CN110503047A - A kind of rds data processing method and processing device based on machine learning - Google Patents
A kind of rds data processing method and processing device based on machine learning Download PDFInfo
- Publication number
- CN110503047A CN110503047A CN201910790311.5A CN201910790311A CN110503047A CN 110503047 A CN110503047 A CN 110503047A CN 201910790311 A CN201910790311 A CN 201910790311A CN 110503047 A CN110503047 A CN 110503047A
- Authority
- CN
- China
- Prior art keywords
- area
- interest
- processed
- picture
- sample set
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (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 provides a kind of rds data processing method and processing device based on machine learning, the method includes obtaining pictures to be processed, intercepts and obtains in the video flowing that the picture to be processed is shot by the camera being laid in around rail;Each picture to be processed in the pictures to be processed is inputted into preset image processing model to obtain the area-of-interest recommendation results of the picture to be processed, the area-of-interest recommendation results include area-of-interest and the corresponding index interested of area-of-interest in picture to be processed;Obtain the post analysis preprocessing rule of pictures to be processed;The corresponding Target Photo of each picture to be processed is obtained according to the area-of-interest recommendation results of post analysis preprocessing rule and each picture to be processed and then obtains Target Photo collection.Present invention is particularly suitable for the information in analysis below around depth excavation railway, reaches and reduced the region that rds data information loses and avoids analysis richness low, promoted the purpose of analysis efficiency.
Description
Technical field
The present invention relates to railway territory more particularly to a kind of rds data processing method and processing devices based on machine learning.
Background technique
With the growth and video technique of railway video monitoring business, the hair of Internet technology and Intellectual Analysis Technology
Exhibition, produces a large amount of image datas comprising Railway Environment information, and the image data of magnanimity automatically analyzes and information excavating
Importance is increasingly prominent, correspondingly, be it is still inadequate to the automatic processing ability of rds data in the prior art, therefore, produce
The loss of a large amount of Railway Environment information has been given birth to, and the automatic processing of rds data is embodied in the automatic metaplasia of area-of-interest
At many aspects such as area-of-interest automated analysis and analysis result are comprehensive.
Summary of the invention
The technical issues of in order to solve automatic processing scarce capacity to rds data in the prior art, is realized interested
The automation in region generates, in order to realize magnanimity image data primary dcreening operation, the present invention provides a kind of based on machine learning
Rds data processing method and processing device.
The present invention is realized with following technical solution:
A kind of rds data processing method based on machine learning, which comprises
Pictures to be processed are obtained, in the video flowing that the picture to be processed is shot by the camera being laid in around rail
Interception obtains;
It is described to obtain that each picture to be processed in the pictures to be processed is inputted into preset image processing model
The area-of-interest recommendation results of picture to be processed, the area-of-interest recommendation results include interested in picture to be processed
Region and the corresponding index interested of area-of-interest;
Obtain the post analysis preprocessing rule of pictures to be processed;
It is obtained according to the area-of-interest recommendation results of the post analysis preprocessing rule and each picture to be processed each
The corresponding Target Photo of picture to be processed;
The corresponding Target Photo collection of the pictures to be processed is obtained according to the corresponding Target Photo of each picture to be processed.
Preferably, further include the steps that training image handles model, the training image handles model, comprising:
Construct training sample set, the training sample set includes and positive example sample set and negative example sample set the positive example sample
This collection is by the sample set including area-of-interest annotation results, and the negative example sample set is in the camera shooting being laid in around rail
The sample set for not being marked area-of-interest is randomly selected in the video flowing of head shooting, the area-of-interest annotation results include
The index interested of area-of-interest and area-of-interest;
It is described to obtain that feature extraction is carried out to each training sample that the training sample is concentrated based on feature extractor
The corresponding sample characteristics collection of training sample set;
The sample characteristics collection is inputted into preset deep learning network model, it is defeated according to the deep learning network model
The area-of-interest recommendation results of each sample out, according to sample described in the area-of-interest recommendation results of each sample
The penalty values between the area-of-interest for being labeled out are concentrated to adjust the parameter of the deep learning network model to obtain figure
As processing model.
