CN108830908A - A kind of magic square color identification method based on artificial neural network - Google Patents
A kind of magic square color identification method based on artificial neural network Download PDFInfo
- Publication number
- CN108830908A CN108830908A CN201810623065.XA CN201810623065A CN108830908A CN 108830908 A CN108830908 A CN 108830908A CN 201810623065 A CN201810623065 A CN 201810623065A CN 108830908 A CN108830908 A CN 108830908A
- Authority
- CN
- China
- Prior art keywords
- color
- picture
- magic square
- neural network
- face
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses the magic square color identification method based on artificial neural network, including step：Build artificial neural network training pattern；Acquire the picture in each face of magic square, color lump is irised out along each color lump profile of magic square on picture with rectangle callout box, each color lump color category is indicated by kind of a class-mark, and the upper left corner of rectangle callout box and bottom right angular coordinate and corresponding kind of class-mark are stored in the corresponding mark file of each picture；Data set is formed by the training of artificial neural network training pattern using mark file and corresponding picture, is formed network model file and is stored to mobile platform；When identification, mobile platform acquires the picture in each face of magic square to be processed with camera in order, after one face picture of every acquisition, call identification with network model file to the measurement picture recognition in the face, the face color array is exported, the output of two-dimensional color rectangle is ultimately formed, completes identification.Magic square robot color recognizer of the present invention introduces neural network and improves discrimination under complex environment.
Description
Technical field
The present invention relates to robot graphics' identification technology fields, and in particular to a kind of magic square face based on artificial neural network
Color recognition methods.
Background technique
Magic square is a kind of protean puzzle, also referred to as Lu Bike square, and 1974 by Budapest, HUN
Architecture institute Shandong Bick teaching inventive.Since its is extremely subtle, there is large quantities of fans, also has to magic square extensively in academia
General and in-depth study, and the research achievement of magic square has also obtained many applications.
Magic square robot is that magic square image is obtained, handled and known based on image analysis and computer vision
Not, and to magic square information it is comprehensively analyzed and is managed, the magic square arbitrarily upset is restored automatically.
Some reduction magic squares robot is higher to environmental requirement when identifying magic square color, is substantially all needs specific
It is carried out under illumination condition, it is not strong to the adaptability of environment.
There are many modes classified to color, and wherein RGB color is most basic, most common color space,
However, for color perception properties, RGB color be it is heterogeneous, the correlation between each color component is stronger, thus
It is generally used in the identification model for target being divided into two classes.Hsv color space is another more commonly used color space, its base
In people psychological response characteristic and establish, belong to polar coordinate space structure, its advantage is that color intuitively can be described, still, by
In there are singular points in hsv color space, meanwhile, when bright and dark light changes, certain component value in light is slightly higher, to color point
Class will appear many mistakes.Therefore it is also not suitable for for realizing classification and quantization to color.
How to realize the classification for the color under environment complicated and changeable, the identification of magic square color is realized, so that magic square machine
Device people can accurately restore magic square, have great importance.
Summary of the invention
In view of the technical drawbacks of the prior art, it is an object of the present invention to provide one kind to be based on artificial neural network
Magic square color identification method.
The technical solution adopted to achieve the purpose of the present invention is：
A kind of magic square color identification method based on artificial neural network, including step：
Artificial neural network training pattern required for building；
The picture for acquiring each face of magic square, with rectangle callout box along each color lump profile circle of magic square on the picture of acquisition
9 color lumps out, and indicate by kind of class-mark the color category of each color lump, by the top left co-ordinate of the rectangle callout box, the lower right corner
Coordinate and corresponding kind of class-mark are stored in the corresponding mark file of each picture；
It is formed by data set using the mark file and corresponding picture, to the artificial neural network built
Network training pattern is trained, and is formed identification and is stored with network model file into mobile platform；
When identification, mobile platform acquires the picture to be determined in each face of magic square to be processed by camera in order, often adopts
After the picture for collecting a face, identification is called, to measurement picture recognition corresponding to the face, to export this face with network model file
Color array, ultimately form the output of two-dimensional color rectangle, complete identification.
The artificial neural network training pattern is built using Linux system, corresponding, and the mobile platform uses
Linux kernel system, in order to call the identification network model file.
Be trained to the artificial neural network training pattern built is instructed using FASTER-RCNN network
Practice.
