CN117541858A - Product detection method and device based on dual-branch network, storage medium and electronic equipment - Google Patents
Product detection method and device based on dual-branch network, storage medium and electronic equipment Download PDFInfo
- Publication number
- CN117541858A CN117541858A CN202311479248.6A CN202311479248A CN117541858A CN 117541858 A CN117541858 A CN 117541858A CN 202311479248 A CN202311479248 A CN 202311479248A CN 117541858 A CN117541858 A CN 117541858A
- Authority
- CN
- China
- Prior art keywords
- network
- branch network
- dual
- product
- boundary point
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 86
- 238000003860 storage Methods 0.000 title claims abstract description 16
- 238000000034 method Methods 0.000 claims abstract description 36
- 238000012545 processing Methods 0.000 claims abstract description 26
- 238000012549 training Methods 0.000 claims description 81
- 238000004590 computer program Methods 0.000 claims description 10
- 230000009977 dual effect Effects 0.000 claims description 5
- 230000008014 freezing Effects 0.000 claims description 3
- 238000007710 freezing Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 description 19
- 230000006870 function Effects 0.000 description 13
- 238000010586 diagram Methods 0.000 description 7
- 238000000605 extraction Methods 0.000 description 7
- 239000013598 vector Substances 0.000 description 7
- 238000009826 distribution Methods 0.000 description 5
- 238000013145 classification model Methods 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 3
- 230000003190 augmentative effect Effects 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 2
- 150000001875 compounds Chemical class 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/778—Active pattern-learning, e.g. online learning of image or video features
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a product detection method, a device, a storage medium and electronic equipment based on a dual-branch network, wherein the method comprises the following steps: acquiring a product image and a dual-branch network, wherein the dual-branch network comprises a first branch network and a second branch network, the first branch network comprises a teacher network, the second branch network comprises a student network and a boundary point classifier, and the boundary point classifier comprises at least one parameterized boundary point corresponding to each known category; inputting the product image into a dual-branch network for processing to obtain a first prediction probability output by a first branch network for an unknown class and a second prediction probability output by a second branch network for a known class; according to the first prediction probability and the second prediction probability, the target category corresponding to the product to be detected is determined from the unknown category and the known category, so that the accurate classification of various industrial products with known categories can be realized, and the known category and the unknown category can be effectively distinguished.
Description
Technical Field
The application belongs to the technical field of industrial detection, and particularly relates to a product detection method and device based on a dual-branch network, a storage medium and electronic equipment.
Background
In recent years, the field of deep learning has been continuously developed, and the deep learning model is also beginning to be applied in the field of industrial detection.
At present, a deep learning model used in the industrial detection field is mainly a classification model and is used for classifying part products, such as defect classification of various products, however, because of various defect types, it is difficult to collect all defect type images for training the classification model, so that when the classification model is used for industrial detection, detection images of unknown defect types are identified as known defect types, false detection and missing detection phenomena are caused, and the model identification accuracy is low.
Disclosure of Invention
The present application aims to solve at least one of the technical problems existing in the prior art. Therefore, the product detection method and device based on the dual-branch network, the storage medium and the electronic equipment can effectively identify unknown type targets and known type targets, and improve model identification accuracy.
In a first aspect, the present application provides a product detection method based on a dual-branch network, including:
Acquiring a product image of a product to be detected and a trained dual-branch network, wherein the dual-branch network comprises a first branch network and a second branch network, the first branch network comprises a teacher network, the second branch network comprises a student network and a boundary point classifier, the boundary point classifier comprises at least one parameterized boundary point corresponding to each known category in a plurality of known categories, and the teacher network and the student network are used for extracting image features;
inputting the product image into the dual-branch network for processing to obtain a first prediction probability output by the first branch network for an unknown class and a second prediction probability output by the second branch network for the known class;
and determining a target category corresponding to the product to be detected from the unknown category and the known category according to the first prediction probability and the second prediction probability so as to carry out industrial detection on the product to be detected.
In some embodiments, the first branch network further includes a prediction module, the inputting the product image into the dual branch network for processing, comprising:
Extracting image features of the product image through the teacher network and the student network respectively to obtain a first image feature corresponding to the teacher network and a second image feature corresponding to the student network;
outputting, by the prediction module, a first prediction probability for an unknown class according to the first image feature and the second image feature;
and outputting second prediction probabilities for a plurality of known categories according to the second image features and the boundary points through the boundary point classifier.
In some embodiments, the outputting, by the prediction module, a first prediction probability for an unknown class from the first image feature and the second image feature comprises:
calculating, by the prediction module, a first distance between the first image feature and the second image feature;
and determining a first prediction probability of the product to be detected belonging to the unknown class according to the first distance.
In some embodiments, the outputting, by the boundary point classifier, a second prediction probability for the known class from the second image feature and the boundary point comprises:
Calculating a second distance between the second image feature and the boundary point corresponding to each known category through the boundary point classifier;
and determining a second prediction probability of the product to be detected belonging to the corresponding known category according to the second distance.
In some embodiments, the determining, according to the first prediction probability and the second prediction probability, a target category corresponding to the product to be detected from the unknown category and the known category includes:
when the first prediction probability is greater than or equal to a preset threshold value, the unknown class is used as a target class corresponding to the product to be detected;
and when the first prediction probability is smaller than the preset threshold value, taking the known category of the boundary point corresponding to the second prediction probability with the largest value as the target category corresponding to the product to be detected.
In some embodiments, further comprising:
constructing the dual-branch network and a typical sample classifier, and constructing at least one initialized boundary point corresponding to each known category by utilizing the boundary point classifier;
acquiring a training sample image set and the known category corresponding to each training sample image in the training sample image set;
Training the dual-branch network according to the training sample image set, the known class and the typical sample classifier, and freezing model parameters of the teacher network in the training process.