Preferably, described to recommend to tie according to the area-of-interest of the post analysis preprocessing rule and each picture to be processed
Fruit obtains the corresponding Target Photo of each picture to be processed, comprising:
Obtain each area-of-interest and the corresponding index interested of the area-of-interest of picture to be processed;
Count the quantity of the area-of-interest;
If the quantity is greater than preset first threshold, according to the index interested of each area-of-interest and interested
The different degree of the areal calculation area-of-interest in region;
Processing is successively carried out to area-of-interest according to different degree descending order interested until area-of-interest sum etc.
Until preset first threshold.
Preferably, the different degree of the area-of-interest can by the weighted value of index interested and area-of-interest come
Characterization.
Preferably, it is described to area-of-interest carry out processing include, comprising:
If the index interested of the area-of-interest is lower than preset third threshold value, the area-of-interest is deleted;
If the index interested of the area-of-interest is not less than preset third threshold value, the area-of-interest is obtained
Adjacent other area-of-interests, and select the highest region of index interested as target in other area-of-interests
Region;
The area-of-interest is merged into the target area.
A kind of rds data processing unit based on machine learning, described device include:
Picture module to be processed, for obtaining pictures to be processed, the picture to be processed is by being laid in around rail
It intercepts and obtains in the video flowing of camera shooting;
Area-of-interest recommendation results obtain module, for each picture to be processed in the pictures to be processed is defeated
Enter preset image processing model to obtain the area-of-interest recommendation results of the picture to be processed, the area-of-interest pushes away
Recommending result includes the area-of-interest and the corresponding index interested of area-of-interest in picture to be processed;
Post analysis preprocessing rule obtains module, for obtaining the post analysis preprocessing rule of pictures to be processed;
Processing module, for being recommended according to the area-of-interest of the post analysis preprocessing rule and each picture to be processed
As a result the corresponding Target Photo of each picture to be processed is obtained;
Target Photo collection obtains module, described to be processed for being obtained according to the corresponding Target Photo of each picture to be processed
The corresponding Target Photo collection of pictures.
It preferably, further include model training module, the model training module includes:
Training sample set construction unit, for constructing training sample set, the training sample set includes and positive example sample set
With negative example sample set, the positive example sample set is by the sample set including area-of-interest annotation results, the negative example sample set
To randomly select the sample set for not being marked area-of-interest, institute in the video flowing for the camera shooting being laid in around rail
State the index interested that area-of-interest annotation results include area-of-interest and area-of-interest;
Sample characteristics collection acquiring unit, each training sample for being concentrated based on feature extractor to the training sample
Feature extraction is carried out to obtain the corresponding sample characteristics collection of the training sample set;
Image processing model generation unit, for the sample characteristics collection to be inputted preset deep learning network model,
According to the area-of-interest recommendation results of each sample of deep learning network model output, according to each sample
Penalty values between the area-of-interest for being labeled out in sample set described in area-of-interest recommendation results adjust the depth
The parameter of learning network model is to obtain image processing model.
Preferably, the processing module, comprising:
Index acquiring unit interested, for obtaining each area-of-interest and the area-of-interest of picture to be processed
Corresponding index interested;
Quantity statistics unit, for counting the quantity of the area-of-interest;
Different degree computing unit, if being greater than preset first threshold for the quantity, according to each area-of-interest
Index and area-of-interest interested areal calculation area-of-interest different degree;
Area-of-interest processing unit, for successively locating according to different degree descending order interested to area-of-interest
Reason is until area-of-interest sum is equal to preset first threshold.
The beneficial effects of the present invention are:
A kind of rds data processing method and processing device based on machine learning can shoot the camera around rail
Obtained picture to be processed carries out the full-automatic identification and full automatic treatment of area-of-interest, and obtains Target Photo.The mesh
The area-of-interest in piece of marking on a map meets the requirement of post analysis rule, the Railway Environment abundant information degree that area-of-interest includes
Height, and area-of-interest fragment is less, so that it is particularly suitable for the information in analysis below around depth excavation railway,
Reach and reduced the region that rds data information loses and avoids analysis richness low, has promoted the purpose of analysis efficiency.
Detailed description of the invention
Fig. 1 is a kind of rds data processing method flow chart based on machine learning provided in this embodiment;
Fig. 2 is the region of interest provided in this embodiment according to the post analysis preprocessing rule and each picture to be processed
Domain recommendation results obtain the corresponding Target Photo flow chart of each picture to be processed;
Fig. 3 is a kind of rds data processing unit block diagram based on machine learning provided in this embodiment.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, the present invention is made into one below in conjunction with attached drawing
Step ground detailed description.
