CN114529523A - Industrial defect degree inference method and system based on uncertainty - Google Patents
Industrial defect degree inference method and system based on uncertainty Download PDFInfo
- Publication number
- CN114529523A CN114529523A CN202210143237.XA CN202210143237A CN114529523A CN 114529523 A CN114529523 A CN 114529523A CN 202210143237 A CN202210143237 A CN 202210143237A CN 114529523 A CN114529523 A CN 114529523A
- Authority
- CN
- China
- Prior art keywords
- uncertainty
- predictor
- industrial
- trained
- detection model
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
- G06N5/041—Abduction
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention provides an uncertainty-based industrial defect degree inference method and system, comprising the following steps of: step 1: obtaining training image data according to original image data of industrial products; step 2: constructing a detection model to be trained according to the convolutional neural network model and the full convolutional neural network; and step 3: training a detection model to be trained by using training image data to obtain a trained detection model; and 4, step 4: and deducing the industrial defect degree of the industrial product according to the uncertainty value output by the trained detection model. Compared with the prior art, the method uses uncertainty to infer the defect degree of the current industrial product, and avoids the problems that the industrial quality inspection index is too single, the quality is roughly classified discretely, the labeling quality of a sample is poor, the performance of the model is influenced, and the like.
Description
Technical Field
The invention relates to the technical field of industrial quality inspection, in particular to an uncertainty-based industrial defect degree inference method and system.
Background
The detection of the surface defects of the industrial products is an important link for evaluating the quality of the products and an important means for ensuring the quality and the production efficiency of the products. Surface defect detection of products is a technical problem of particular interest in the manufacturing industry in recent years. As an essential step in the production and manufacturing process, surface defect detection is widely applied to various industrial fields, including industries such as 3C, semiconductors, electronics, automobiles, chemical engineering, medicine, light industry, military industry and the like, and a plurality of upstream and downstream enterprises are promoted. Surface defect inspection has gone through roughly three stages since the beginning of the 20 th century, namely manual visual inspection, single electromechanical or optical inspection, and machine vision inspection. With the rapid development of photoelectric components and the intensive research of algorithms such as image processing, artificial intelligence and the like in computer technology, advanced methods represented by machine vision are more and more widely applied to industrial quality inspection.
The existing industrial product surface defect detection method mainly carries out industrial quality detection by directly detecting related industrial defect characteristics, but indexes are too single, the quality of industrial products is simply classified into a plurality of discrete categories such as excellent, good and unqualified, the fact that the industrial defect degree is continuous is not considered, the defect degree is different among the same categories, and samples with poor annotation quality in samples interfere with the learning of the models, so that the prediction results of the models are poor.
Patent document CN111754497A discloses an industrial defect detection method and system based on geometric algebra. The system introduces GA-U-net to score the severity of industrial product surface defects in the image, then obtains image pixels with scores higher than a preset threshold value through a defect filter, then obtains an image with complete defects through a connected domain analysis module, and then performs defect type evaluation on the 3D image with the complete defects through a stack type deep neural network based on geometric algebra fuzzy pooling to complete detection and classification of the industrial defects. However, the method does not effectively solve the problems that the index is too single, the quality of industrial products is simply classified into a plurality of discrete categories, the fact that the industrial defect degree is continuous and the defect degrees are different among the same category is not considered, and the samples with poor labeling quality in the samples interfere with the learning of the model, so that the prediction result of the model is poor.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide an industrial defect degree inference method and system based on uncertainty.
The invention provides an uncertainty-based industrial defect degree inference method, which comprises the following steps of:
step 1: obtaining training image data according to original image data of industrial products;
step 2: constructing a detection model to be trained according to the convolutional neural network model and the full convolutional neural network;
and step 3: training a detection model to be trained by using training image data to obtain a trained detection model;
and 4, step 4: and deducing the industrial defect degree of the industrial product according to the uncertainty value output by the trained detection model.
