CN112381127A - Pearl sorting method based on human bifurcation intervention - Google Patents
Pearl sorting method based on human bifurcation intervention Download PDFInfo
- Publication number
- CN112381127A CN112381127A CN202011210928.4A CN202011210928A CN112381127A CN 112381127 A CN112381127 A CN 112381127A CN 202011210928 A CN202011210928 A CN 202011210928A CN 112381127 A CN112381127 A CN 112381127A
- Authority
- CN
- China
- Prior art keywords
- pearl
- classification
- pearls
- bifurcation
- intervention
- 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
- 238000000034 method Methods 0.000 title claims abstract description 18
- 238000013528 artificial neural network Methods 0.000 claims abstract description 8
- 238000012360 testing method Methods 0.000 claims abstract description 8
- 239000011049 pearl Substances 0.000 claims description 56
- 230000003993 interaction Effects 0.000 claims description 2
- 238000013527 convolutional neural network Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 230000007547 defect Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 239000010984 cultured pearl Substances 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- 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
Abstract
A pearl sorting method based on human bifurcation intervention, comprising: step 1: two independent and advanced neural networks capable of realizing pearl sorting are selected: deploying a ResNet50 model and a SE-ResNet50 model on the server and setting them as a system divergence algorithm; step 2: independently training the two models by using the downloaded pearl data sets, and respectively storing the optimal models; and step 3: setting an arbitrator to process different prediction output results; and 4, step 4: testing the system bifurcation algorithm by using a test set in a pearl data set; and 5: and outputting a final classification result. The invention corrects the prediction output of the two independent systems by the intervention of the diverged driving person, thereby improving the classification precision of the whole pearl sorting.
Description
Technical Field
The invention relates to a pearl detection and classification method, which is suitable for solving the problem that the improvement of pearl sorting precision is limited by pure deep learning by utilizing a human intervention scheme.
Background
Traditionally, pearls are sorted by hand, but the manual sorting of a large number of pearls is undoubtedly inefficient and costly, and the accuracy of the manual sorting is also easily affected by individuals due to the increase in the number and the fatigue of the sorters. The artificially cultured pearls are generally required to be classified into different grades according to the quality and then sold, the reputation of an enterprise and the trust level of a customer can be influenced when high-grade pearls are mixed with low-grade pearls, unnecessary additional cost is caused when the low-grade pearls are mixed with high-grade pearls, and therefore higher requirements are put on the classification precision.
The convolutional neural network CNN can extract the features in the pictures and identify the information in the new pictures, so that the pictures with unknown labels can be classified after a large number of pictures with known labels are trained. The current automatic classification system trained by the deep learning method achieves the sorting precision of 92.57%, but the requirement of enterprises is still insufficient. Therefore, the improvement is made aiming at the prior technical scheme, two neural network algorithms which independently operate are introduced to simultaneously act on the pearl sorting system, and whether the prediction of the system is wrong is judged from the divergence of the predictions of the two algorithms: if the two independent algorithms give the same result, the current system is considered to predict correctly with confidence; if the results diverge, it is reasonable to assume that the accuracy of the current predictions of the system need to be carefully assessed and we opt to be re-judged by a person who is more adept at the task.
Although deep learning can have higher classification precision in data processing capacity compared with other traditional machine learning algorithms, and human labor force is greatly liberated, the current artificial intelligence technology based on deep learning still has the defects of difficult interpretation, poor robustness and the like, so that the algorithm with higher precision is difficult to train, the existing error output cannot be predicted, and the reliability of high-precision application scenes such as pearl sorting is influenced, and economic and credible double losses are brought.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a pearl sorting method based on human bifurcation intervention.
The method can find out the place where the system is likely to make a prediction error by utilizing the divergence between two advanced independent neural network algorithms under the condition of not improving the algorithms, so that the pearl is judged again based on human intervention, and an idea is provided for improving the precision of the pearl sorting method.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a pearl sorting method based on human bifurcation intervention comprises the following steps:
step 1: two independent neural networks capable of realizing pearl sorting are selected: deploying a ResNet50 model and a SE-ResNet50 model on the server and setting them as a system divergence algorithm;
step 2: independently training the two models by using the downloaded pearl data sets, and respectively storing the optimal models;
and step 3: setting an arbitrator to process different prediction output results;
and 4, step 4: testing the system bifurcation algorithm by using a test set in a pearl data set;
and 5: and outputting a final classification result.