Embodiment 1:
The embodiment of the present invention discloses a kind of rds data processing method based on machine learning, as shown in Figure 1, the method
Include:
S101. pictures to be processed, the view that the picture to be processed is shot by the camera being laid in around rail are obtained
It intercepts and obtains in frequency stream.
S103. each picture to be processed in the pictures to be processed is inputted into preset image processing model to obtain
The area-of-interest recommendation results of the picture to be processed, the area-of-interest recommendation results include the sense in picture to be processed
Interest region and the corresponding index interested of area-of-interest.
Specifically, the preset image processing model can be used existing machine mould and be obtained by training.This
Inventive embodiments disclose the training method of one of machine mould, as shown in Figure 2, which comprises
S1. training sample set is constructed, the training sample set includes and positive example sample set and negative example sample set the positive example
Sample set is by the sample set including area-of-interest annotation results, and the negative example sample set is to be laid in taking the photograph around rail
The sample set for not being marked area-of-interest, the area-of-interest annotation results packet are randomly selected in the video flowing shot as head
Include the index interested of area-of-interest and area-of-interest.
S3. feature extraction is carried out to each training sample that the training sample is concentrated to obtain based on feature extractor
State the corresponding sample characteristics collection of training sample set.
Specifically, the feature extractor includes sequentially connected picture input layer, the first convolution process layer, the first pond
Layer, the second convolution process layer, the second pond layer, full articulamentum and normalization layer.Wherein, the feature extractor uses existing
Machine learning algorithm training obtains, and the embodiment of the present invention does not repeat them here.The embodiment of the present invention is, it is emphasized that in order to which Enhanced feature mentions
Effect is taken, the embodiment of the present invention improves the loss function of the feature extractor, increases to reach for one
The purpose of the difference of each characteristic pattern of the output of feature extractor described in the picture of input.
The loss function of the embodiment of the present invention can be obtained by adding selection item on primary loss functional foundations, institute
It states the loss that each characteristic pattern that selection item is used to export feature extractor generates to be adjusted, so that it is big to increase generation loss
Characteristic pattern influence power.
In a feasible embodiment, the loss function of the embodiment of the present invention can be represented asWherein α, β, M, δiThe weight for respectively indicating control selection degree, can
To be set according to actual needs;Parameter is selected, can be a lesser numerical value, such as 0.025;Second process of convolution
The number of the characteristic pattern of layer output;The characteristic pattern in i-th of channel that the second convolution process layer generates is being completely used for described in training
Average activity in the sample of feature extractor.Wherein LlostConcrete form for the loss function of script, loss function can
To refer to the prior art, the embodiment of the present invention is not repeated them here, for example, it can beWherein sum,
Out(s1),T(s1) it is total sample number, the output label corresponding with some sample that some sample generates, wherein for determining
Label is given value for training set and machine mould.
Further, the characteristic pattern that the embodiment of the present invention provides i-th of channel is being completely used for training the feature extraction
The calculation method of average activity in the sample of device, which comprises
S10. each sample is obtained in whole feature set of graphs of each channel output of the second convolution process layer.
S30. the feature atlas in i-th of channel is obtained according to whole feature set of graphs, the feature atlas is described
Source is the set of the characteristic pattern composition in i-th of channel of the second convolution process layer in All Eigenvalues set.
S50. the pixel of each characteristic pattern is concentrated to be integrated respectively along X direction and y direction to obtain the characteristic pattern
To activity.
S70. it is being completely used for training the spy divided by the characteristic pattern that total sample number obtains i-th of channel by the activity
Levy the average activity in the sample of extractor.
S5. the sample characteristics collection is inputted into preset deep learning network model, according to the deep learning network mould
The area-of-interest recommendation results of each sample of type output, according to the area-of-interest recommendation results of each sample
The penalty values between area-of-interest for being labeled out in sample set adjust the parameter of the deep learning network model to obtain
To image processing model.
Specifically, the sense including area-of-interest and the area-of-interest in the area-of-interest recommendation results is emerging
Interesting index.
S105. the post analysis preprocessing rule of pictures to be processed is obtained.
S107. it is obtained according to the area-of-interest recommendation results of the post analysis preprocessing rule and each picture to be processed
The corresponding Target Photo of each picture to be processed.
S109. the corresponding target figure of the pictures to be processed is obtained according to the corresponding Target Photo of each picture to be processed
Piece collection.