A kind of magic square color identification method based on artificial neural network proposed by the present invention, offer platform base it
On, the picture that camera is read infers that the varying environment difference light selected when according to training pattern is not using network model
With the picture of color lump combination, the shortcomings that tradition OpenCV recognition methods is illuminated by the light influence, the identification for magic square profile can avoid
It handles also more intelligent.Infer that speed is fast simultaneously, and accurate, carries out place mat work for the resolving of next step magic square.It does not need
The relative position of magic square and camera is accurately adjusted, only need to guarantee that magic square is wrapped within the scope of camera view,
Guaranteeing the constant situation of original magic square robot architecture, improving color identification module, so that whole system is more accurate more steady
It is fixed.
Detailed description of the invention
Fig. 1 is the flow chart of the magic square color identification method based on artificial neural network；
Fig. 2 is schematic diagram of the magic square color lump in rectangle mark frame in image labeling processing；
Fig. 3 is the control of the magic square color identification method progress magic square recovery using of the invention based on artificial neural network
The hardware structure diagram of system.
Specific embodiment
The present invention is described in further detail below in conjunction with the drawings and specific embodiments.It should be appreciated that described herein
Specific embodiment be only used to explain the present invention, be not intended to limit the present invention.
Referring to shown in Fig. 1-2, a kind of magic square color identification method based on artificial neural network includes the following steps：
A. artificial neural network training pattern is built in LINUX operating system.
B. using the picture of camera acquisition magic square, that is, acquire what the different color blocks under varying environment and different illumination combined
The picture in the single face of magic square.Since the position of magic square robot camera and magic square is relatively-stationary, so the picture of acquisition
General Central position be magic square wherein one side, peripheral part has a small amount of environment sundries to influence, by the way that environment sundries is arranged, comes
Make influence of the Network Study Environment for magic square color lump value.
C. the picture of step B acquisition is handled as follows：
Picture is overturn, each color lump profile of magic square using rectangle callout box, and respectively on each picture irises out 9 colors
Block, and indicate the color category of color lump, for example, it is red, orange, yellow, green, black, white, blue be respectively set to kind of class-mark 0,1,2,3,4,5,
6, by the top left co-ordinate of rectangle callout box, bottom right angular coordinate and corresponding color category number, it is corresponding to be stored in each picture
Mark file.Because there are 9 color lumps on the face of a magic square, therefore have in a mark file and only 9 rectangle callout box.
It is as follows to mark file part format：Wherein name is type, is the coordinate of rectangle callout box in bndbox.Such as Fig. 2
The shown processing mode described to the picture containing magic square color lump, 2 indicate magic square color lump, and 1 is rectangle callout box.It needs with 9
A rectangle callout box separately includes the color lump and mark of different colours.The picture and corresponding mark file constitute training
The data set (or training set) of sample.
D. the neural network training model of invocation step A creation, is trained using the data set that step C is obtained, training
After the completion, the network model file for training and saving is imported into mobile platform, specific implementation is trained by python, raw
At corresponding network model file.Mobile platform equally uses linux kernel system, is convenient for calling model.
The present invention uses FASTER-RCNN network, and FASTER-RCNN is by Area generation network RPN and FAST RCNN network
Zu Cheng.Here four step coaching methods are used：
1) the image labeling data that step C is generated are input to be trained in RPN basic network and concentrate candidate region
It extracts, i.e. the extraction of rectangle callout box passes through the input of the pre-training model load networks FAST RCNN network of open source.To mesh
Before until, the two networks shared convolutional layer not yet, but receive training respectively.
2) RPN is trained further according to training set, fixes the shared convolution layer parameter of two networks, only update RPN its
The parameter of his unique portion, i.e., the parameter in addition to convolutional layer
3) candidate region extracted again according to RPN is finely adjusted FAST RCNN network parameter, fixes two networks
Shared shared convolutional layer parameter constant, is finely adjusted only for the parameter of FAST RCNN unique portion.
Specific training parameter is as follows：Training uses original FASTER RCNN network, sets the number of iterations as 5000.It learns
Habit rate is set as 0.001.Training uses original network, but is combined to for trained sample image using different magic square color lumps
As input, BATCH SIZE is set as 64.Final training result is to generate a unique network model file.
E. camera acquire picture, be passed to mobile platform control module-single-chip microcontroller, single-chip microcontroller invocation step D-shaped at
Network model file, carries out identifying processing, and the network model file obtained by step D training pattern irises out 9 rectangle marks
Frame, and press and export type from top to bottom from left to right to get the color array in a face out.
