CN110222680A - A kind of domestic waste article outer packing Method for text detection - Google Patents
A kind of domestic waste article outer packing Method for text detection Download PDFInfo
- Publication number
- CN110222680A CN110222680A CN201910416098.1A CN201910416098A CN110222680A CN 110222680 A CN110222680 A CN 110222680A CN 201910416098 A CN201910416098 A CN 201910416098A CN 110222680 A CN110222680 A CN 110222680A
- Authority
- CN
- China
- Prior art keywords
- text
- image
- detected
- text filed
- filed
- 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
- 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
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
- G06V10/225—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition based on a marking or identifier characterising the area
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
A kind of domestic waste article outer packing Method for text detection: the image data set of acquisition municipal refuse article outer packing, and the text filed of each image concentrated to image data is labeled;The each image concentrated to the image data that mark is completed generates Text Score characteristic pattern and multichannel position feature figure according to text filed mark, constitutes the training label of each image;The image that image data is concentrated is divided into training set and test set according to the ratio of 9:1;It constructs full convolutional neural networks model and is trained, obtain trained full convolutional neural networks model;The prediction for obtaining image to be detected using trained full convolutional neural networks model is text filed;Threshold value screening stage;The non-maxima suppression stage obtains final text filed testing result.Strong robustness of the present invention, detection accuracy are high.There is more strong adaptability using complex article surface of the text detection strategy divided based on pixel for distortion, detection effect is also more preferable.
Description
Technical field
The present invention relates to a kind of text detection methods of article outer packing.More particularly to a kind of city based on pixel segmentation
House refuse article outer packing Method for text detection.
Background technique
With the rapid development of industrialization and urbanization, huge change: the consumer goods is also had occurred in the consumption life of the common people
It is many kinds of, it continues to introduce new.Cause domestic waste also while explosive increase is presented, most big and medium-sized cities occur tight
The garbage-surrounded city problem of weight, more than 600 of China big and medium-sized cities 2/3rds are fallen among rubbish encirclement, and the city of a quarter is
Place is stacked through unsuitable garbage loading embeading.Worse, domestic waste is still persistently increased with the speed of 8%-10%
It is long, and the municipal refuse amount of clearing speedup is only 3.3%, municipal refuse accumulates storage over the years and is up to more than 80 hundred million tons.The city of flood tide
Sanitary fill causes high risks to surrounding city natural ecological environment and residents ' health.
" garbage-surrounded city " is the common difficulty that world community faces, and not merely appears in China, and other countries, the world are also same
Sample is worried for garbage problem.It is put into effect to cope with " garbage-surrounded city " western developed country and area for the national conditions of this country
Various strategies simultaneously achieve many beneficial effects.Although the garbage disposal strategy of these countries has a certain difference, core
Theory is surprising similar, i.e., cracks " garbage-surrounded city " quagmire by garbage classification, realizes garbage as resource, minimizing, harmless
Change processing.
But consider the special national conditions in China: its people, which put on by classification, to be realized not strong and can not be obviously improved in a short time
(from Japan with American-European through examining, the change of consciousness needs 20 years or so time), house refuse predominantly mixing is launched.For
Realization waste resources maximally utilize, and there is an urgent need to develop, miniaturization, economical, detection accuracy is high, fireballing based on intelligence
The waste items essence sorting system of energy vision-based detection.
High-precision vision classification is the key link of waste items essence sorting system development.About waste items image point
Class just has some related ends to report, but be all still in infancy external recent years.Main achievement has: 2016,
Sakr etc. proposes the rubbish image classification algorithms based on the AlexNet network architecture, and in the 2000 rubbish figures voluntarily acquired
As being tested on data set (undisclosed, to contain plastics, paper, 3 class article of metal), accuracy rate 83%.2017, Stamford was big
The it is proposeds such as Yang are learned to divide 6 classes (glass, paper, hardboard, plastics, metal, Other Waste) waste items with classics SVM
Class processing, test data set total sample number are 2527 (publication)s, recognition accuracy 63%.Then, 2018
It etc. proposing RecycleNet algorithm, and is tested on Stanford University's public data collection, accuracy rate is increased to
81%.