Specifically, during analyzing Target Photo collection in the later period, the Target Photo of Target Photo concentration is needed
Meeting post analysis rule can make analysis process smoothly execute, for example, the number of the area-of-interest in each Target Photo
Amount cannot be less than preset second threshold no more than preset first threshold, the area of each area-of-interest, and each sense is emerging
The index interested in interesting region cannot be below preset third threshold value.
Correspondingly, for handling the area-of-interest recommendation results of each picture to be processed, so that it is corresponded to
Target Photo meet post analysis rule processing rule be the embodiment of the present invention in post analysis preprocessing rule.
In a feasible embodiment, the sense according to the post analysis preprocessing rule and each picture to be processed
Interest region recommendation results obtain the corresponding Target Photo of each picture to be processed, as shown in Figure 2, comprising:
S1071. each area-of-interest and the corresponding index interested of the area-of-interest of picture to be processed are obtained.
S1072. the quantity of the area-of-interest is counted.
If S1073. the quantity is greater than preset first threshold, according to the index interested of each area-of-interest and
The different degree of the areal calculation area-of-interest of area-of-interest.
Specifically, the different degree of the area-of-interest can by the weighted value of index interested and area-of-interest come
Characterization.
S1074. processing is successively carried out to area-of-interest until area-of-interest according to different degree descending order interested
Until sum is equal to preset first threshold.
Specifically, carrying out processing to area-of-interest includes:
If S10741. the index interested of the area-of-interest is lower than preset third threshold value, it is emerging to delete the sense
Interesting region.
If S10743. the index interested of the area-of-interest is not less than preset third threshold value, the sense is obtained
The adjacent other area-of-interests in interest region, and the highest region of index interested is selected in other area-of-interests
As target area.
S10745. the area-of-interest is merged into the target area.
The rds data processing method based on machine learning that the embodiment of the invention discloses a kind of, so as to for rail
The picture to be processed that the camera of surrounding is shot carries out the full-automatic identification and full automatic treatment of area-of-interest, and obtains
Target Photo.Area-of-interest in the Target Photo meets the requirement of post analysis rule, the iron that area-of-interest includes
Road environmental information richness is high, and area-of-interest fragment is less, digs to be particularly suitable for the depth in analysis below
The information around railway is dug, has reached and has reduced the region that rds data information loses and avoids analysis richness low, promotion point
Analyse the purpose of efficiency.
The rds data processing unit based on machine learning that the present invention also provides a kind of, as shown in figure 3, described device packet
It includes:
Picture module 201 to be processed, for obtaining pictures to be processed, the picture to be processed is by being laid in around rail
Camera shooting video flowing in intercept and obtain;
Area-of-interest recommendation results obtain module 203, for by each figure to be processed in the pictures to be processed
Piece inputs preset image processing model to obtain the area-of-interest recommendation results of the picture to be processed, the region of interest
Domain recommendation results include area-of-interest and the corresponding index interested of area-of-interest in picture to be processed;
Post analysis preprocessing rule obtains module 205, for obtaining the post analysis preprocessing rule of pictures to be processed;
Processing module 207, for the area-of-interest according to the post analysis preprocessing rule and each picture to be processed
Recommendation results obtain the corresponding Target Photo of each picture to be processed;
Target Photo collection obtains module 209, for according to the corresponding Target Photo of each picture to be processed obtain it is described to
Handle the corresponding Target Photo collection of pictures.
It specifically, further include model training module, the model training module includes:
Training sample set construction unit, for constructing training sample set, the training sample set includes and positive example sample set
With negative example sample set, the positive example sample set is by the sample set including area-of-interest annotation results, the negative example sample set
To randomly select the sample set for not being marked area-of-interest, institute in the video flowing for the camera shooting being laid in around rail
State the index interested that area-of-interest annotation results include area-of-interest and area-of-interest;
Sample characteristics collection acquiring unit, each training sample for being concentrated based on feature extractor to the training sample
Feature extraction is carried out to obtain the corresponding sample characteristics collection of the training sample set;
Image processing model generation unit, for the sample characteristics collection to be inputted preset deep learning network model,
According to the area-of-interest recommendation results of each sample of deep learning network model output, according to each sample
Penalty values between the area-of-interest for being labeled out in sample set described in area-of-interest recommendation results adjust the depth
The parameter of learning network model is to obtain image processing model.