Preferably, step 2, comprises:
step 201: extracting image features according to a convolutional neural network model of a coding and decoding structure to obtain a feature map;
step 202: constructing a classification predictor, a regression predictor and an uncertainty predictor according to the full convolution neural network;
step 203: obtaining classification information, regression information and uncertainty information according to the feature map, the classification predictor, the regression predictor and the uncertainty predictor;
step 204: and obtaining a detection model to be trained according to the convolutional neural network model, the classification predictor, the regression predictor and the uncertainty predictor of the coding and decoding structure.
Preferably, step 3, comprises:
step 301: obtaining a loss function according to the training graphic data and the detection model to be trained;
step 302: obtaining the gradient of the learnable parameters in the detection model to be trained through a gradient back propagation algorithm;
step 303: updating the learning parameters according to the relevant configuration of the optimizer, and continuing to perform forward propagation;
step 304: and when the iteration is carried out until the loss function meets the preset condition, finishing the training of the detection model to be trained to obtain the trained detection model.
Preferably, step 4, comprises:
step 401: inputting the industrial image to be detected into the trained detection model to obtain classification information output by the classification predictor, regression information output by the regression predictor and uncertain information output by the uncertainty predictor;
step 402: filtering the classified information by using a filter to obtain a filtering result;
step 403: extracting corresponding regression numerical values and uncertainty numerical values according to the filtering result;
step 404: and deducing the industrial defect degree of the industrial image to be detected according to the uncertainty value.
Preferably, step 1, comprises:
step 101: acquiring original image data of an industrial product;
step 102: and carrying out image processing on the original image data to obtain training image data.
The invention provides an uncertainty-based industrial defect degree inference system, which comprises:
module M1: obtaining training image data according to original image data of an industrial product;
module M2: constructing a detection model to be trained according to the convolutional neural network model and the full convolutional neural network;
module M3: training a detection model to be trained by using training image data to obtain a trained detection model;
module M4: and deducing the industrial defect degree of the industrial product according to the uncertainty value output by the trained detection model.
Preferably, the module M2, comprises:
submodule M201: extracting image features according to a convolutional neural network model of a coding and decoding structure to obtain a feature map;
submodule M202: constructing a classification predictor, a regression predictor and an uncertainty predictor according to the full convolution neural network;
submodule M203: obtaining classification information, regression information and uncertainty information according to the feature map, the classification predictor, the regression predictor and the uncertainty predictor;
submodule M204: and obtaining a detection model to be trained according to the convolutional neural network model, the classification predictor, the regression predictor and the uncertainty predictor of the coding and decoding structure.
Preferably, the module M3 includes:
submodule M301: obtaining a loss function according to the training graphic data and the detection model to be trained;
submodule M302: obtaining the gradient of the learnable parameters in the detection model to be trained through a gradient back propagation algorithm;
submodule M303: updating the learning parameters according to the relevant configuration of the optimizer, and continuing to perform forward propagation;
submodule M304: and when the iteration is carried out until the loss function meets the preset condition, finishing the training of the detection model to be trained to obtain the trained detection model.
Preferably, the module M4, comprises:
submodule M401: inputting an industrial image to be detected into a trained detection model to obtain classification information output by a classification predictor, regression information output by a regression predictor and uncertain information output by an uncertainty predictor;
submodule M402: filtering the classified information by using a filter to obtain a filtering result;
submodule M403: extracting corresponding regression numerical values and uncertainty numerical values according to the filtering result;
sub-module M404: and deducing the industrial defect degree of the industrial image to be detected according to the uncertainty value.
Preferably, the module M1, comprises:
submodule M101: acquiring original image data of an industrial product;
sub-module M102: and carrying out image processing on the original image data to obtain training image data.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention completes the defect degree inference through the uncertainty value of the model prediction, and enriches the industrial quality inspection method.
2. The uncertainty given by the invention is a continuous numerical value, is not a plurality of discrete classifications, and is adaptive to the continuous defect degree of industrial products.
3. According to the invention, the weight of the sample with poor labeling quality is reduced in a self-adaptive manner through the uncertainty of model prediction, so that the model focuses on the sample with high labeling quality, and the model precision is better.
Drawings
Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments with reference to the following drawings:
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 is a logic diagram of the present invention.