Although above method can predict the classification of waste items, high-purity rubbish may be not achieved only with the technology
The testing requirements of recycling (recognition accuracy is greater than 95%).Reason is: waste items are different from commodity for sale, often turn round
The complicated states such as curved change, tearing, therefore the simple graphical information using article carries out classification to article material and is difficult to be expired
The effect of meaning.We have discovered that, other than using the graphical information of article, the text information in article outer packing is (such as in the recent period
Brand, common advertising slogan, product operation instruction etc.) very valuable clue can also be provided for the Accurate Prediction of goods categories.
Therefore, it is highly desirable to carry out waste items outer packing text detection Study of recognition.
The Method for text detection of early stage is based primarily upon the text in the Image Feature Detection scene of more bottom, existing inspection
Survey method can substantially be divided into three classes: edge is based on, based on region and based on the detection method of texture.Widely apply result table
Bright, these traditional detection methods are only applicable to that imaging clearly, clean background, font be simple, file and picture of arranged regular.And object
Text on article wrapper belongs to scene text (Scene text).Document text is compared in the font of scene text, color, arrangement
It is much more complex, cause effect of traditional text detection techniques in scene text detection very unsatisfactory.In recent years, deep learning
Technology is also initially introduced into scene text detection.Method for text detection based on deep learning can be divided into two classes: be based on
The Method for text detection of anchor frame and the Method for text detection divided based on pixel.The usual precision of Method for text detection based on anchor frame
It is higher but detection speed it is slower, and based on pixel segmentation method precision be slightly below the method based on anchor frame but detection speed compared with
Fastly.Waste items outer packing text is a kind of special scene text, has no that the method based on deep learning is applied to rubbish at present
The relevant report of rubbish article outer packing text detection.Since sorting amount is big in rubbish assembly line Nonexpendable Item sort process,
Therefore very high requirement is proposed to the real-time of detection.
It is simple not high using article graphical information progress waste items material nicety of grading, it is difficult to reach high-purity rubbish and return
The testing requirements of receipts, and the text information on article can provide new valuable clue for garbage classification precision.
Summary of the invention
The technical problem to be solved by the invention is to provide a kind of domestic waste articles that can be improved detection accuracy
Outer packing Method for text detection.
The technical scheme adopted by the invention is that: a kind of domestic waste article outer packing Method for text detection, including
Following steps:
1) image data set of municipal refuse article outer packing, and the text for each image concentrated to image data are acquired
Region is labeled;
2) Text Score spy is generated according to text filed mark to each image that the image data that mark is completed is concentrated
Sign figure and multichannel position feature figure, constitute the training label of each image;
3) image that image data is concentrated is divided into training set and test set according to the ratio of 9:1;
4) it constructs full convolutional neural networks model and is trained using the training set for indicating trained label, trained
Full convolutional neural networks model;
5) prediction for obtaining image to be detected using trained full convolutional neural networks model is text filed;
6) threshold value screening stage, it is text filed to the prediction of image to be detected according to the Text Score confidence threshold value of setting
Effectively text filed screening is carried out, the effective text filed of image to be detected is obtained;
7) the non-maxima suppression stage is effective text filed non-maximum suppression of progress to obtained image to be detected
System operation obtains final text filed testing result to remove partial redundance or in vain text filed in image to be detected.
Mark described in step 1), which is the method that manually marks, uses four for text filed on image each in data set
Side shape collimation mark, which outpours, to be come, that is, marks four text filed apex coordinate C={ (xi,yi), 1≤i≤4 are opened from the image upper left corner
Begin and is arranged according to clockwise direction.
Training label described in step 2) is to generate to mark text filed text/non-textual classification, four apex coordinates
Offset and text filed rotation angle.