In some embodiments, the training the dual-branch network according to the training sample image set, the known class, and the canonical sample classifier includes:
extracting image features of the training sample images through the teacher network and the student network respectively to obtain first sample image features corresponding to the teacher network and second sample image features corresponding to the student network;
determining a first loss value corresponding to the first branch network according to the first sample image characteristic and the second sample image characteristic;
determining a second loss value corresponding to the boundary point classifier according to the second sample image characteristic, the boundary point and the known category;
determining a third loss value corresponding to the typical sample classifier according to the second sample image features and the known category;
and reversely adjusting model parameters of the student network and the boundary point classifier in the dual-branch network according to the first loss value, the second loss value and the third loss value so as to train the dual-branch network.
In a second aspect, the present application provides a product detection apparatus based on a dual-branch network, including:
an acquisition unit, configured to acquire a product image of a product to be detected, and a trained dual-branch network, where the dual-branch network includes a first branch network and a second branch network, the first branch network includes a teacher network, the second branch network includes a student network and a boundary point classifier, the boundary point classifier includes at least one parameterized boundary point corresponding to each of a plurality of known categories, and the teacher network and the student network are configured to extract image features;
the processing unit is used for inputting the product image into the dual-branch network for processing so as to obtain a first prediction probability output by the first branch network for an unknown class and a second prediction probability output by the second branch network for the known class;
and the determining unit is used for determining a target category corresponding to the product to be detected from the unknown category and the known category according to the first prediction probability and the second prediction probability so as to carry out industrial detection on the product to be detected.
In some embodiments, the first branch network further comprises a prediction module, and the processing unit is specifically configured to:
extracting image features of the product image through the teacher network and the student network respectively to obtain a first image feature corresponding to the teacher network and a second image feature corresponding to the student network;
outputting, by the prediction module, a first prediction probability for an unknown class according to the first image feature and the second image feature;
and outputting second prediction probabilities for a plurality of known categories according to the second image features and the boundary points through the boundary point classifier.
In some embodiments, the processing unit is specifically configured to:
calculating, by the prediction module, a first distance between the first image feature and the second image feature;
and determining a first prediction probability of the product to be detected belonging to the unknown class according to the first distance.
In some embodiments, the processing unit is specifically configured to:
calculating a second distance between the second image feature and the boundary point corresponding to each known category through the boundary point classifier;
And determining a second prediction probability of the product to be detected belonging to the corresponding known category according to the second distance.
In some embodiments, the determining unit is specifically configured to:
when the first prediction probability is greater than or equal to a preset threshold value, the unknown class is used as a target class corresponding to the product to be detected;
and when the first prediction probability is smaller than the preset threshold value, taking the known category of the boundary point corresponding to the second prediction probability with the largest value as the target category corresponding to the product to be detected.
In some embodiments, the training unit is further configured to:
constructing the dual-branch network and a typical sample classifier, and constructing at least one initialized boundary point corresponding to each known category by utilizing the boundary point classifier;
acquiring a training sample image set and the known category corresponding to each training sample image in the training sample image set;
training the dual-branch network according to the training sample image set, the known class and the typical sample classifier, and freezing model parameters of the teacher network in the training process.
In some embodiments, the training unit is specifically configured to:
extracting image features of the training sample images through the teacher network and the student network respectively to obtain first sample image features corresponding to the teacher network and second sample image features corresponding to the student network;
determining a first loss value corresponding to the first branch network according to the first sample image characteristic and the second sample image characteristic;
determining a second loss value corresponding to the boundary point classifier according to the second sample image characteristic, the boundary point and the known category;
determining a third loss value corresponding to the typical sample classifier according to the second sample image features and the known category;
and reversely adjusting model parameters of the student network and the boundary point classifier in the dual-branch network according to the first loss value, the second loss value and the third loss value so as to train the dual-branch network.
In a third aspect, the present application provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the dual-branch network based product detection method of any of the above.
In a fourth aspect, the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor executes the program to implement the product detection method based on the dual-branch network according to any one of the above.
According to the product detection method, the device, the storage medium and the electronic equipment based on the dual-branch network, through obtaining the product image of the product to be detected and the trained dual-branch network, the dual-branch network comprises a first branch network and a second branch network, the first branch network comprises a teacher network, the second branch network comprises a student network and a boundary point classifier, the boundary point classifier comprises at least one parameterized boundary point corresponding to each known category in a plurality of known categories, and the teacher network and the student network are used for extracting image features; inputting the product image into a dual-branch network for processing to obtain a first prediction probability output by a first branch network for an unknown class and a second prediction probability output by a second branch network for a known class; according to the first prediction probability and the second prediction probability, the target class corresponding to the product to be detected is determined from the unknown class and the known class so as to carry out industrial detection on the product to be detected, thereby not only realizing the accurate classification of the industrial products of various known classes, but also effectively distinguishing the known class and the unknown class, greatly improving the industrial detection capability, reducing the omission rate and the false detection rate in the industrial detection, being applicable to various complicated and diverse industrial detection scenes, and having wide application range and high flexibility.
Drawings
The foregoing and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, wherein:
fig. 1 is a schematic flow chart of a product detection method based on a dual-branch network according to an embodiment of the present application;
FIG. 2 is a schematic diagram of a display process for industrial detection using a dual-branch network according to an embodiment of the present application;
FIG. 3 is another flow chart of a method for detecting products based on a dual-branch network according to an embodiment of the present application;
fig. 4 is a schematic diagram of a training flow of a dual-branch network according to an embodiment of the present application;
fig. 5 is a schematic structural diagram of a product detection device based on a dual-branch network according to an embodiment of the present application;
fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present application;
fig. 7 is a schematic hardware structure of an electronic device according to an embodiment of the present application.
Detailed Description
Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary only for the purpose of explaining the present application and are not to be construed as limiting the present application.
