentee after: University of Science and Technology of China

Patentee after: Kang Yu

Patentee after: Lv Wenjun

Address before: 230026 Jinzhai Road, Baohe District, Hefei, Anhui Province, No. 96

Patentee before: University of Science and Technology of China

TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20230109

Address after: Room 1001, Building 16, Wenqingyuan, Keda Garden, No. 130, Tongcheng South Road, Baohe District, Hefei City, Anhui Province, 230051

Patentee after: Kang Yu

Patentee after: Lv Wenjun

Address before: 230026 Jinzhai Road, Baohe District, Hefei, Anhui Province, No. 96

Patentee before: University of Science and Technology of China

Patentee before: Kang Yu

Patentee before: Lv Wenjun

TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20230228

Address after: Building 7, Yousi Tiancheng Industrial Park, 1800 Dabieshan Road, High-tech Zone, Hefei, Anhui Province, 230088

Patentee after: Hefei Zhongke Liheng Intelligent Technology Co.,Ltd.

Address before: Room 1001, Building 16, Wenqingyuan, Keda Garden, No. 130, Tongcheng South Road, Baohe District, Hefei City, Anhui Province, 230051

Patentee before: Kang Yu

Patentee before: Lv Wenjun