Full convolutional neural networks model described in step 4) includes that characteristic extracting module, Fusion Features module and text are pre-
Module is surveyed, characteristic extracting module extracts image using ResNet-50 network for extracting the characteristic pattern of input picture step by step
High-level characteristic figure and low-level feature figure;Fusion Features module is for melting the characteristic pattern that characteristic extracting module is extracted step by step
It closes, i.e., merges the high-level characteristic figure that characteristic extracting module extracts with low-level feature figure;Text prediction module for pair
The characteristic pattern obtained after the operation of Fusion Features module carries out Text Score prediction and text position regression forecasting.
Loss function uses more Classification Loss functions, including Classification Loss and recurrence loss when training in step 4).
Step 5) includes being detected using trained full convolutional neural networks model to image to be detected, obtain to
The Text Score characteristic pattern and multichannel position feature figure of the single channel Pixel-level of detection image, according to image to be detected text point
All predictions that number characteristic pattern and multichannel position feature figure obtain image to be detected are text filed.
The gray value of each pixel in the Text Score characteristic pattern is the pre- of image to be detected where the pixel
Survey text filed Text Score;The gray value of each pixel in the multichannel position feature figure is the pixel away from this
Image to be detected is pre- where the distance on text filed four vertex of prediction of image to be detected where pixel or the pixel
Survey text filed rotation angle.
Step 6) be by the text filed Text Score of each prediction of image to be detected respectively with the text confidence of setting
Degree threshold value is compared, wherein greater than the text filed effective text area for image to be detected of prediction of text confidence threshold value
Domain.
A kind of domestic waste article outer packing Method for text detection of the invention, strong robustness, detection accuracy are high.By
Often there are the complicated states such as torsional deformation, tearing in the outer packing of house refuse article, therefore using the text divided based on pixel
Inspection policies have more strong adaptability, detection effect compared to complex article surface of the detection method based on anchor frame for distortion
More preferably.
The text of article outer packing can be efficiently extracted using method of the invention, this is provided for the classification of waste items material
It is raw can be obviously improved city for very valuable clue, article material sorting technique of the subsequent combination based on article graphical information
Waste items living recycle nicety of grading, have extensive market application potential.
Detailed description of the invention
Fig. 1 is a kind of flow chart of domestic waste article outer packing Method for text detection of the present invention;
Fig. 2 is the schematic network structure of full convolutional neural networks model in the present invention.
Specific embodiment
Below with reference to embodiment and attached drawing to a kind of domestic waste article outer packing Method for text detection of the invention
It is described in detail.
Just technical term of the invention is explained and illustrated first below:
Full convolutional network: network model is made of convolutional layer and nonlinear transformation layer entirely, is different from convolutional neural networks, entirely
The end of convolution eliminates full articulamentum, therefore does not have size limitation to input picture, can be realized pixel scale end to end
Prediction.
ResNet-50:ResNet-50 generally comprises 5 parts, wherein first part is by the volume using 7*7 convolution kernel
Lamination composition, then by the pond layer that convolution kernel is 3*3, step-length is 2, rear four parts are each by convolution kernel in varying numbers
For the convolutional layer of 3*3 and a pond layer composition.ResNet-50 has powerful character representation ability, often in different computers
It is used as basic network in visual task.
As shown in Figure 1, a kind of domestic waste article outer packing Method for text detection of the invention, including walk as follows
It is rapid:
1) image data set of municipal refuse article outer packing, and the text for each image concentrated to image data are acquired
Region is labeled;
The mark, which is the method that manually marks, uses quadrangle for text filed on text image each in data set
Collimation mark, which outpours, to be come, that is, marks four text filed apex coordinate C={ (xi,yi), 1≤i≤4, since the image upper left corner simultaneously
It is arranged according to clockwise direction.
2) Text Score spy is generated according to text filed mark to each image that the image data that mark is completed is concentrated
Sign figure and multichannel position feature figure, constitute the training label of each image;
The training label is to generate to mark text filed text/non-textual classification, four apex coordinate offsets
With text filed rotation angle.