Specifically, the processing module, comprising:
Index acquiring unit interested, for obtaining each area-of-interest and the area-of-interest of picture to be processed
Corresponding index interested;
Quantity statistics unit, for counting the quantity of the area-of-interest;
Different degree computing unit, if being greater than preset first threshold for the quantity, according to each area-of-interest
Index and area-of-interest interested areal calculation area-of-interest different degree;
Area-of-interest processing unit, for successively locating according to different degree descending order interested to area-of-interest
Reason is until area-of-interest sum is equal to preset first threshold.
A kind of rds data processing unit based on machine learning disclosed by the embodiments of the present invention is based on embodiment of the method
Identical inventive concept.
The above disclosure is only the preferred embodiments of the present invention, cannot limit the right model of the present invention with this certainly
It encloses, therefore equivalent changes made in accordance with the claims of the present invention, is still within the scope of the present invention.
Claims (8)
1. a kind of rds data processing method based on machine learning, which is characterized in that the described method includes:
Pictures to be processed are obtained, are intercepted in the video flowing that the picture to be processed is shot by the camera being laid in around rail
It obtains;
It is described wait locate to obtain that each picture to be processed in the pictures to be processed is inputted into preset image processing model
The area-of-interest recommendation results of picture are managed, the area-of-interest recommendation results include the area-of-interest in picture to be processed
And the corresponding index interested of area-of-interest;
Obtain the post analysis preprocessing rule of pictures to be processed;
It is obtained according to the area-of-interest recommendation results of the post analysis preprocessing rule and each picture to be processed each wait locate
Manage the corresponding Target Photo of picture;
The corresponding Target Photo collection of the pictures to be processed is obtained according to the corresponding Target Photo of each picture to be processed.
2. the method according to claim 1, wherein further including the steps that training image handles model, the instruction
Practice image processing model, comprising:
Construct training sample set, the training sample set includes and positive example sample set and negative example sample set the positive example sample set
For by the sample set including area-of-interest annotation results, the negative example sample set is to clap in the camera being laid in around rail
The sample set for not being marked area-of-interest is randomly selected in the video flowing taken the photograph, the area-of-interest annotation results include feeling emerging
The index interested of interesting region and area-of-interest;
Feature extraction is carried out to obtain the training to each training sample that the training sample is concentrated based on feature extractor
The corresponding sample characteristics collection of sample set;
The sample characteristics collection is inputted into preset deep learning network model, according to deep learning network model output
The area-of-interest recommendation results of each sample, according in sample set described in the area-of-interest recommendation results of each sample
The penalty values between area-of-interest for being labeled out adjust the parameter of the deep learning network model to obtain at image
Manage model.
3. according to the method described in claim 2, it is characterized in that, it is described according to the post analysis preprocessing rule and it is each to
The area-of-interest recommendation results of processing picture obtain the corresponding Target Photo of each picture to be processed, comprising:
Obtain each area-of-interest and the corresponding index interested of the area-of-interest of picture to be processed;
Count the quantity of the area-of-interest;
If the quantity is greater than preset first threshold, according to the index and area-of-interest interested of each area-of-interest
Areal calculation area-of-interest different degree;
Processing is successively carried out to area-of-interest according to different degree descending order interested until area-of-interest sum is equal in advance
If first threshold until.
4. according to the method described in claim 3, it is characterized by:
The different degree of the area-of-interest can be characterized by the weighted value of index interested and area-of-interest.
5. according to the method described in claim 3, it is characterized in that, it is described to area-of-interest carry out processing include, comprising:
If the index interested of the area-of-interest is lower than preset third threshold value, the area-of-interest is deleted;
If the index interested of the area-of-interest is not less than preset third threshold value, it is adjacent to obtain the area-of-interest
Other area-of-interests, and select the highest region of index interested as target area in other area-of-interests
Domain;
The area-of-interest is merged into the target area.
6. a kind of rds data processing unit based on machine learning, which is characterized in that described device includes:
Picture module to be processed, for obtaining pictures to be processed, the picture to be processed is by the camera shooting that is laid in around rail
It intercepts and obtains in the video flowing of head shooting;
Area-of-interest recommendation results obtain module, pre- for inputting each picture to be processed in the pictures to be processed
If image processing model to obtain the area-of-interest recommendation results of the picture to be processed, the area-of-interest recommends knot
Fruit includes area-of-interest and the corresponding index interested of area-of-interest in picture to be processed;
Post analysis preprocessing rule obtains module, for obtaining the post analysis preprocessing rule of pictures to be processed;
Processing module, for the area-of-interest recommendation results according to the post analysis preprocessing rule and each picture to be processed
Obtain the corresponding Target Photo of each picture to be processed;
Target Photo collection obtains module, for obtaining the picture to be processed according to the corresponding Target Photo of each picture to be processed
Collect corresponding Target Photo collection.