Detailed Description
The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that it would be obvious to those skilled in the art that various changes and modifications can be made without departing from the spirit of the invention. All falling within the scope of the present invention.
Fig. 1 is a schematic flow chart of the present invention, and as shown in fig. 1, the present invention provides an uncertainty-based industrial defect level inference method, which includes the following steps:
step 1: and obtaining training image data according to the original image data of the industrial product.
Preferably, step 1, comprises: step 101: acquiring original image data of an industrial product; step 102: and carrying out image processing on the original image data to obtain training image data.
Specifically, in order to obtain image data that can be used for inferring the industrial defect degree, relevant hardware such as a lens light source is selected for a specific industrial quality inspection application scene, so that the defect degree of a corresponding industrial product can be reflected in the acquired image data. After the hardware part is built, parameters such as camera shooting resolution and the like are set, and original image data of an actual industrial quality inspection application scene are obtained through shooting. Further, image processing is carried out on the original image data to obtain data used for training the neural network. And according to the actual industrial quality inspection scene requirements, completing the labeling of classification information and regression information in the original image data to obtain training image data.
The specific type of image processing in the present invention is not limited, and may be, for example, uniform format conversion, data enhancement, and the like.
Step 2: and constructing a detection model to be trained according to the convolutional neural network model and the full convolutional neural network.
Preferably, step 2, comprises: step 201: extracting image features according to a convolutional neural network model of a coding and decoding structure to obtain a feature map; step 202: constructing a classification predictor, a regression predictor and an uncertainty predictor according to the full convolution neural network; step 203: obtaining classification information, regression information and uncertainty information according to the feature map, the classification predictor, the regression predictor and the uncertainty predictor; step 204: and obtaining a detection model to be trained according to the convolutional neural network model, the classification predictor, the regression predictor and the uncertainty predictor of the coding and decoding structure.
Specifically, a neural network for deducing the industrial defect degree is built, firstly, a convolutional neural network with a coding and decoding structure is used for extracting a feature map of an input image, the backbone network adopts the idea of transfer learning, and model parameters obtained after pre-training on a large universal visual data set are used as initial parameters during actual industrial quality inspection model training. The hourglass network and the residual error network variants and the like can be used as backbone network extraction features. Next, classification information, regression information, and uncertainty information are predicted separately using three parallel branches. The models of the three parallel branches are all full convolution neural networks, and after the network building is completed, the detection model to be trained is obtained.
Wherein the three parallel branches comprise: the method comprises the following steps of classifying branches, regression branches and uncertainty branches, wherein the classification branches predict confidence coefficients that different positions in input original image data belong to various target classes; the regression branch predicts the target size information in the industrial quality inspection image; and the uncertainty branch is used for adaptively giving uncertainty corresponding to the prediction result and is used for deducing the industrial defect degree.
And step 3: and training the detection model to be trained by using the training image data to obtain the trained detection model.
Preferably, step 3, comprises: step 301: obtaining a loss function according to the training graphic data and the detection model to be trained; step 302: obtaining the gradient of the learnable parameters in the detection model to be trained through a gradient back propagation algorithm; step 303: updating the learning parameters according to the relevant configuration of the optimizer, and continuing to perform forward propagation; step 304: and when the iteration is carried out until the loss function meets the preset condition, finishing the training of the detection model to be trained to obtain the trained detection model.
Specifically, training image data is used, a detection model to be trained is input, and forward calculation of the model is completed.
Wherein, the classification branch adopts the focus loss for supervision, which can be expressed by formula (1):
wherein L isKRepresents a classification loss; n represents the number of objects in the image;representing a prediction result;representing a target result; alpha represents a hyper-parameter; beta represents a hyper-ginseng; x represents the abscissa; y represents the ordinate; and c represents a channel. Coefficient of performanceAndthe method is used for solving the problem that difficult and easy samples are not balanced in an industrial quality inspection scene, reducing the weight of a large number of simple background samples in the image and enabling the training process to pay more attention to an important foreground part.