3) image that image data is concentrated is divided into training set and test set according to the ratio of 9:1;
4) it constructs full convolutional neural networks model and is trained using the training set for indicating trained label, trained
Full convolutional neural networks model, training when loss function use more Classification Loss functions, including Classification Loss and return loss.
As shown in Fig. 2, the full convolutional neural networks model includes characteristic extracting module, Fusion Features module and text
This prediction module, characteristic extracting module use ResNet- as shown in Table 1 for extracting the characteristic pattern of input picture step by step
The high-level characteristic figure and low-level feature figure of 50 networks extraction image;Fusion Features module for extracting characteristic extracting module step by step
Characteristic pattern merged, i.e., the high-level characteristic figure that characteristic extracting module extracts is merged with low-level feature figure;Text
Prediction module is used to carry out Text Score prediction to the characteristic pattern obtained after the operation of Fusion Features module and text position returns in advance
It surveys.
1 ResNet-50 network configuration of table
5) prediction for obtaining image to be detected using trained full convolutional neural networks model is text filed;Including,
Image to be detected is detected using trained full convolutional neural networks model, obtains the list of image to be detected
The Text Score characteristic pattern and multichannel position feature figure of channel Pixel-level, according to image to be detected Text Score characteristic pattern and more
All predictions that channel position characteristic pattern obtains image to be detected are text filed.
The gray value of each pixel in the Text Score characteristic pattern is the pre- of image to be detected where the pixel
Survey text filed Text Score;The gray value of each pixel in the multichannel position feature figure is the pixel away from this
Image to be detected is pre- where the distance on text filed four vertex of prediction of image to be detected where pixel or the pixel
Survey text filed rotation angle.
6) threshold value screening stage, it is text filed to the prediction of image to be detected according to the Text Score confidence threshold value of setting
Effectively text filed screening is carried out, the effective text filed of image to be detected is obtained;
Be by the text filed Text Score of each prediction of image to be detected respectively with the text confidence threshold value of setting
It is compared, wherein text filed greater than the prediction of text confidence threshold value for the effective text filed of image to be detected.
7) the non-maxima suppression stage is effective text filed non-maximum suppression of progress to obtained image to be detected
System operation obtains final text filed testing result to remove partial redundance or in vain text filed in image to be detected.
Specific example is given below:
1, data set employed in present example is 4000 domestic waste images of items of acquisition, every figure
Contain a kind of waste items as in.The text filed of each image concentrated to image data is labeled.Specifically, manually
The method of mark marks out text filed on text image each in data set with quadrilateral frame to mark text filed
Four apex coordinate C={ (xi,yi), 1≤i≤4 are arranged since the image upper left corner and according to clockwise direction.
2, Text Score spy is generated according to text filed mark to each image that the image data that mark is completed is concentrated
Sign figure and multichannel position feature figure, constitute the training label of each image;
The training label is to generate to mark text filed text/non-textual classification, four apex coordinate offsets
With text filed rotation angle.
3, data set is divided into training set and test set according to the ratio of 9:1.
4, it constructs full convolutional neural networks model and is trained using the training set for indicating trained label, wherein losing letter
Number uses more Classification Loss functions, including Classification Loss and recurrence loss.Classification Loss intersects entropy loss using two classification, returns
Loss is lost using IoU.Optimizer selects Adam, and initial learning rate is 0.001, and every 27300 step of training falls to original ten
/ mono-, until model restrains deconditioning.Specifically, which includes three modules: feature extraction
Module, Fusion Features module, text prediction module.Characteristic extracting module uses ResNet-50 basic network architectures, and configuration is such as
Shown in table 1, including 5 parts, respectively conv1, conv2_x, conv3_x, conv4_x, conv5_x, for extracting step by step
The high-level semantics features of image, each part includes convolutional layer in varying numbers and a pond layer, for the image of input, warp
After crossing each part processing of ResNet-50, the resolution ratio of output characteristic pattern reduces 2 times.Fusion Features module is used for will be special
The characteristic pattern of 4 parts (conv2_x, conv3_x, conv4_x, conv5_x) output is up-sampled after in sign extraction module
Deng operation after merged.Text prediction module includes classification branch and returns branch, is respectively used to category score prediction and sits
Offset regression forecasting is marked, is operated wherein classification branch carries out the convolution kernel that convolution kernel is 1*1 to the characteristic pattern of input, obtains 1
A characteristic pattern indicates the text filed category score of the candidate of prediction, returns branch and is to the characteristic pattern progress convolution kernel of input
The convolution operation of 1*1 obtains 5 characteristic patterns, respectively indicates the candidate of prediction text filed coordinate shift amount and rotation angle.