7. device according to claim 6, which is characterized in that it further include model training module, the model training module
Include:
Training sample set construction unit, for constructing training sample set, the training sample set includes and positive example sample set and negative
Example sample set, the positive example sample set be by the sample set including area-of-interest annotation results, the negative example sample set for
It is laid in the video flowing of the camera shooting around rail and randomly selects the sample set for not being marked area-of-interest, the sense
Interest region annotation results include the index interested of area-of-interest and area-of-interest;
Sample characteristics collection acquiring unit, for being carried out based on feature extractor to each training sample that the training sample is concentrated
Feature extraction is to obtain the corresponding sample characteristics collection of the training sample set;
Image processing model generation unit, for the sample characteristics collection to be inputted preset deep learning network model, according to
The area-of-interest recommendation results of each sample of the deep learning network model output, the sense according to each sample are emerging
Penalty values between the area-of-interest for being labeled out in sample set described in interesting region recommendation results adjust the deep learning
The parameter of network model is to obtain image processing model.
8. device according to claim 7, which is characterized in that the processing module, comprising:
Index acquiring unit interested, each area-of-interest and the area-of-interest for obtaining picture to be processed are corresponding
Index interested;
Quantity statistics unit, for counting the quantity of the area-of-interest;
Different degree computing unit, if being greater than preset first threshold for the quantity, according to the sense of each area-of-interest
The different degree of the areal calculation area-of-interest of interest index and area-of-interest;
Area-of-interest processing unit is straight for successively carrying out handling to area-of-interest according to different degree descending order interested
Until area-of-interest sum is equal to preset first threshold.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910790311.5A CN110503047A (en) | 2019-08-26 | 2019-08-26 | A kind of rds data processing method and processing device based on machine learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910790311.5A CN110503047A (en) | 2019-08-26 | 2019-08-26 | A kind of rds data processing method and processing device based on machine learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110503047A true CN110503047A (en) | 2019-11-26 |
Family
ID=68589476
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910790311.5A Pending CN110503047A (en) | 2019-08-26 | 2019-08-26 | A kind of rds data processing method and processing device based on machine learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110503047A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111079819A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Method for judging state of coupler knuckle pin of railway wagon based on image recognition and deep learning |
CN111191059A (en) * | 2019-12-31 | 2020-05-22 | 腾讯科技(深圳)有限公司 | Image processing method, image processing device, computer storage medium and electronic equipment |
CN111648174A (en) * | 2020-06-01 | 2020-09-11 | 柳七峰 | Automatic straightening equipment for rails |
CN111797832A (en) * | 2020-07-14 | 2020-10-20 | 成都数之联科技有限公司 | Automatic generation method and system of image interesting region and image processing method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105976392A (en) * | 2016-05-30 | 2016-09-28 | 北京智芯原动科技有限公司 | Maximum-output-probability-based vehicle tyre detection method and apparatus |
CN106355188A (en) * | 2015-07-13 | 2017-01-25 | 阿里巴巴集团控股有限公司 | Image detection method and device |
CN107403183A (en) * | 2017-07-21 | 2017-11-28 | 桂林电子科技大学 | The intelligent scissor method that conformity goal is detected and image segmentation is integrated |
US20180314943A1 (en) * | 2017-04-27 | 2018-11-01 | Jianming Liang | Systems, methods, and/or media, for selecting candidates for annotation for use in training a classifier |
US20180322635A1 (en) * | 2017-05-05 | 2018-11-08 | Shanghai United Imaging Healthcare Co., Ltd. | Systems and methods for image processing |
CN108875730A (en) * | 2017-05-16 | 2018-11-23 | 中兴通讯股份有限公司 | A kind of deep learning sample collection method, apparatus, equipment and storage medium |
-
2019