Specific identification process is as follows：Image to be determined is input in the network model file of step D output, this same net
Network model includes RPN and FAST RCNN model, initially enters the extraction that RPN network carries out candidate region, is existed according to characteristics of image
Different feature vectors is generated at sliding window, and the full articulamentum that each feature vector is sent into FAST RCNN is given a forecast, and is detected
The magic square square of different colours whether there is, and then classify to the target area that may be present of prediction, and to target area
Domain carries out non-maxima suppression, and the last judgement of image result is carried out according to the accuracy rate threshold value of identification, exports type, forms face
Color array.One face color recognition result of final output is the one-dimensional color array being made of 9 elements.
F. the other faces for converting magic square repeat step E and successively carry out other identifications, finally obtain the two-dimensional color square of 6*9
Battle array.
As shown in figure 3, describing the hardware system of entire recognition process unit, image transmitting is acquired by image capture module
To single chip control module, the neural network model file process image of importing is called in single chip control module, is known
Not, color array is then calculated according to algorithm, the steering engine module progress control flaps for being sent to the manipulator of magic square robot are motor-driven
Make, to control corresponding manipulator behavior, realizes the recovery operation processing of magic square.
The magic square identifying schemes based on artificial neural network proposed in the present invention are focused on to not sharing the same light under varying environment
Image according under carries out feature extraction and study, is compared and is learnt by mass data, show that one can identify under varying environment
The classifier of color, to carry out the identification decision of color.And traditional OpenCV identifies color of image, only resides within adjustment
Threshold value defines color gamut, adjusts once changing a kind of environment and will carry out new threshold value, is otherwise likely to occur identification mistake.
The above is only a preferred embodiment of the present invention, it is noted that for the common skill of the art
For art personnel, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications
Also it should be regarded as protection scope of the present invention.
Claims (3)
1. a kind of magic square color identification method based on artificial neural network, which is characterized in that including step：
Artificial neural network training pattern required for building；
The picture for acquiring each face of magic square, irises out 9 with each color lump profile of the rectangle callout box on the picture of acquisition along magic square
A color lump, and indicate by kind of class-mark the color category of each color lump, by the top left co-ordinate of the rectangle callout box, bottom right angular coordinate
And corresponding kind of class-mark, it is stored in the corresponding mark file of each picture；
It is formed by data set using the mark file and corresponding picture, the artificial neural network built is instructed
Practice model to be trained, forms identification and stored with network model file into mobile platform；
When identification, mobile platform acquires the picture to be determined in each face of magic square to be processed, every acquisition one by camera in order
After the picture in a face, identification is called, to measurement picture recognition corresponding to the face, to export the face in this face with network model file
Chromatic number group ultimately forms a two-dimensional color rectangle output, completes identification.
2. the magic square color identification method based on artificial neural network as described in claim 1, which is characterized in that the artificial mind
It is built through network training model using Linux system, corresponding, the mobile platform uses linux kernel system, in order to adjust
With identification network model file.
3. the magic square color identification method based on artificial neural network as described in claim 1, which is characterized in that the institute built
The artificial neural network training pattern stated, which is trained, to be trained using FASTER-RCNN network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810623065.XA CN108830908A (en) | 2018-06-15 | 2018-06-15 | A kind of magic square color identification method based on artificial neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810623065.XA CN108830908A (en) | 2018-06-15 | 2018-06-15 | A kind of magic square color identification method based on artificial neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108830908A true CN108830908A (en) | 2018-11-16 |
Family
ID=64142337
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810623065.XA Pending CN108830908A (en) | 2018-06-15 | 2018-06-15 | A kind of magic square color identification method based on artificial neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108830908A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109684971A (en) * | 2018-12-18 | 2019-04-26 | 北京理工大学珠海学院 | Magic cube-solving robot algorithm executes method |