Further, the predicted results of the regression branch and the uncertainty branch together form a predicted gaussian distribution, which can be expressed by formula (2):
wherein theta represents the nomenclature of all trainable parameters in the convolutional neural network,representing predictionsThe degree of uncertainty of (a) is,represents the predicted distribution, δxIndicating a lateral displacement. The target distribution during training is a dirac function, which can be expressed by equation (3):
wherein the content of the first and second substances,representing a target distribution;representing a dirac function; deltaxRepresents a lateral displacement;representing the target lateral displacement.
Specifically, the joint loss function of the regression branch and the uncertainty branch is obtained based on the predicted gaussian distribution and the KL divergence (Kullback-Leibler divergence) of the target distribution, and can be represented by formula (4):
wherein the content of the first and second substances,indicating a KL divergence of the predicted distribution from the target distribution;representing the entropy of the target distribution.
After derivation, specifically, derivation includes removing the constant part, replacing variables, etc., to obtain the loss functions of the regression branch and the uncertainty branch, which can be expressed by formula (5):
wherein L isDURepresenting the joint loss of regression branches and uncertainty branches; n represents the number of objects in the image;representing a weight coefficient;representing a weight coefficient;representing a target lateral displacement;representing a predicted lateral displacement;represents the lateral prediction uncertainty;representing longitudinal prediction uncertainty;representing a target longitudinal displacement;indicating the predicted longitudinal displacement.
The loss function calculates the difference between the actual output and the expected output of the detection model to be trained, the gradient of learnable parameters in the detection model to be trained can be calculated through a gradient back propagation algorithm, the learnable parameters are updated according to the relevant configuration of the optimizer, forward propagation and multiple iterations are carried out continuously, and after the iterations meet the relevant requirements, the model is trained to obtain the trained detection model.
And 4, step 4: and deducing the industrial defect degree of the industrial product according to the uncertainty value output by the trained detection model.
Preferably, step 4, comprises: step 401: inputting the industrial image to be detected into the trained detection model to obtain classification information output by the classification predictor, regression information output by the regression predictor and uncertain information output by the uncertainty predictor; step 402: filtering the classified information by using a filter to obtain a filtering result; step 403: extracting corresponding regression numerical values and uncertainty numerical values according to the filtering result; step 404: and deducing the industrial defect degree of the industrial image to be detected according to the uncertainty value.
Specifically, the obtained trained detection model is used in an actual industrial defect degree inference scene. Inputting an industrial image to be detected shot in an industrial quality inspection scene into a trained detection model, and respectively outputting predicted classification information, regression information and uncertainty information by predictors of three parallel branches of the trained detection model. The channels with different classification prediction tensors represent different classes, different positions correspond to positions of the original image after down-sampling, and specific numerical values represent confidence coefficients of the current position belonging to the corresponding classes. And removing the result with the classification confidence coefficient lower than a certain threshold value by using a filter, and extracting the corresponding regression value and uncertainty value to form a final model detection result. And the uncertainty output by the trained detection model is used for deducing the industrial defect degree, and the larger the uncertainty value is, namely the larger the difference between the current sample and the normal qualified sample is, the more serious the defect degree of the industrial sample is.
FIG. 2 is a logic diagram of the present invention, as shown in FIG. 2, which includes inputting an image to a feature extractor for encoding and decoding, extracting features, obtaining image features, inputting the image features into three predictors in parallel branches, namely, a classification predictor of a classification branch, a regression predictor of a regression branch, and an uncertainty predictor of an uncertainty branch, obtaining corresponding predicted classification tensor, predicted regression tensor, and predicted uncertainty tensor, further, removing a result that a classification confidence of the predicted classification tensor is lower than a certain threshold value through a filter, simultaneously, extracting the corresponding predicted regression tensor and predicted uncertainty tensor to obtain corresponding regression information and uncertainty information, forming a final detection result, wherein the uncertainty is used for deducing an industrial defect degree, and the larger an uncertainty value represents that a current sample and a normal qualified sample come in and go out, the more severe the industrial sample is defective.
The invention also provides an uncertainty-based industrial defect degree inference system, which comprises:
module M1: and obtaining training image data according to the original image data of the industrial product.