5, the prediction for obtaining image to be detected using trained full convolutional neural networks model is text filed.Specifically,
Image to be detected is detected using trained full convolutional neural networks model, obtains the Text Score of single channel Pixel-level
The gray value of characteristic pattern and multichannel position feature figure, each pixel in the Text Score characteristic pattern is the pixel institute
In the text filed Text Score of the prediction of image to be detected;The gray scale of each pixel in the multichannel position feature figure
Value is distance or the pixel institute on prediction text filed four vertex of the pixel away from image to be detected where the pixel
In the text filed rotation angle of the prediction of image to be detected.
6, threshold value screening stage, the text confidence threshold value that we set herein is 0.8.I.e. by each of image to be detected
Predict that text filed Text Score is compared with 0.8 respectively, it is to have that wherein Text Score is text filed greater than 0.8 prediction
It imitates text filed.
7, the non-maxima suppression stage, by obtained effectively text filed the progresss non-maxima suppression of previous step operate with
Partial redundance or in vain text filed are removed, final text filed testing result is obtained.
Claims (8)
1. a kind of domestic waste article outer packing Method for text detection, which comprises the steps of:
1) image data set of municipal refuse article outer packing is acquired, and to the text filed of each image of image data concentration
It is labeled;
2) each image concentrated to the image data that mark is completed generates Text Score characteristic pattern according to text filed mark
With multichannel position feature figure, the training label of each image is constituted;
3) image that image data is concentrated is divided into training set and test set according to the ratio of 9:1;
4) it constructs full convolutional neural networks model and is trained using the training set for indicating trained label, obtained trained complete
Convolutional neural networks model;
5) prediction for obtaining image to be detected using trained full convolutional neural networks model is text filed;
6) threshold value screening stage, according to the progress text filed to the prediction of image to be detected of the Text Score confidence threshold value of setting
Effective text filed screening, obtains the effective text filed of image to be detected;
7) the non-maxima suppression stage is effective text filed progress non-maxima suppression behaviour to obtained image to be detected
Make, to remove partial redundance or in vain text filed in image to be detected, obtains final text filed testing result.
2. a kind of domestic waste article outer packing Method for text detection according to claim 1, which is characterized in that step
It is rapid 1) described in mark to be the method that manually marks mark text filed on image each in data set with quadrilateral frame
Out, that is, four text filed apex coordinate C={ (x are markedi,yi), 1≤i≤4, since the image upper left corner and according to suitable
Clockwise arrangement.
3. a kind of domestic waste article outer packing Method for text detection according to claim 1, which is characterized in that step
It is rapid 2) described in training label be to generate to mark text filed text/non-textual classification, four apex coordinate offsets and text
One's respective area rotates angle.
4. a kind of domestic waste article outer packing Method for text detection according to claim 1, which is characterized in that step
It is rapid 4) described in full convolutional neural networks model include characteristic extracting module, Fusion Features module and text prediction module, it is special
Extraction module is levied for extracting the characteristic pattern of input picture step by step, i.e., extracts the high-level characteristic of image using ResNet-50 network
Figure and low-level feature figure;Fusion Features module, i.e., will be special for being merged to the characteristic pattern that characteristic extracting module is extracted step by step
The high-level characteristic figure that sign extraction module extracts is merged with low-level feature figure;Text prediction module is used for Fusion Features mould
The characteristic pattern obtained after block operation carries out Text Score prediction and text position regression forecasting.