The embodiment of the application provides a product detection method and device based on a dual-branch network, a storage medium and electronic equipment.
Referring to fig. 1, fig. 1 is a flowchart of a product detection method based on a dual-branch network according to an embodiment of the present application. The product detection method based on the dual-branch network is applied to electronic equipment, wherein the electronic equipment comprises a mobile phone, a tablet personal computer, a personal computer (personal computer, PC) and the like. Specifically, the product detection method based on the dual-branch network comprises the following steps 101-103, wherein:
101. the method comprises the steps of obtaining a product image of a product to be detected, and a trained dual-branch network, wherein the dual-branch network comprises a first branch network and a second branch network, the first branch network comprises a teacher network, the second branch network comprises a student network and a boundary point classifier, the boundary point classifier comprises at least one parameterized boundary point corresponding to each known category in a plurality of known categories, and the teacher network and the student network are used for extracting image features.
The product image may be an image shot for a product to be detected on an industrial production line. The double-branch network is trained through a large number of sample images in advance, and not only can accurately identify which known category a product of a known category specifically belongs to, but also can detect a product of an unknown category, namely, the known category is distinguished from the unknown category. Referring to fig. 2, fig. 2 is a schematic diagram of a display process of industrial detection using a dual-branch network according to an embodiment of the present application. In the industrial detection process, the first branch network and the second branch network both can carry out classification decision (namely output respective prediction probabilities) on the product image, wherein the first branch network can predict whether a target object belongs to an unknown class or a known class, the second branch network can predict which known class the target object specifically belongs to, and the known classes are mainly some classes marked by training data during model training.
In general, the teacher network and the student network are backbone networks of a dual-branch network, which adopt the same network structure, and the network structure can flexibly select proper scale and size according to the scale and complexity of training data. The boundary point classifier is used for leading the classification decision of the second branch network, and in the image feature space, boundary points of a plurality of categories are learned, which is different from the conventional classifier for mapping the image feature space to the linear classification space, and the boundary point classifier is directly used for fitting the classification decision in the image feature space.
It should be noted that, the teacher network and the student network can be pre-trained by using big data in advance, and the pre-trained learning of the big data enables the teacher network and the student network to have strong general image feature extraction capability, and the extracted general image features have no distribution bias and are suitable for the image feature extraction of the unified feature space of the switching set. Meanwhile, in the training process of the dual-branch network aiming at sample data in the industrial detection field, the teacher network freezes own model parameters, the weight parameters are not updated according to the reverse propagation of training errors, the student network follows and updates own model parameters in the training process of the dual-branch network, namely, the student network only learns to fit the teacher network in a closed-set image feature space, the open-set image features are not subjected to fitting learning, the output of the student network and the teacher network has differences, the open-closed-set category is decided by utilizing the degree of the output differences, the differences are greatly predicted to be unknown categories, and the differences are little predicted to be known categories.
102. The product image is input into the dual-branch network for processing to obtain a first prediction probability output by the first branch network for an unknown class and a second prediction probability output by the second branch network for the known class.
The first prediction probability can be considered as a difference value between output image features of a teacher student network, the teacher and the student network (a main network) are used for leading decision classification of an open-close set, the teacher network extracts unbiased image features of the open-close set, the student network only fits and learns the feature extraction of the teacher network in the image feature space of the closed-set, and the difference classification of the open-close set detection images between the output image features of the teacher student network is utilized.
In some embodiments, please refer to fig. 3 and the above-mentioned fig. 2, fig. 3 is another flow chart of a product detection method based on a dual-branch network provided in the embodiment of the present application, where the first-branch network further includes a prediction module, and the above-mentioned step 102 may specifically include the following steps 1021 and 1022, where:
1021. extracting image features of the product image through the teacher network and the student network respectively to obtain a first image feature corresponding to the teacher network and a second image feature corresponding to the student network;
1022. And outputting, by a prediction module, a first prediction probability for an unknown class according to the first image feature and the second image feature, and outputting, by the boundary point classifier, a second prediction probability for a plurality of known classes according to the second image feature and the boundary point.
Further, the step of outputting, by the prediction module, a first prediction probability for an unknown class according to the first image feature and the second image feature may specifically include:
calculating, by the prediction module, a first distance between the first image feature and the second image feature;
and determining a first prediction probability of the product to be detected belonging to the unknown class according to the first distance.
The first prediction probability can be regarded as the distance between the first image feature and the second image feature, and the larger the distance is, the larger the value of the first prediction probability is, and the more likely the product to be detected belongs to the unknown class; the smaller the distance is, the smaller the value of the second prediction probability is, and the more likely the product to be detected belongs to the known category, so that the accurate classification of the unknown category and the known category is realized.
Further, the step of outputting, by the boundary point classifier, the second prediction probability for the known class according to the second image feature and the boundary point may specifically include:
Calculating a second distance between the second image feature and the boundary point corresponding to each known category through the boundary point classifier;
and determining a second prediction probability of the product to be detected belonging to the corresponding known category according to the second distance.
The second prediction probability may be regarded as a distance between the second image feature and the boundary point of each known class, where the larger the distance is, the larger the value of the second prediction probability is, and at this time, the more likely the product to be detected belongs to the corresponding known class, the smaller the distance is, and the smaller the value of the second prediction probability is, the less likely the product to be detected belongs to the known class, so that the boundary point classifier may also assist in predicting the known class and the unknown class.
It is easy to understand that the dual-branch network needs to be trained by a large amount of data in the industrial detection field in advance, that is, the product detection method further comprises the following steps 104-106, wherein:
104. the above-mentioned dual-branch network and the typical sample classifier are constructed, and at least one initialized boundary point corresponding to each known class is constructed using the boundary point classifier.