Preferably, the module M1, comprises: submodule M101: acquiring original image data of an industrial product; submodule M102: and carrying out image processing on the original image data to obtain training image data.
Module M2: and constructing a detection model to be trained according to the convolutional neural network model and the full convolutional neural network.
Preferably, the module M2, comprises: submodule M201: extracting image features according to a convolutional neural network model of a coding and decoding structure to obtain a feature map; submodule M202: constructing a classification predictor, a regression predictor and an uncertainty predictor according to the full convolution neural network; submodule M203: obtaining classification information, regression information and uncertainty information according to the feature map, the classification predictor, the regression predictor and the uncertainty predictor; submodule M204: and obtaining a detection model to be trained according to the convolutional neural network model, the classification predictor, the regression predictor and the uncertainty predictor of the coding and decoding structure.
Module M3: and training the detection model to be trained by using the training image data to obtain the trained detection model.
Preferably, the module M3, comprises: submodule M301: obtaining a loss function according to the training graphic data and the detection model to be trained; submodule M302: obtaining the gradient of the learnable parameters in the detection model to be trained through a gradient back propagation algorithm; submodule M303: updating the learning parameters according to the relevant configuration of the optimizer, and continuing to perform forward propagation; submodule M304: and when the iteration is carried out until the loss function meets the preset condition, finishing the training of the detection model to be trained to obtain the trained detection model.
Module M4: and deducing the industrial defect degree of the industrial product according to the uncertainty value output by the trained detection model.
Preferably, the module M4, comprises: submodule M401: inputting the industrial image to be detected into the trained detection model to obtain classification information output by the classification predictor, regression information output by the regression predictor and uncertain information output by the uncertainty predictor; submodule M402: filtering the classified information by using a filter to obtain a filtering result; submodule M403: extracting corresponding regression numerical values and uncertainty numerical values according to the filtering result; sub-module M404: and deducing the industrial defect degree of the industrial image to be detected according to the uncertainty value.
The technical problem solved by the invention is as follows:
1. the industrial quality inspection is carried out by directly detecting the characteristics of the related industrial defects, and the indexes are too single.
2. The quality of industrial products is simply classified into a plurality of discrete categories, such as excellent, good, unqualified and the like, and the condition that the industrial defect degree is continuous and the defect degree is different among the same categories is not considered.
3. The samples with poor labeling quality in the samples interfere with the learning of the model, so that the prediction result of the model is poor.
The technical principle of the invention is as follows:
1. the invention provides a novel method for deducing the industrial defect degree, which deduces the severity of the industrial defect by predicting uncertainty and solves the following problems in an industrial quality inspection scene: the detection index is too single, the quality is roughly classified discretely, and the performance of the model is influenced by the samples with poor labeling quality.
2. The invention provides a novel detection model for deducing the industrial defect degree, and the classification information, regression information and uncertainty information of an industrial target are predicted by using a convolutional neural network structure with three parallel branches, so that the problems that an industrial quality inspection model is difficult to super-parameter adjust, the model is poor in robustness and cannot solve a complex industrial scene are solved.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention provides a new industrial quality inspection index, which completes defect degree inference through uncertainty numerical values predicted by a model, and enriches industrial quality inspection methods.
2. The uncertainty given by the invention is a continuous numerical value, is not a plurality of discrete classifications, and is adaptive to the continuous defect degree of industrial products.
3. According to the invention, the weight of the sample with poor labeling quality is reduced in a self-adaptive manner through the uncertainty of prediction, so that the model focuses on the sample with high labeling quality, and the model precision is better.
Those skilled in the art will appreciate that, in addition to implementing the systems, apparatus, and various modules thereof provided by the present invention in purely computer readable program code, the same procedures can be implemented entirely by logically programming method steps such that the systems, apparatus, and various modules thereof are provided in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Therefore, the system, the device and the modules thereof provided by the present invention can be considered as a hardware component, and the modules included in the system, the device and the modules thereof for implementing various programs can also be considered as structures in the hardware component; modules for performing various functions may also be considered to be both software programs for performing the methods and structures within hardware components.