5. a kind of domestic waste article outer packing Method for text detection according to claim 1, which is characterized in that step
It is rapid 4) in training when loss function use more Classification Loss functions, including Classification Loss and return loss.
6. a kind of domestic waste article outer packing Method for text detection according to claim 1, which is characterized in that step
It is rapid 5) to include, image to be detected is detected using trained full convolutional neural networks model, obtains image to be detected
The Text Score characteristic pattern and multichannel position feature figure of single channel Pixel-level, according to image to be detected Text Score characteristic pattern and
All predictions that multichannel position feature figure obtains image to be detected are text filed.
7. a kind of domestic waste article outer packing Method for text detection according to claim 6, which is characterized in that institute
State each pixel in Text Score characteristic pattern gray value be image to be detected where the pixel prediction it is text filed
Text Score;The gray value of each pixel in the multichannel position feature figure is the pixel away from where the pixel
The prediction of image to be detected is text filed where the distance on text filed four vertex of prediction of image to be detected or the pixel
Rotation angle.
8. a kind of domestic waste article outer packing Method for text detection according to claim 1, which is characterized in that step
6) rapid is to carry out the text filed Text Score of each prediction of image to be detected with the text confidence threshold value of setting respectively
Compare, wherein text filed greater than the prediction of text confidence threshold value for the effective text filed of image to be detected.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910416098.1A CN110222680A (en) | 2019-05-19 | 2019-05-19 | A kind of domestic waste article outer packing Method for text detection |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910416098.1A CN110222680A (en) | 2019-05-19 | 2019-05-19 | A kind of domestic waste article outer packing Method for text detection |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110222680A true CN110222680A (en) | 2019-09-10 |
Family
ID=67821473
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910416098.1A Pending CN110222680A (en) | 2019-05-19 | 2019-05-19 | A kind of domestic waste article outer packing Method for text detection |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110222680A (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110738131A (en) * | 2019-09-20 | 2020-01-31 | 广州游艺云物联网技术有限公司 | Garbage classification management method and device based on deep learning neural network |
CN110880000A (en) * | 2019-11-27 | 2020-03-13 | 上海智臻智能网络科技股份有限公司 | Picture character positioning method and device, computer equipment and storage medium |
CN110884791A (en) * | 2019-11-28 | 2020-03-17 | 石家庄邮电职业技术学院(中国邮政集团公司培训中心) | Vision garbage classification system and classification method based on TensorFlow |
CN111444918A (en) * | 2020-04-01 | 2020-07-24 | 中移雄安信息通信科技有限公司 | Image inclined text line detection model training and image inclined text line detection method |
CN111476226A (en) * | 2020-02-29 | 2020-07-31 | 新华三大数据技术有限公司 | Text positioning method and device and model training method |
CN111598041A (en) * | 2020-05-25 | 2020-08-28 | 青岛联合创智科技有限公司 | Image generation text method for article searching |
CN111598011A (en) * | 2020-05-19 | 2020-08-28 | 科大讯飞股份有限公司 | Garbage classification processing method, related equipment and readable storage medium |
CN111767921A (en) * | 2020-06-30 | 2020-10-13 | 上海媒智科技有限公司 | Express bill positioning and correcting method and device |
CN112307962A (en) * | 2020-10-30 | 2021-02-02 | 成都福立盟环保大数据有限公司 | Method for detecting soil dirt on outer surface of carriage of construction waste transport vehicle |