| CN110991374A (en) * | 2019-12-10 | 2020-04-10 | 电子科技大学 | Fingerprint singular point detection method based on RCNN |
| CN111383352A (en) * | 2020-03-20 | 2020-07-07 | 北京工业大学 | Automatic color filling and abstracting method for three-order magic cube |
| CN111950654A (en) * | 2020-08-25 | 2020-11-17 | 桂林电子科技大学 | Magic cube color block color reduction method based on SVM classification |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150186726A1 (en) * | 2012-09-14 | 2015-07-02 | Huawei Technologies Co., Ltd. | Scene Recognition Method and Apparatus |
| CN107729801A (en) * | 2017-07-11 | 2018-02-23 | 银江股份有限公司 | A kind of vehicle color identifying system based on multitask depth convolutional neural networks |
| CN107909580A (en) * | 2017-11-01 | 2018-04-13 | 深圳市深网视界科技有限公司 | A kind of pedestrian wears color identification method, electronic equipment and storage medium clothes |
| CN108079581A (en) * | 2017-12-15 | 2018-05-29 | 北京小米移动软件有限公司 | Intelligent magic cube, magic square control method, equipment and storage medium |
-
2018
- 2018-06-15 CN CN201810623065.XA patent/CN108830908A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150186726A1 (en) * | 2012-09-14 | 2015-07-02 | Huawei Technologies Co., Ltd. | Scene Recognition Method and Apparatus |
| CN107729801A (en) * | 2017-07-11 | 2018-02-23 | 银江股份有限公司 | A kind of vehicle color identifying system based on multitask depth convolutional neural networks |
| CN107909580A (en) * | 2017-11-01 | 2018-04-13 | 深圳市深网视界科技有限公司 | A kind of pedestrian wears color identification method, electronic equipment and storage medium clothes |
| CN108079581A (en) * | 2017-12-15 | 2018-05-29 | 北京小米移动软件有限公司 | Intelligent magic cube, magic square control method, equipment and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 张雪娇: ""智能魔方机器人的视觉感知与复原算法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109684971A (en) * | 2018-12-18 | 2019-04-26 | 北京理工大学珠海学院 | Magic cube-solving robot algorithm executes method |
| CN109684971B (en) * | 2018-12-18 | 2023-01-24 | 北京理工大学珠海学院 | Magic cube solving robot algorithm execution method |
| CN110991374A (en) * | 2019-12-10 | 2020-04-10 | 电子科技大学 | Fingerprint singular point detection method based on RCNN |
| CN111383352A (en) * | 2020-03-20 | 2020-07-07 | 北京工业大学 | Automatic color filling and abstracting method for three-order magic cube |
| CN111383352B (en) * | 2020-03-20 | 2023-09-26 | 北京工业大学 | Automatic color filling and abstraction method for third-order magic cube |
| CN111950654A (en) * | 2020-08-25 | 2020-11-17 | 桂林电子科技大学 | Magic cube color block color reduction method based on SVM classification |
| CN111950654B (en) * | 2020-08-25 | 2022-07-05 | 桂林电子科技大学 | Magic cube color block color reduction method based on SVM classification |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108108761B (en) | Rapid traffic signal lamp detection method based on deep feature learning | |
| CN108830908A (en) | A kind of magic square color identification method based on artificial neural network | |
| CN108717524B (en) | Gesture recognition system based on double-camera mobile phone and artificial intelligence system | |
| CN104484658A (en) | Face gender recognition method and device based on multi-channel convolution neural network | |
| CN106384117B (en) | A kind of vehicle color identification method and device | |
| CN104598924A (en) | Target matching detection method | |
| CN108671534A (en) | A kind of robot Chinese chess beginning pendulum chess method and system based on objective contour and framework characteristic | |
| CN110032932B (en) | Human body posture identification method based on video processing and decision tree set threshold | |
| CN110287787B (en) | Image recognition method, image recognition device and computer-readable storage medium | |
| CN110956099B (en) | Dynamic gesture instruction identification method | |
| CN105426924B (en) | A kind of scene classification method based on image middle level features | |
| CN107240112A (en) | Individual X Angular Point Extracting Methods under a kind of complex scene | |
| CN109598200B (en) | Intelligent image identification system and method for molten iron tank number | |
| CN109325408A (en) | A kind of gesture judging method and storage medium | |
| CN110399888A (en) | A kind of go judgment system based on MLP neural network and computer vision | |
| CN108764447A (en) | A kind of group robot Majiang game intelligence dynamicization system and mahjong identification learning algorithm | |
| CN109284759A (en) | One kind being based on the magic square color identification method of support vector machines (svm) | |
| CN108154496B (en) | Electric equipment appearance change identification method suitable for electric power robot | |
| GUO et al. | Traffic signs recognition based on visual attention mechanism | |
| CN108274476A (en) | A kind of method of anthropomorphic robot crawl sphere | |
| Fernandes et al. | Identification and Sorting of Objects based on Shape and Colour using robotic arm | |
| CN106650824B (en) | Moving object classification method based on support vector machines | |
| Mukhopadhyay et al. | Comparing shape descriptor methods for different color space and lighting conditions | |
| CN104318267B (en) | A kind of automatic identification system of Tibetan mastiff pup purity | |
| CN110689001B (en) | Method for generating license plate training sample in complex 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: 20181116 |