The typical sample classifier is used as a compound classifier, the feature extraction operation of the image is restricted only in the model training process, the stability of boundary point learning is enhanced, the processing is not participated in the industrial detection link, and the classification decision is not made. Referring to fig. 4, fig. 4 is a schematic diagram of a training flow of a dual-branch network provided in an embodiment of the present application, where decision boundaries of an open-close set feature space of an image feature space are directly modeled by a boundary point classifier, and uncertainty of the open-set feature space is constrained. The closed-set feature space may be considered as a feature space obtained by learning a training sample image set (known class), which is furthest from the boundary points of the corresponding known class, and the open-set feature space is a feature space of an unknown class, which is closer to each boundary point, and the open-set feature space and the closed-set feature space are complementary.
A plurality of corresponding boundary points can be constructed for each known class by adopting random initialization parameters to simulate an unknown open set feature space, wherein the boundary points are vectors with the size of c×n×m, c is the number of the known classes, n is the number of the boundary points corresponding to each known class, and m is the parameter size (equal to the image feature size extracted by a student network) of each boundary point.
105. A training sample image set and a known category corresponding to each training sample image in the training sample image set are obtained.
The training sample image is mainly a product image in the industrial detection field, different detection scenes can be provided with different product images and categories, for example, if the detection scenes are product quality inspection, the training sample image can comprise product images with various defect types and product images without defects, and the categories can be simply set to be two types of defects, namely, defects with defects and defects without defects, and can also limit types of defects.
In other embodiments, the detection scene may also be product sorting, where the training sample image may include multiple types of product images, for example, for a screw product, the training sample image may include multiple sets of screw product images with different types, for example, a type a screw, a type B screw, a type C screw, and the like, and the types are corresponding types, that is, a type a, B type, type C, and the like.
106. The dual-branch network is trained based on the training sample image set, the known class, and the representative sample classifier, and model parameters of the teacher network are frozen during training.
The model training process, that is, the process of continuously and iteratively updating the model parameters, changes the model parameters of the student network and the boundary point classifier by learning the training sample image, and changes (parameterizes) the boundary point along with the changes of the model parameters of the boundary point classifier. The student network learning fits the diversity distribution feature extraction of the teacher network, only judges whether the training sample image is a learned training sample image (closed set image sample), and the learning sample is an open set image sample; the student network does not learn the category attribute of diversity, does not need to train the student network through the category label, avoids the decision of a single model from the theory itself to detect the closed-set category attribute of the image, has the capability of identifying the unknown category image sample, and greatly avoids identifying the same category closed-set image with large intra-category distribution variation as the unknown category image sample.
In the training process of the double-branch network, the decision boundary of the image feature open-close set feature space is directly modeled through a boundary point classifier, the uncertainty of the open-set feature space is restrained, the image feature is restrained in a certain range through the assistance of a typical sample classifier, the stability of the boundary point fitting classification decision boundary is improved, meanwhile, the teacher network freezes model parameters in the training process, and the student network carries out self model parameter updating along with training, namely, the student network classifies learning of two languages, namely, the language I is that the image feature is extracted in the closed-set image feature space through loss value (loss) fitting, and the extracted image feature generates closed-set image feature distribution bias; the second language is a closed set image feature extraction mode of the teacher network. The double-branch network obtained through training in the mode can accurately identify the detection targets of the known type, greatly improve the identification capability of the double-branch network on the targets of the unknown type, and distinguish the detection images with very high similarity with the training sample images from the known type even if the detection images exist in the follow-up industrial detection, so that the detection omission and false detection are avoided, and the model classification accuracy is high.
In some embodiments, the step 106 may specifically include:
extracting image features of the training sample image through the teacher network and the student network respectively to obtain first sample image features corresponding to the teacher network and second sample image features corresponding to the student network;
determining a first loss value corresponding to the first branch network according to the first sample image characteristic and the second sample image characteristic;
determining a second loss value corresponding to the boundary point classifier according to the second sample image feature, the boundary point and the known class;
determining a third loss value corresponding to the typical sample classifier according to the second sample image feature and the known class;
and reversely adjusting model parameters of the student network and the boundary point classifier in the dual-branch network according to the first loss value, the second loss value and the third loss value so as to train the dual-branch network.
In the model training process, the teacher network and the student network are used as a backbone network of the dual-branch network and are used for extracting image features, the boundary point classifier is used for parameterizing the boundary point according to the image features extracted by the student network, and the typical sample classifier is used for restricting the extraction of the image features in the parameterization process of the boundary point.
In general, the teacher network and the student network output sample image features of the same size, for example, the teacher network outputs a vector of size t=b×n, and the student network outputs a vector of size s=b×n; where b is the number of training image samples and n is the size of each sample image feature. When the backbone network is trained, the training data does not need to be labeled with categories, the learning of category diversity attribute is not performed, and the training sample images with the same category diversity are prevented from being identified as unknown categories. In a closed set sample space (i.e. a feature space learned according to a training sample image set), student network learning fits a teacher network, and for a closed set detection image (i.e. a detection image belonging to a known class), the difference between two vectors output by the teacher network and the student network is smaller; for an unweared open set image (namely, a detection image belonging to an unknown class), the difference of two vectors output by a teacher network and a student network is large, and an open-close set detection result is decided by outputting the difference.
With continued reference to fig. 4, the Loss function of the backbone network (i.e., teacher network and student network) may be expressed as Loss ST ＝||||s||-||t|||| 2 ，Loss ST And s is the second sample image characteristic extracted by the student network, and t is the first sample image characteristic extracted by the teacher network. The distance between the two sample image features may be calculated as the first loss value after the two sample image features have been normalized. The better the student network fitting learning teacher network is, the first The smaller the distance between the present image feature and the second sample image feature, the less ST The smaller. In addition, the training fit of the boundary point classifier, which is used for classifying the closed training class, and the typical sample classifier, which is used for enhancing the stability of model training, also drives the learning of the student network.