- 2019-08-26 CN CN201910790311.5A patent/CN110503047A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106355188A (en) * | 2015-07-13 | 2017-01-25 | 阿里巴巴集团控股有限公司 | Image detection method and device |
CN105976392A (en) * | 2016-05-30 | 2016-09-28 | 北京智芯原动科技有限公司 | Maximum-output-probability-based vehicle tyre detection method and apparatus |
US20180314943A1 (en) * | 2017-04-27 | 2018-11-01 | Jianming Liang | Systems, methods, and/or media, for selecting candidates for annotation for use in training a classifier |
US20180322635A1 (en) * | 2017-05-05 | 2018-11-08 | Shanghai United Imaging Healthcare Co., Ltd. | Systems and methods for image processing |
CN108875730A (en) * | 2017-05-16 | 2018-11-23 | 中兴通讯股份有限公司 | A kind of deep learning sample collection method, apparatus, equipment and storage medium |
CN107403183A (en) * | 2017-07-21 | 2017-11-28 | 桂林电子科技大学 | The intelligent scissor method that conformity goal is detected and image segmentation is integrated |
Non-Patent Citations (2)
Title |
---|
SUNMOK KIM.ET AL: ""An Effective Sign Language Learning with Object Detection Based ROI Segmentation"", 《IEEE》 * |
叶银兰: ""基于感兴趣区域和神经网络的图像检索"", 《河南师范大学学报(自然科学版)》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111079819A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Method for judging state of coupler knuckle pin of railway wagon based on image recognition and deep learning |
CN111079819B (en) * | 2019-12-12 | 2021-03-23 | 哈尔滨市科佳通用机电股份有限公司 | Method for judging state of coupler knuckle pin of railway wagon based on image recognition and deep learning |
CN111191059A (en) * | 2019-12-31 | 2020-05-22 | 腾讯科技(深圳)有限公司 | Image processing method, image processing device, computer storage medium and electronic equipment |
CN111191059B (en) * | 2019-12-31 | 2023-05-05 | 腾讯科技(深圳)有限公司 | Image processing method, device, computer storage medium and electronic equipment |
CN111648174A (en) * | 2020-06-01 | 2020-09-11 | 柳七峰 | Automatic straightening equipment for rails |
CN111797832A (en) * | 2020-07-14 | 2020-10-20 | 成都数之联科技有限公司 | Automatic generation method and system of image interesting region and image processing method |
CN111797832B (en) * | 2020-07-14 | 2024-02-02 | 成都数之联科技股份有限公司 | Automatic generation method and system for image region of interest and image processing method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110503047A (en) | A kind of rds data processing method and processing device based on machine learning | |
CN103207910B (en) | Image retrieval method based on hierarchical features and genetic programming relevance feedback | |
CN104933428B (en) | A kind of face identification method and device based on tensor description | |
CN106612457B (en) | video sequence alignment method and system | |
CN105808610A (en) | Internet picture filtering method and device | |
CN109102579A (en) | A kind of intelligent meeting based on recognition of face and big data analysis is registered equipment | |
CN107993250A (en) | A kind of fast multi-target pedestrian tracking and analysis method and its intelligent apparatus | |
CN114170137B (en) | Pepper disease identification method, identification system and computer readable storage medium | |
CN108090070A (en) | A kind of Chinese entity attribute abstracting method | |
CN110321956B (en) | Grass pest control method and device based on artificial intelligence | |
CN110750567A (en) | Wearing and matching data generation method and system based on intelligent wardrobe | |
CN108229588A (en) | A kind of machine learning recognition methods based on deep learning | |
CN108334878A (en) | Video images detection method and apparatus | |
CN106650795A (en) | Sorting method of hotel room type images | |
CN114359727A (en) | Tea disease identification method and system based on lightweight optimization Yolo v4 | |
CN109741301A (en) | A kind of intensive object count method based on deep learning faster | |
CN110110663A (en) | A kind of age recognition methods and system based on face character | |
CN111191531A (en) | Rapid pedestrian detection method and system | |
CN109271848A (en) | A kind of method for detecting human face and human face detection device, storage medium | |
CN110689039A (en) | Trunk texture identification method based on four-channel convolutional neural network | |
CN110516734A (en) | A kind of image matching method, device, equipment and storage medium | |
Sakurai et al. | Two-step Transfer Learning for Semantic Plant Segmentation. | |
CN107516102A (en) | View data is classified and establishes disaggregated model method, apparatus and system | |
Zhou et al. | Vine spread for superpixel segmentation | |
CN104680193A (en) | Online target classification method and system based on fast similarity network fusion algorithm |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191126 |