The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes or modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention. The embodiments and features of the embodiments of the present application may be combined with each other arbitrarily without conflict.
Claims (10)
1. An uncertainty-based industrial defect level inference method, the method comprising:
step 1: obtaining training image data according to original image data of industrial products;
and 2, step: constructing a detection model to be trained according to the convolutional neural network model and the full convolutional neural network;
and step 3: training the detection model to be trained by using the training image data to obtain a trained detection model;
and 4, step 4: and deducing the industrial defect degree of the industrial product according to the uncertainty value output by the trained detection model.
2. The uncertainty-based industrial fault level inference method of claim 1, wherein said step 2, comprises:
step 201: extracting image features according to the convolutional neural network model of the coding and decoding structure to obtain a feature map;
step 202: constructing a classification predictor, a regression predictor and an uncertainty predictor according to the full-convolution neural network;
step 203: obtaining classification information, regression information and uncertainty information according to the feature map, the classification predictor, the regression predictor and the uncertainty predictor;
step 204: and obtaining the detection model to be trained according to the convolutional neural network model, the classification predictor, the regression predictor and the uncertainty predictor of the coding and decoding structure.
3. The uncertainty-based industrial defect level inference method of claim 1 or 2, wherein said step 3, comprises:
step 301: obtaining a loss function according to the training graphic data and the detection model to be trained;
step 302: obtaining the gradient of the learnable parameters in the detection model to be trained through a gradient back propagation algorithm;
step 303: updating the learning parameters according to the related configuration of the optimizer, and continuing to perform forward propagation;
step 304: and when the iteration is carried out until the loss function meets the preset condition, finishing the training of the detection model to be trained to obtain the trained detection model.
4. The uncertainty-based industrial defect level inference method of claim 1 or 2, wherein said step 4, comprises:
step 401: inputting the industrial image to be detected into the trained detection model to obtain the classification information output by the classification predictor, the regression information output by the regression predictor and the uncertain information output by the uncertainty predictor;
step 402: filtering the classified information by using a filter to obtain a filtering result;
step 403: extracting corresponding regression numerical values and uncertainty numerical values according to the filtering result;
step 404: and deducing the industrial defect degree of the industrial image to be detected according to the uncertainty value.
5. The uncertainty-based industrial defect level inference method of claim 4, wherein said step 1, comprises:
step 101: obtaining the raw image data of the industrial product;
step 102: and performing image processing on the original image data to obtain the training image data.
6. An uncertainty-based industrial defect level inference system, the system comprising:
module M1: obtaining training image data according to original image data of industrial products;
module M2: constructing a detection model to be trained according to the convolutional neural network model and the full convolutional neural network;
module M3: training the detection model to be trained by using the training image data to obtain a trained detection model;
module M4: and deducing the industrial defect degree of the industrial product according to the uncertainty value output by the trained detection model.
7. The uncertainty-based industrial fault severity inference system of claim 6, wherein said module M2, comprising:
submodule M201: extracting image features according to the convolutional neural network model of the coding and decoding structure to obtain a feature map;
submodule M202: constructing a classification predictor, a regression predictor and an uncertainty predictor according to the full-convolution neural network;
submodule M203: obtaining classification information, regression information and uncertainty information according to the feature map, the classification predictor, the regression predictor and the uncertainty predictor;
submodule M204: and obtaining the detection model to be trained according to the convolutional neural network model, the classification predictor, the regression predictor and the uncertainty predictor of the coding and decoding structure.
8. The uncertainty-based industrial defect level inference system of claim 6 or 7, wherein the module M3 comprises:
submodule M301: obtaining a loss function according to the training graphic data and the detection model to be trained;
submodule M302: obtaining the gradient of the learnable parameters in the detection model to be trained through a gradient back propagation algorithm;
submodule M303: updating the learning parameters according to the related configuration of the optimizer, and continuing to perform forward propagation;
submodule M304: and when the iteration is carried out until the loss function meets the preset condition, finishing the training of the detection model to be trained to obtain the trained detection model.