The invention provides a method for judging pearls again based on human intervention by finding out the possible error prediction place of the system by using the divergence between two advanced independent neural network algorithms without improving the algorithm, thereby providing an idea for improving the precision of the pearl sorting method. The convolutional neural network CNN can extract the features in the pictures and identify the information in the new pictures, so that the pictures with unknown labels can be classified after a large number of pictures with known labels are trained; the effective intervention of the human can correct the prediction deviation problem existing in the system.
Compared with the prior art, the technical scheme of the invention has the advantages that:
(1) the situation that only one type of algorithm is used for predicting classification errors can be detected more simply and effectively by utilizing the divergence of two independent algorithm networks;
(2) the method combines human intervention to correct error classification, effectively utilizes human advantages to make up for machine errors, and is easy to obtain higher sorting precision.
Drawings
FIG. 1: flow chart of the method of the invention.
Detailed Description
In order to facilitate the understanding and implementation of the present invention for those of ordinary skill in the art, the present invention is further described in detail below with reference to the accompanying drawings and examples.
A pearl sorting method based on human bifurcation intervention comprises the following steps:
step 1: two independent and advanced neural networks capable of realizing pearl sorting are selected: a ResNet50 model and an SE-ResNet50 model are deployed on a server and are set as a system bifurcation algorithm, the two models are required to independently finish a pearl sorting task, and no information interaction is generated during operation;
step 2: independently training the two models by using the downloaded pearl data sets, respectively storing the optimal models, and when the learning rate is 0.0001 and the single-batch training sample is 64, respectively achieving the accuracy of the ResNet50 model and the SE-ResNet50 model to 92.69% and 92.55%, wherein the accuracy is similar and is highest;
and step 3: an arbiter is provided to handle the different predicted output results generated by the two models: comparing the highest probability categories output by each model prediction, and if the highest probability categories are the same, defaulting the system output result; if not, the decision is made by intervention of a person;
and 4, step 4: testing the system bifurcation algorithm by using a test set in a pearl data set;
and 5: outputting a final classification result: if the classification result given by the arbiter selection system is the pearl type, otherwise, the prediction result is given again after the classification is judged by the human, and the result is considered as the final type.
The characteristics corresponding to the pearl surface picture can be divided into 7 categories, and firstly, the characteristics can be roughly divided into two categories according to a rough classification standard: 1) one type is flat or with obvious flaws; 2) the other is more rounded and less flawed; secondly, according to a more refined classification standard, the pearl classification with poor first major classification can be divided into three minor classifications: A1) a pearl having a plurality of flat surfaces; A2) pearls of symmetrical shape; A3) other pearls; the second major category of better pearl classification can be divided into four subclasses: B1) pearls having a ratio of minor to major radii greater than 0.7; B2) a light colored pearl; B3) pearls with hidden speckles; B4) the rest of pearl.
The Convolutional Neural Network and Pearl data set referred to herein are disclosed in "Automatic Pearl Classification Machine base ON a Multistream conditional Neural Network" of IEEE TRANSACTIONS ON INDUSTRIAL ELECTRICAL 2018 by Qi Xuan, Binwei Fang, et al.
The invention corrects the prediction output of the two independent systems by the intervention of the diverged driving person, thereby improving the classification precision of the whole pearl sorting.
The embodiments described in this specification are merely illustrative of implementations of the inventive concept and the scope of the present invention should not be considered limited to the specific forms set forth in the embodiments but rather by the equivalents thereof as may occur to those skilled in the art upon consideration of the present inventive concept.
Claims (2)
1. A pearl sorting method based on human bifurcation intervention comprises the following steps:
step 1: two independent neural networks capable of realizing pearl sorting are selected: a ResNet50 model and an SE-ResNet50 model are deployed on a server and are set as a system bifurcation algorithm, the two models are required to independently finish a pearl sorting task, and no information interaction is generated during operation;
step 2: independently training the two models by using the downloaded pearl data sets, respectively storing the optimal models, and when the learning rate is 0.0001 and the single-batch training sample is 64, respectively achieving the accuracy of the ResNet50 model and the SE-ResNet50 model to 92.69% and 92.55%, wherein the accuracy is similar and is highest;
and step 3: an arbiter is provided to handle the different predicted output results generated by the two models: comparing the highest probability categories output by each model prediction, and if the highest probability categories are the same, defaulting the system output result; if not, the decision is made by intervention of a person;
and 4, step 4: testing the system bifurcation algorithm by using a test set in a pearl data set;
and 5: outputting a final classification result: if the classification result given by the arbiter selection system is the pearl type, otherwise, the prediction result is given again after the classification is judged by the human, and the result is considered as the final type.