In other embodiments, the teacher and student networks may output multi-scale feature vectors, where b is the number of training image samples, c is the size of the spatial feature for each image location, and h and w are the width and height of the multi-scale image features. The method is characterized in that the image features are sampled from the original input image by 16 times and 32 times, the feature images sampled from the 32 times are interpolated and sampled to 16 times, and the features of the two scales are combined to form a multi-scale feature t scale ＝b×(c 16 +c 32 ) X h x w, and characterizing the multiscale feature t scale As output of a teacher network or a student network.
The Loss function of the boundary point classifier can be expressed as the formula Loss R ＝CE(||x-p||,y)+max(||x i -p i ||-r,0)，Loss R For the second loss value described above, x is the second sample image feature, p is the boundary point, y is the known class (i.e., class label), and CE () is the cross entropy, i.e., the relative measure between the two corresponding input vectors. The better the boundary point fitting learning, the greater the distance between the second sample image feature and the boundary point of the corresponding known class should be, the smaller the error in the relative metric. max () is maximum value, x i For the image feature corresponding to the ith class label, p i And r is a parameter variable which is obtained dynamically through model training and is a boundary point corresponding to the ith class label. I x-p i is the distance between the second sample image feature and the boundary point, i x i -p i I is the distance between the second sample image feature to the boundary point corresponding to the respective known class. Loss (Low Density) R The second term of the formula is mainly used for limiting the distance to the range of r, and although the open set feature space is uncertain, the open set feature space and the closed set feature space are complementary, so that the distance can be limited to the range of r from the limiting boundary point to the closed set feature spaceThe range is used for realizing the constraint of uncertainty of the open-set feature space, and the robustness of the constraint of uncertainty of the open-set feature space is improved.
The Loss function of the exemplary sample classifier may be a cross entropy function for calculating the distance from the second sample image feature to the exemplary image feature and the cross entropy between the known classes as the third Loss value Loss C . The exemplary sample classifier samples a small portion of data from the original training sample image set as an exemplary sample, where the data can represent the features of the original training sample image set as much as possible, for example, assuming that n is the number of known classes, each of the known classes corresponds to m training sample images, and for each of the known classes, the exemplary sample classifier may sample the sample image features of k training sample images (i.e., exemplary samples) from the sample image features of m training sample images through coreset (core set), as exemplary image features, to characterize the overall feature spatial distribution of the known classes, where k and m may be determined according to requirements.
Whole dual-branch network training fitting's Loss compounds teacher student network Loss ST Boundary point classifier Loss R And a typical sample classifier Loss C . The mathematical formula for Loss is expressed as: loss=loss ST +μLoss R +λLoss C . Wherein, teacher student network Loss ST Calculating the similarity between the output image features of the dual-branch network, and the more similar the Loss ST The smaller; boundary point classifier Loss R Calculating the distance from the image feature to the boundary point, wherein the larger the distance is, the less R The smaller; typical sample classifier Loss C The distance from the image feature to the representative image feature is calculated, with the smaller the distance less the Loss. μ and λ are Loss complex weights with default values of 1.0 and 0.5, respectively.
103. And determining a target category corresponding to the product to be detected from the unknown category and the known category according to the first prediction probability and the second prediction probability so as to carry out industrial detection on the product to be detected.
For example, for defect detection, products with known defects can be detected, products with unknown defects and high similarity can be accurately distinguished from the known defects, the situation that products with unknown defects are forcedly identified as the known defects in the traditional classification model is avoided, the probability of false detection and missing detection of the model is effectively reduced, and the model classification accuracy is improved.
In some embodiments, referring to fig. 3, the step 103 may specifically include:
when the first prediction probability is greater than or equal to a preset threshold value, the unknown class is used as a target class corresponding to the product to be detected;
and when the first prediction probability is smaller than the preset threshold, taking the known category of the boundary point corresponding to the second prediction probability with the largest value as the target category corresponding to the product to be detected.
The preset threshold may be set manually, for example, 0.75, and once the first prediction probability is greater than or equal to the preset threshold, the product to be detected may be considered to belong to an unknown class, or else, to belong to a known class. Since the first branch network is mainly used for predicting whether the product to be detected belongs to a known class or an unknown class, and the second branch network is mainly used for predicting which known class the product to be detected belongs to, the values of the predicted results of the two branch networks are usually opposite, that is, if the first predicted probability value is larger, the second predicted probability value should be smaller, for example, if the first predicted probability output through the first branch network is 0.9, the product to be detected can be considered to belong to the unknown class, and at this time, the second predicted probability value is usually smaller, that is, the probability that the product to be detected belongs to any one of the known classes is smaller. If the first prediction probability output by the first branch network is 0.3, the probability that the product to be detected belongs to the unknown class is considered to be smaller, and the probability that the product to be detected belongs to the known class is considered to be larger, and if the probability value of the product to be detected belongs to the class C in the second prediction probability of each known class output by the second branch network is largest, for example, 0.85, the product to be detected can be considered to belong to the known class C.
As can be seen from the foregoing, in the product detection method based on the dual-branch network provided in the embodiments of the present application, by acquiring a product image of a product to be detected and a trained dual-branch network, the dual-branch network includes a first branch network and a second branch network, the first branch network includes a teacher network, the second branch network includes a student network and a boundary point classifier, the boundary point classifier includes at least one parameterized boundary point corresponding to each of a plurality of known categories, and the teacher network and the student network are used for extracting image features; inputting the product image into a dual-branch network for processing to obtain a first prediction probability output by a first branch network for an unknown class and a second prediction probability output by a second branch network for a known class; according to the first prediction probability and the second prediction probability, the target class corresponding to the product to be detected is determined from the unknown class and the known class so as to carry out industrial detection on the product to be detected, thereby not only realizing the accurate classification of the industrial products of various known classes, but also effectively distinguishing the known class and the unknown class, greatly improving the industrial detection capability, reducing the omission rate and the false detection rate in the industrial detection, being applicable to various complicated and diverse industrial detection scenes, and having wide application range and high flexibility.