9. The uncertainty-based industrial defect level inference system of claim 6 or 7, wherein the module M4 comprises:
submodule M401: inputting the industrial image to be detected into the trained detection model to obtain the classification information output by the classification predictor, the regression information output by the regression predictor and the uncertain information output by the uncertainty predictor;
submodule M402: filtering the classified information by using a filter to obtain a filtering result;
submodule M403: extracting corresponding regression numerical values and uncertainty numerical values according to the filtering result;
submodule M404: and deducing the industrial defect degree of the industrial image to be detected according to the uncertainty value.
10. The uncertainty-based industrial fault severity inference system of claim 9, wherein said module M1, comprising:
submodule M101: obtaining the raw image data of the industrial product;
sub-module M102: and carrying out image processing on the original image data to obtain the training image data.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210143237.XA CN114529523A (en) | 2022-02-16 | 2022-02-16 | Industrial defect degree inference method and system based on uncertainty |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210143237.XA CN114529523A (en) | 2022-02-16 | 2022-02-16 | Industrial defect degree inference method and system based on uncertainty |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114529523A true CN114529523A (en) | 2022-05-24 |
Family
ID=81622861
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210143237.XA Pending CN114529523A (en) | 2022-02-16 | 2022-02-16 | Industrial defect degree inference method and system based on uncertainty |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114529523A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116129123A (en) * | 2023-02-27 | 2023-05-16 | 中国矿业大学 | End-to-end chromosome segmentation method based on uncertainty calibration and region decomposition |
-
2022
- 2022-02-16 CN CN202210143237.XA patent/CN114529523A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116129123A (en) * | 2023-02-27 | 2023-05-16 | 中国矿业大学 | End-to-end chromosome segmentation method based on uncertainty calibration and region decomposition |
CN116129123B (en) * | 2023-02-27 | 2024-01-05 | 中国矿业大学 | End-to-end chromosome segmentation method based on uncertainty calibration and region decomposition |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109543502B (en) | Semantic segmentation method based on deep multi-scale neural network | |
CN108334835B (en) | Method for detecting visible components in vaginal secretion microscopic image based on convolutional neural network | |
CN112381763A (en) | Surface defect detection method | |
CN110852417B (en) | Single-depth neural network model robustness improving method for application of Internet of things | |
CN111753986A (en) | Dynamic testing method and device for deep learning model | |
CN115131558B (en) | Semantic segmentation method in environment with few samples | |
CN114529523A (en) | Industrial defect degree inference method and system based on uncertainty | |
CN117132827A (en) | Hot rolled steel strip surface defect detection method based on improved YOLOv5s network | |
CN111612803B (en) | Vehicle image semantic segmentation method based on image definition | |
US20230326191A1 (en) | Method and Apparatus for Enhancing Performance of Machine Learning Classification Task | |
CN109800793B (en) | Target detection method and system based on deep learning | |
CN116229323A (en) | Human body behavior recognition method based on improved depth residual error network | |
CN115631186A (en) | Industrial element surface defect detection method based on double-branch neural network | |
CN113269786B (en) | Assembly image segmentation method and device based on deep learning and guided filtering | |
CN111652083B (en) | Weak supervision time sequence action detection method and system based on self-adaptive sampling | |
CN112016403B (en) | Video abnormal event detection method | |
CN114332099A (en) | Deep privilege semantic segmentation method based on multi-modal contrast learning | |
CN114120202A (en) | Semi-supervised video target segmentation method based on multi-scale target model and feature fusion | |
CN114494284A (en) | Scene analysis model and method based on explicit supervision area relation | |
Runhu et al. | Semantic segmentation using DeepLabv3+ model for fabric defect detection | |
CN113008998A (en) | Hidden engineering internal defect judgment method based on PCNN | |
CN113505751B (en) | Human skeleton action recognition method based on difference map convolutional neural network | |
CN113744306B (en) | Video target segmentation method based on time sequence content perception attention mechanism | |
CN114821206B (en) | Multi-modal image fusion classification method and system based on confrontation complementary features | |
CN112052626B (en) | Automatic design system and method for neural network |
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 |