2. The method for detecting and classifying pearls according to claim 1, wherein: the characteristics corresponding to the pearl surface picture can be divided into 7 categories, and firstly, the characteristics can be roughly divided into two categories according to a rough classification standard: 1) one type is flat or with obvious flaws; 2) the other is more rounded and less flawed; secondly, according to a more refined classification standard, the pearl classification with poor first major classification can be divided into three minor classifications: A1) a pearl having a plurality of flat surfaces; A2) pearls of symmetrical shape; A3) other pearls; the second major category of better pearl classification can be divided into four subclasses: B1) pearls having a ratio of minor to major radii greater than 0.7; B2) a light colored pearl; B3) pearls with hidden speckles; B4) the rest of pearl.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011210928.4A CN112381127A (en) | 2020-11-03 | 2020-11-03 | Pearl sorting method based on human bifurcation intervention |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011210928.4A CN112381127A (en) | 2020-11-03 | 2020-11-03 | Pearl sorting method based on human bifurcation intervention |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112381127A true CN112381127A (en) | 2021-02-19 |
Family
ID=74578848
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011210928.4A Pending CN112381127A (en) | 2020-11-03 | 2020-11-03 | Pearl sorting method based on human bifurcation intervention |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112381127A (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106529568A (en) * | 2016-10-11 | 2017-03-22 | 浙江工业大学 | Pearl multi-classification method based on BP neural network |
CN106650721A (en) * | 2016-12-28 | 2017-05-10 | 吴晓军 | Industrial character identification method based on convolution neural network |
CN109684922A (en) * | 2018-11-20 | 2019-04-26 | 浙江大学山东工业技术研究院 | A kind of recognition methods based on the multi-model of convolutional neural networks to finished product dish |
CN110263774A (en) * | 2019-08-19 | 2019-09-20 | 珠海亿智电子科技有限公司 | A kind of method for detecting human face |
-
2020
- 2020-11-03 CN CN202011210928.4A patent/CN112381127A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106529568A (en) * | 2016-10-11 | 2017-03-22 | 浙江工业大学 | Pearl multi-classification method based on BP neural network |
CN106650721A (en) * | 2016-12-28 | 2017-05-10 | 吴晓军 | Industrial character identification method based on convolution neural network |
CN109684922A (en) * | 2018-11-20 | 2019-04-26 | 浙江大学山东工业技术研究院 | A kind of recognition methods based on the multi-model of convolutional neural networks to finished product dish |
CN110263774A (en) * | 2019-08-19 | 2019-09-20 | 珠海亿智电子科技有限公司 | A kind of method for detecting human face |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111179251B (en) | Defect detection system and method based on twin neural network and by utilizing template comparison | |
CN102682760B (en) | Overlapped voice detection method and system | |
CN103136504B (en) | Face identification method and device | |
CN112465040B (en) | Software defect prediction method based on class unbalance learning algorithm | |
CN109919252B (en) | Method for generating classifier by using few labeled images | |
CN106096661B (en) | The zero sample image classification method based on relative priority random forest | |
CN111752775B (en) | Disk fault prediction method and system | |
JP2020123330A (en) | Method for acquiring sample image for label acceptance inspection from among auto-labeled images utilized for neural network learning, and sample image acquisition device utilizing the same | |
CN112150450B (en) | Image tampering detection method and device based on dual-channel U-Net model | |
CN103544506A (en) | Method and device for classifying images on basis of convolutional neural network | |
CN110688888B (en) | Pedestrian attribute identification method and system based on deep learning | |
CN110503161B (en) | Ore mud ball target detection method and system based on weak supervision YOLO model | |
US20230385333A1 (en) | Method and system for building training database using automatic anomaly detection and automatic labeling technology | |
CN111522951A (en) | Sensitive data identification and classification technical method based on image identification | |
CN111127454A (en) | Method and system for generating industrial defect sample based on deep learning | |
CN114627383A (en) | Small sample defect detection method based on metric learning | |
US20220222807A1 (en) | Ai-based new learning model generation system for vision inspection on product production line | |
CN110569780A (en) | high-precision face recognition method based on deep transfer learning | |
CN112200392B (en) | Service prediction method and device | |
CN111950645A (en) | Method for improving class imbalance classification performance by improving random forest | |
CN117011563B (en) | Road damage inspection cross-domain detection method and system based on semi-supervised federal learning | |
Melhart et al. | A study on affect model validity: Nominal vs ordinal labels | |
CN115063664A (en) | Model learning method, training method and system for industrial vision detection | |
CN112381127A (en) | Pearl sorting method based on human bifurcation intervention | |
JP7123306B2 (en) | Image processing device and image processing method |
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 |