According to the method described in the above embodiment, the embodiment of the present application further provides a product detection device based on a dual-branch network, which is configured to perform the steps in the product detection method based on the dual-branch network. Referring to fig. 5, fig. 5 is a schematic structural diagram of a product detection device based on a dual-branch network according to an embodiment of the present application. The product detection device 200 based on the dual-branch network is applied to an electronic device, and the image processing device 200 comprises an acquisition unit 201, a processing unit 202 and a determining unit 203, wherein:
an acquiring unit 201, configured to acquire a product image of a product to be detected, and a trained dual-branch network, where the dual-branch network includes a first branch network and a second branch network, the first branch network includes a teacher network, and the second branch network includes a student network and a boundary point classifier, where the boundary point classifier includes at least one parameterized boundary point corresponding to each of a plurality of known categories, and the teacher network and the student network are configured to extract image features;
a processing unit 202, configured to input the product image into the dual-branch network for processing, so as to obtain a first prediction probability output by the first branch network for an unknown class, and a second prediction probability output by the second branch network for the known class;
The determining unit 203 is configured to determine, according to the first prediction probability and the second prediction probability, a target category corresponding to the product to be detected from the unknown category and the known category, so as to perform industrial detection on the product to be detected.
In some embodiments, the first branch network further includes a prediction module, and the processing unit 202 is specifically configured to:
extracting image features of the product image through the teacher network and the student network respectively to obtain a first image feature corresponding to the teacher network and a second image feature corresponding to the student network;
outputting, by the prediction module, a first prediction probability for an unknown class according to the first image feature and the second image feature;
and outputting, by the boundary point classifier, second prediction probabilities for a plurality of known classes based on the second image feature and the boundary point.
In some embodiments, the processing unit 202 is specifically configured to:
calculating, by the prediction module, a first distance between the first image feature and the second image feature;
and determining a first prediction probability of the product to be detected belonging to the unknown class according to the first distance.
In some embodiments, the processing unit 202 is specifically configured to:
Calculating a second distance between the second image feature and the boundary point corresponding to each known category through the boundary point classifier;
and determining a second prediction probability of the product to be detected belonging to the corresponding known category according to the second distance.
In some embodiments, the determining unit 203 is specifically configured to:
when the first prediction probability is greater than or equal to a preset threshold value, the unknown class is used as a target class corresponding to the product to be detected;
and when the first prediction probability is smaller than the preset threshold, taking the known category of the boundary point corresponding to the second prediction probability with the largest value as the target category corresponding to the product to be detected.
In some embodiments, the product detection apparatus 200 of the dual-branch network further comprises a training unit for:
constructing the dual-branch network and a typical sample classifier, and constructing at least one initialized boundary point corresponding to each known category by utilizing the boundary point classifier;
acquiring a training sample image set and known categories corresponding to each training sample image in the training sample image set;
the dual-branch network is trained based on the training sample image set, the known class, and the representative sample classifier, and model parameters of the teacher network are frozen during training.
In some embodiments, the training unit is specifically configured to:
extracting image features of the training sample image through the teacher network and the student network respectively to obtain first sample image features corresponding to the teacher network and second sample image features corresponding to the student network;
determining a first loss value corresponding to the first branch network according to the first sample image characteristic and the second sample image characteristic;
determining a second loss value corresponding to the boundary point classifier according to the second sample image feature, the boundary point and the known class;
determining a third loss value corresponding to the typical sample classifier according to the second sample image feature and the known class;
and reversely adjusting model parameters of the student network and the boundary point classifier in the dual-branch network according to the first loss value, the second loss value and the third loss value so as to train the dual-branch network.
It should be noted that, the specific details of each module unit in the dual-branch network-based product detection apparatus 200 are described in detail in the embodiment of the dual-branch network-based product detection method, which is not described herein.
In some embodiments, the product detection apparatus based on the dual-branch network in the embodiments of the present application may be an electronic device, or may be a component in the electronic device, such as an integrated circuit or a chip. The electronic device may be a terminal device. By way of example, the electronic device may be a mobile phone, tablet computer, notebook computer, palm computer, vehicle-mounted electronic device, mobile internet appliance (Mobile Internet Device, MID), augmented reality (augmented reality, AR)/Virtual Reality (VR) device, robot, wearable device, ultra-mobile personal computer, UMPC, netbook or personal digital assistant (personal digital assistant, PDA), etc., but may also be a server, network attached storage (Network Attached Storage, NAS), personal computer (personal computer, PC), television (TV), teller machine or self-service machine, etc., and the embodiments of the present application are not limited in particular.
In some embodiments, as shown in fig. 6, the embodiment of the present application further provides an electronic device 300, including a processor 301, a memory 302, and a computer program stored in the memory 302 and capable of running on the processor 301, where the program when executed by the processor 301 implements the processes of the above-mentioned dual-branch network-based product detection method embodiment, and the same technical effects can be achieved, so that repetition is avoided and redundant description is omitted herein.
The electronic device in the embodiment of the application includes the mobile electronic device and the non-mobile electronic device described above.
Fig. 7 is a schematic hardware structure of an electronic device implementing an embodiment of the present application.
The electronic device 400 includes, but is not limited to: radio frequency unit 401, network module 402, audio output unit 403, input unit 404, sensor 405, display unit 406, user input unit 407, interface unit 408, memory 409, and processor 410.
Those skilled in the art will appreciate that the electronic device 400 may also include a power source (e.g., a battery) for powering the various components, which may be logically connected to the processor 410 by a power management system to perform functions such as managing charge, discharge, and power consumption by the power management system. The electronic device structure shown in fig. 7 does not constitute a limitation of the electronic device, and the electronic device may include more or less components than shown, or may combine certain components, or may be arranged in different components, which are not described in detail herein.