CN112329765A (en) * | 2020-10-09 | 2021-02-05 | 中保车服科技服务股份有限公司 | Text detection method and device, storage medium and computer equipment |
CN112784737A (en) * | 2021-01-21 | 2021-05-11 | 上海云从汇临人工智能科技有限公司 | Text detection method, system and device combining pixel segmentation and line segment anchor |
CN112801045A (en) * | 2021-03-18 | 2021-05-14 | 北京世纪好未来教育科技有限公司 | Text region detection method, electronic equipment and computer storage medium |
CN113033593A (en) * | 2019-12-25 | 2021-06-25 | 上海智臻智能网络科技股份有限公司 | Text detection training method and device based on deep learning |
CN113705673A (en) * | 2021-08-27 | 2021-11-26 | 四川医枢科技有限责任公司 | Character detection method, device, equipment and storage medium |
CN113780038A (en) * | 2020-06-10 | 2021-12-10 | 深信服科技股份有限公司 | Picture auditing method and device, computing equipment and storage medium |
CN115935462A (en) * | 2022-10-18 | 2023-04-07 | 美的集团股份有限公司 | External package modeling method and device |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107609549A (en) * | 2017-09-20 | 2018-01-19 | 北京工业大学 | The Method for text detection of certificate image under a kind of natural scene |
CN109299274A (en) * | 2018-11-07 | 2019-02-01 | 南京大学 | A kind of natural scene Method for text detection based on full convolutional neural networks |
-
2019
- 2019-05-19 CN CN201910416098.1A patent/CN110222680A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107609549A (en) * | 2017-09-20 | 2018-01-19 | 北京工业大学 | The Method for text detection of certificate image under a kind of natural scene |
CN109299274A (en) * | 2018-11-07 | 2019-02-01 | 南京大学 | A kind of natural scene Method for text detection based on full convolutional neural networks |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110738131A (en) * | 2019-09-20 | 2020-01-31 | 广州游艺云物联网技术有限公司 | Garbage classification management method and device based on deep learning neural network |
CN110880000A (en) * | 2019-11-27 | 2020-03-13 | 上海智臻智能网络科技股份有限公司 | Picture character positioning method and device, computer equipment and storage medium |
CN110880000B (en) * | 2019-11-27 | 2022-09-02 | 上海智臻智能网络科技股份有限公司 | Picture character positioning method and device, computer equipment and storage medium |
CN110884791A (en) * | 2019-11-28 | 2020-03-17 | 石家庄邮电职业技术学院(中国邮政集团公司培训中心) | Vision garbage classification system and classification method based on TensorFlow |
CN113033593A (en) * | 2019-12-25 | 2021-06-25 | 上海智臻智能网络科技股份有限公司 | Text detection training method and device based on deep learning |
CN113033593B (en) * | 2019-12-25 | 2023-09-01 | 上海智臻智能网络科技股份有限公司 | Text detection training method and device based on deep learning |
CN111476226A (en) * | 2020-02-29 | 2020-07-31 | 新华三大数据技术有限公司 | Text positioning method and device and model training method |
CN111476226B (en) * | 2020-02-29 | 2022-08-30 | 新华三大数据技术有限公司 | Text positioning method and device and model training method |
CN111444918A (en) * | 2020-04-01 | 2020-07-24 | 中移雄安信息通信科技有限公司 | Image inclined text line detection model training and image inclined text line detection method |
CN111598011A (en) * | 2020-05-19 | 2020-08-28 | 科大讯飞股份有限公司 | Garbage classification processing method, related equipment and readable storage medium |
CN111598041A (en) * | 2020-05-25 | 2020-08-28 | 青岛联合创智科技有限公司 | Image generation text method for article searching |
CN113780038A (en) * | 2020-06-10 | 2021-12-10 | 深信服科技股份有限公司 | Picture auditing method and device, computing equipment and storage medium |
CN111767921A (en) * | 2020-06-30 | 2020-10-13 | 上海媒智科技有限公司 | Express bill positioning and correcting method and device |
CN112329765B (en) * | 2020-10-09 | 2024-05-24 | 中保车服科技服务股份有限公司 | Text detection method and device, storage medium and computer equipment |
CN112329765A (en) * | 2020-10-09 | 2021-02-05 | 中保车服科技服务股份有限公司 | Text detection method and device, storage medium and computer equipment |
CN112307962A (en) * | 2020-10-30 | 2021-02-02 | 成都福立盟环保大数据有限公司 | Method for detecting soil dirt on outer surface of carriage of construction waste transport vehicle |