It should be appreciated that in embodiments of the present application, the input unit 404 may include a graphics processor (Graphics Processing Unit, GPU) 4041 and a microphone 4042, with the graphics processor 4041 processing image data of still pictures or video obtained by an image capture device (e.g., a camera) in a video capture mode or an image capture mode. The display unit 406 may include a display panel 4061, and the display panel 4061 may be configured in the form of a liquid crystal display, an organic light emitting diode, or the like. The user input unit 407 includes at least one of a touch panel 4071 and other input devices 4072. The touch panel 4071 is also referred to as a touch screen. The touch panel 4071 may include two parts, a touch detection device and a touch controller. Other input devices 4072 may include, but are not limited to, a physical keyboard, function keys (e.g., volume control keys, switch keys, etc.), a trackball, a mouse, a joystick, and so forth, which are not described in detail herein.
Memory 409 may be used to store software programs as well as various data. The memory 409 may mainly include a first memory area storing programs or instructions and a second memory area storing data, wherein the first memory area may store an operating system, application programs or instructions (such as a sound playing function, an image playing function, etc.) required for at least one function, and the like. Further, the memory 409 may include volatile memory or nonvolatile memory, or the memory 409 may include both volatile and nonvolatile memory. The nonvolatile Memory may be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically Erasable EPROM (EEPROM), or a flash Memory. The volatile memory may be random access memory (Random Access Memory, RAM), static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (ddr SDRAM), enhanced SDRAM (Enhanced SDRAM), synchronous DRAM (SLDRAM), and Direct RAM (DRRAM). Memory 409 in embodiments of the present application includes, but is not limited to, these and any other suitable types of memory.
Processor 410 may include one or more processing units; the processor 410 integrates an application processor that primarily processes operations involving an operating system, user interface, application programs, etc., and a modem processor that primarily processes wireless communication signals, such as a baseband processor. It will be appreciated that the modem processor described above may not be integrated into the processor 410.
The embodiment of the present application further provides a non-transitory computer readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements each process of the above-mentioned product detection method embodiment based on the dual-branch network, and can achieve the same technical effect, so that repetition is avoided, and details are not repeated here.
Wherein the processor is a processor in the electronic device described in the above embodiment. The readable storage medium includes computer readable storage medium such as computer readable memory ROM, random access memory RAM, magnetic or optical disk, etc.
The embodiment of the application also provides a computer program product, which comprises a computer program, and the computer program realizes the product detection method based on the dual-branch network when being executed by a processor.
Wherein the processor is a processor in the electronic device described in the above embodiment. The readable storage medium includes computer readable storage medium such as computer readable memory ROM, random access memory RAM, magnetic or optical disk, etc.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element. Furthermore, it should be noted that the scope of the methods and apparatus in the embodiments of the present application is not limited to performing the functions in the order shown or discussed, but may also include performing the functions in a substantially simultaneous manner or in an opposite order depending on the functions involved, e.g., the described methods may be performed in an order different from that described, and various steps may also be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solutions of the present application may be embodied essentially or in a part contributing to the prior art in the form of a computer software product stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk), comprising several instructions for causing a terminal (which may be a mobile phone, a computer, a server, or a network device, etc.) to perform the methods described in the embodiments of the present application.
The embodiments of the present application have been described above with reference to the accompanying drawings, but the present application is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and many forms may be made by those of ordinary skill in the art without departing from the spirit of the present application and the scope of the claims, which are also within the protection of the present application.
The terms first, second and the like in the description and in the claims, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged, as appropriate, such that embodiments of the present application may be implemented in sequences other than those illustrated or described herein, and that the objects identified by "first," "second," etc. are generally of a type and not limited to the number of objects, e.g., the first object may be one or more. Furthermore, in the description and claims, "and/or" means at least one of the connected objects, and the character "/", generally means that the associated object is an "or" relationship.
In the description of the present application, the meaning of "plurality" is two or more.
In the description of the present specification, reference to the terms "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present application. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that: many changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the application, the scope of which is defined by the claims and their equivalents.
Claims (10)
1. A method for detecting a product based on a dual-branch network, comprising:
acquiring a product image of a product to be detected and a trained dual-branch network, wherein the dual-branch network comprises a first branch network and a second branch network, the first branch network comprises a teacher network, the second branch network comprises a student network and a boundary point classifier, the boundary point classifier comprises at least one parameterized boundary point corresponding to each known category in a plurality of known categories, and the teacher network and the student network are used for extracting image features;
inputting the product image into the dual-branch network for processing to obtain a first prediction probability output by the first branch network for an unknown class and a second prediction probability output by the second branch network for the known class;
And determining a target category corresponding to the product to be detected from the unknown category and the known category according to the first prediction probability and the second prediction probability so as to carry out industrial detection on the product to be detected.
2. The dual-branch network-based product detection method as set forth in claim 1, wherein the first branch network further comprises a prediction module, and wherein the inputting the product image into the dual-branch network for processing comprises:
extracting image features of the product image through the teacher network and the student network respectively to obtain a first image feature corresponding to the teacher network and a second image feature corresponding to the student network;
outputting, by the prediction module, a first prediction probability for an unknown class according to the first image feature and the second image feature;
and outputting second prediction probabilities for a plurality of known categories according to the second image features and the boundary points through the boundary point classifier.
3. The dual branch network based product detection method as claimed in claim 2, wherein said outputting, by said prediction module, a first prediction probability for an unknown class based on said first image feature and said second image feature comprises:
Calculating, by the prediction module, a first distance between the first image feature and the second image feature;
and determining a first prediction probability of the product to be detected belonging to the unknown class according to the first distance.