CN112784737A (en) * | 2021-01-21 | 2021-05-11 | 上海云从汇临人工智能科技有限公司 | Text detection method, system and device combining pixel segmentation and line segment anchor |
CN112784737B (en) * | 2021-01-21 | 2023-10-20 | 上海云从汇临人工智能科技有限公司 | Text detection method, system and device combining pixel segmentation and line segment anchor |
CN112801045B (en) * | 2021-03-18 | 2021-07-16 | 北京世纪好未来教育科技有限公司 | Text region detection method, electronic equipment and computer storage medium |
CN112801045A (en) * | 2021-03-18 | 2021-05-14 | 北京世纪好未来教育科技有限公司 | Text region detection method, electronic equipment and computer storage medium |
CN113705673A (en) * | 2021-08-27 | 2021-11-26 | 四川医枢科技有限责任公司 | Character detection method, device, equipment and storage medium |
CN113705673B (en) * | 2021-08-27 | 2023-12-12 | 四川医枢科技有限责任公司 | Text detection method, text detection device, text detection equipment and storage medium |
CN115935462A (en) * | 2022-10-18 | 2023-04-07 | 美的集团股份有限公司 | External package modeling method and device |
CN115935462B (en) * | 2022-10-18 | 2023-12-26 | 美的集团股份有限公司 | Method and device for modeling outer package |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110222680A (en) | A kind of domestic waste article outer packing Method for text detection | |
Wang | Garbage recognition and classification system based on convolutional neural network vgg16 | |
Liang et al. | Traffic sign detection via improved sparse R‐CNN for autonomous vehicles | |
CN110210635A (en) | A kind of intelligent classification recovery system that can identify waste | |
CN109859190A (en) | A kind of target area detection method based on deep learning | |
CN107945153A (en) | A kind of road surface crack detection method based on deep learning | |
Mohammed et al. | Automated waste-sorting and recycling classification using artificial neural network and features fusion: A digital-enabled circular economy vision for smart cities | |
CN112528997A (en) | Tibetan-Chinese bilingual scene text detection method based on text center region amplification | |
CN116542932A (en) | Injection molding surface defect detection method based on improved YOLOv5s | |
Gan et al. | Research on the algorithm of urban waste classification and recycling based on deep learning technology | |
AR et al. | Automatic waste detection by deep learning and disposal system design | |
CN115330720A (en) | Unmanned aerial vehicle image earth surface solid waste detection model and detection method | |
Lun et al. | Skip-YOLO: Domestic garbage detection using deep learning method in complex multi-scenes | |
Chen et al. | Classification and recycling of recyclable garbage based on deep learning | |
CN101594314A (en) | A kind of spam image-recognizing method and device based on high-order autocorrelation characteristic | |
Sirimewan et al. | Deep learning-based models for environmental management: Recognizing construction, renovation, and demolition waste in-the-wild | |
CN110458203B (en) | Advertisement image material detection method | |
CN114066861B (en) | Coal gangue identification method based on intersection algorithm edge detection theory and visual characteristics | |
Wu et al. | A novel object detection method to facilitate the recycling of waste small electrical and electronic equipment | |
CN115035442A (en) | Garbage classification collection and transportation supervision method based on improved YOLOv3 network | |
Qiao et al. | A Waste Classification model in Low-illumination scenes based on ConvNeXt | |
Rashida et al. | Implementation of faster region-based convolutional neural network for waste type classification | |
Liu | An intelligent garbage classifier based on deep learning models | |
Saravanan et al. | Enhancing the Accuracy of Meal Waste from Innovative Trash Data Using Random Forest by Comparing Support Vector Machine Algorithms | |
Moradi et al. | FPGA implementation of feature extraction and MLP neural network classifier for Farsi handwritten digit recognition |
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: 20190910 |