4. The dual branch network based product detection method as claimed in claim 2, wherein said outputting, by said boundary point classifier, a second prediction probability for said known class based on said second image feature and said boundary point, comprises:
calculating a second distance between the second image feature and the boundary point corresponding to each known category through the boundary point classifier;
and determining a second prediction probability of the product to be detected belonging to the corresponding known category according to the second distance.
5. The dual branch network-based product detection method according to claim 1, wherein determining the target category corresponding to the product to be detected from the unknown category and the known category according to the first prediction probability and the second prediction probability comprises:
when the first prediction probability is greater than or equal to a preset threshold value, the unknown class is used as a target class corresponding to the product to be detected;
And when the first prediction probability is smaller than the preset threshold value, taking the known category of the boundary point corresponding to the second prediction probability with the largest value as the target category corresponding to the product to be detected.
6. The dual branch network-based product detection method according to any one of claims 1-5, further comprising:
constructing the dual-branch network and a typical sample classifier, and constructing at least one initialized boundary point corresponding to each known category by utilizing the boundary point classifier;
acquiring a training sample image set and the known category corresponding to each training sample image in the training sample image set;
training the dual-branch network according to the training sample image set, the known class and the typical sample classifier, and freezing model parameters of the teacher network in the training process.
7. The method of claim 6, wherein training the dual-branch network based on the training sample image set, the known class, and the representative sample classifier comprises:
Extracting image features of the training sample images through the teacher network and the student network respectively to obtain first sample image features corresponding to the teacher network and second sample image features corresponding to the student network;
determining a first loss value corresponding to the first branch network according to the first sample image characteristic and the second sample image characteristic;
determining a second loss value corresponding to the boundary point classifier according to the second sample image characteristic, the boundary point and the known category;
determining a third loss value corresponding to the typical sample classifier according to the second sample image features and the known category;
and reversely adjusting model parameters of the student network and the boundary point classifier in the dual-branch network according to the first loss value, the second loss value and the third loss value so as to train the dual-branch network.
8. A dual-branch network-based product detection apparatus, comprising:
an acquisition unit, configured to acquire a product image of a product to be detected, and a trained dual-branch network, where the dual-branch network includes a first branch network and a second branch network, the first branch network includes a teacher network, the second branch network includes a student network and a boundary point classifier, the boundary point classifier includes at least one parameterized boundary point corresponding to each of a plurality of known categories, and the teacher network and the student network are configured to extract image features;
The processing unit is used for inputting the product image into the dual-branch network for processing so as to obtain a first prediction probability output by the first branch network for an unknown class and a second prediction probability output by the second branch network for the known class;
and the determining unit is used for determining a target category corresponding to the product to be detected from the unknown category and the known category according to the first prediction probability and the second prediction probability so as to carry out industrial detection on the product to be detected.
9. A non-transitory computer readable storage medium, having stored thereon a computer program, which when executed by a processor implements the dual-branch network based product detection method according to any of claims 1-7.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the dual-branch network based product detection method of any of claims 1-7 when the program is executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311479248.6A CN117541858A (en) | 2023-11-08 | 2023-11-08 | Product detection method and device based on dual-branch network, storage medium and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311479248.6A CN117541858A (en) | 2023-11-08 | 2023-11-08 | Product detection method and device based on dual-branch network, storage medium and electronic equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117541858A true CN117541858A (en) | 2024-02-09 |
Family
ID=89785388
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311479248.6A Pending CN117541858A (en) | 2023-11-08 | 2023-11-08 | Product detection method and device based on dual-branch network, storage medium and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117541858A (en) |
-
2023
- 2023-11-08 CN CN202311479248.6A patent/CN117541858A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112488073A (en) | Target detection method, system, device and storage medium | |
| CN109544592B (en) | Moving object detection algorithm for camera movement | |
| Cuevas et al. | Improved background modeling for real-time spatio-temporal non-parametric moving object detection strategies | |
| US11170512B2 (en) | Image processing apparatus and method, and image processing system | |
| CN112036381B (en) | Visual tracking method, video monitoring method and terminal equipment | |
| CN111368634B (en) | Human head detection method, system and storage medium based on neural network | |
| CN112651467B (en) | Training method and system and prediction method and system for convolutional neural network | |
| CN112818888A (en) | Video auditing model training method, video auditing method and related device | |
| CN117726862A (en) | Model training method, device and storage medium applied to industrial detection | |
| CN118115927B (en) | Target tracking method, apparatus, computer device, storage medium and program product | |
| CN117114648A (en) | Production line equipment fault prediction method and device, electronic equipment and storage medium | |
| CN113395584B (en) | Video data processing method, device, equipment and medium | |
| US20200311401A1 (en) | Analyzing apparatus, control method, and program | |
| CA3170297A1 (en) | Generating performance predictions with uncertainty intervals | |
| CN112580584A (en) | Method, device and system for detecting standing behavior and storage medium | |
| Ni et al. | An improved kernelized correlation filter based visual tracking method | |
| CN116797973A (en) | Data mining method and system applied to sanitation intelligent management platform | |
| CN117541858A (en) | Product detection method and device based on dual-branch network, storage medium and electronic equipment | |
| JP4879257B2 (en) | Moving object tracking device, moving object tracking method, and moving object tracking program | |
| CN115393755A (en) | Visual target tracking method, device, equipment and storage medium | |
| CN114170271A (en) | Multi-target tracking method with self-tracking consciousness, equipment and storage medium | |
| Chang et al. | Improved adaptive vibe and the application for segmentation of complex background | |
| CN112434629A (en) | Online time sequence action detection method and equipment | |
| Zhou et al. | Dynamic foreground detection based on improved Codebook model | |
| Sivabalakrishnan et al. | Performance analysis of fuzzy logic-based background subtraction in dynamic environments |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication |