CN116630749A - Industrial equipment fault detection method, device, equipment and storage medium - Google Patents
Industrial equipment fault detection method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN116630749A CN116630749A CN202310721247.1A CN202310721247A CN116630749A CN 116630749 A CN116630749 A CN 116630749A CN 202310721247 A CN202310721247 A CN 202310721247A CN 116630749 A CN116630749 A CN 116630749A
- Authority
- CN
- China
- Prior art keywords
- industrial equipment
- fault
- text
- detection
- preset
- 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 143
- 238000003860 storage Methods 0.000 title claims abstract description 16
- 239000013598 vector Substances 0.000 claims abstract description 164
- 230000000007 visual effect Effects 0.000 claims abstract description 95
- 238000000034 method Methods 0.000 claims abstract description 22
- 238000012549 training Methods 0.000 claims description 78
- 238000012795 verification Methods 0.000 claims description 24
- 238000002372 labelling Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 12
- 238000005282 brightening Methods 0.000 claims description 8
- 238000011478 gradient descent method Methods 0.000 claims description 8
- 230000009467 reduction Effects 0.000 claims description 7
- 230000002708 enhancing effect Effects 0.000 claims 1
- 238000003745 diagnosis Methods 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 14
- 230000006870 function Effects 0.000 description 10
- 238000004590 computer program Methods 0.000 description 7
- 238000004891 communication Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 6
- 238000011179 visual inspection Methods 0.000 description 6
- 238000000605 extraction Methods 0.000 description 5
- 239000012212 insulator Substances 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000013507 mapping Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000013209 evaluation strategy Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 238000010200 validation analysis 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
- G06V10/77—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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- 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
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
-
- 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
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The application discloses an industrial equipment fault detection method, a device, equipment and a computer readable storage medium, wherein the method comprises the following steps: acquiring a detection image of industrial equipment; inputting the detection image into a preset visual detection model to obtain a visual information feature vector of the target to be identified; obtaining an reasoning result according to the visual information feature vector and a preset multi-mode knowledge graph reasoning; and generating fault text according to the reasoning result. The technical scheme provided by the application combines the visual detection model and the multi-mode knowledge graph, generates the interpretable fault text according to the detection image of the industrial equipment, can meet the fault diagnosis requirement of the industrial equipment in a complex scene with less data quantity, improves the application range of the fault diagnosis of the industrial equipment, and improves the fault diagnosis accuracy of the industrial equipment.
Description
Technical Field
The present application relates to the technical field of the present application, and in particular, to a method, an apparatus, a device, and a storage medium for detecting a fault of an industrial device.
Background
The original industrial equipment fault evaluation strategies can be divided into a means for detecting by using manual experience and a method for detecting by using a computer model as a main body, and the traditional detection methods can only confirm possible fault points by using a photographing mode, so that the problems of omission and the like are possibly caused, various influencing factors influencing the accuracy and the stability of fault detection are difficult to remove in a multi-angle, multi-field and multi-level manner by the traditional detection methods, and the industrial equipment fault detection is low in accuracy and narrow in adaptability.
Therefore, the existing industrial equipment fault detection method is low in detection accuracy and narrow in adaptability, and is a problem to be solved urgently.
Disclosure of Invention
The application mainly aims to provide an industrial equipment fault detection method, an industrial equipment fault detection device, industrial equipment fault detection equipment and a storage medium, and aims to solve the technical problems of low detection accuracy and narrow adaptability of the existing industrial equipment fault detection method.
In order to achieve the above object, the present application provides an industrial equipment failure detection method, including:
acquiring a detection image of industrial equipment;
inputting the detection image into a preset visual detection model to obtain a visual information feature vector of a target to be identified;
obtaining an inference result according to the visual information feature vector and a preset multi-mode knowledge graph;
and generating fault text according to the reasoning result.
Optionally, the step of obtaining the reasoning result according to the visual information feature vector and a preset multi-mode knowledge graph includes:
extracting keywords from the visual information feature vector to obtain a feature text vector;
sorting the characteristic text vectors with the confidence coefficient higher than a preset threshold value from high to low to obtain a text vector set;
And obtaining the reasoning result according to the text vector set and a preset multi-mode knowledge graph.
Optionally, the step of obtaining the reasoning result according to the text vector set and a preset multi-mode knowledge graph includes:
sequentially performing cosine similarity matching on the text vectors in the text vector set and entities, attributes and relations in a preset multi-mode knowledge graph to obtain fault vectors with highest similarity corresponding to each text vector;
obtaining a fault triplet set according to the fault vector and the semantic matching template;
and adopting path question sentences and multi-hop question sentences to carry out the graph searching and the joint-by-joint strategy reasoning on the fault triplet set to generate the reasoning result.
Optionally, the step of generating the fault text according to the reasoning result includes:
and inputting the reasoning result into a text generation model to generate a fault text.
Optionally, the step of acquiring the detection image of the industrial device comprises, before:
collecting sample images of industrial equipment, and dividing the sample images into a training set and a verification set;
carrying out data enhancement on the sample images in the training set to obtain enhanced image data;
and training the network model through the enhanced image data, and verifying the training result through the verification set to obtain a preset visual detection model.
Optionally, the step of performing data enhancement on the sample images in the training set to obtain enhanced image data includes:
labeling industrial equipment in the sample images in the training set to obtain a labeling frame of the industrial equipment;
image enlargement is carried out on the sample image and the annotation frame through an enlargement formula, and an enlarged image is obtained;
and carrying out noise reduction, sharpening and brightening on the enlarged image to obtain the enhanced image data.
Optionally, the step of training the network model through the enhanced image data and verifying the training result through the verification set to obtain a preset visual detection model includes:
setting initialization weights and parameters of a training process;
inputting the enhanced image data into a network model for model training, updating and optimizing each group of parameters in the network model by using a random gradient descent method, inputting a verification set into the trained network model for calculating mAP values of the network model after each round of model training, and updating weights of the current optimal mAP values;
after training the preset number of rounds, selecting a network model corresponding to the weight of the optimal mAP value as a preset visual detection model.
In a second aspect, the present application also provides an industrial equipment failure detection apparatus, including:
the acquisition module is used for acquiring a detection image of the industrial equipment;
the first obtaining module is used for inputting the detection image into a preset visual detection model to obtain a visual information feature vector of the target to be identified;
the second obtaining module is used for obtaining an reasoning result according to the visual information feature vector and a preset multi-mode knowledge graph;
and the generating module is used for generating a fault text according to the reasoning result.
In a third aspect, the present application also provides an industrial equipment fault detection device, including a processor, a memory, and an industrial equipment fault detection program stored on the memory and executable by the processor, where the industrial equipment fault detection program, when executed by the processor, implements the steps of the industrial equipment fault detection method as described above.
In a fourth aspect, the present application also provides a computer readable storage medium, wherein the computer readable storage medium stores an industrial equipment fault detection program, and the industrial equipment fault detection program, when executed by a processor, implements the steps of the industrial equipment fault detection method as described above.
In the technical scheme of the application, a detection image of industrial equipment is obtained; inputting the detection image into a preset visual detection model to obtain a visual information feature vector of the target to be identified; obtaining an inference result according to the visual information feature vector and a preset multi-mode knowledge graph; and generating fault text according to the reasoning result. According to the technical scheme provided by the application, the visual information feature vector is obtained based on the preset visual detection model, the visual information feature vector is mapped with the multi-mode knowledge graph to obtain the reasoning result, and the reasoning result is processed to generate the fault text.
Drawings
Fig. 1 is a schematic diagram of a hardware structure of a terminal according to an embodiment of the present application;
FIG. 2 is a schematic flow chart of a first embodiment of the industrial equipment fault detection method of the present application;
FIG. 3 is a detailed schematic diagram of the steps for obtaining a fault triplet set according to the visual information feature vector and a preset multi-modal knowledge graph;
FIG. 4 is a detailed flowchart of the step of obtaining the reasoning result according to the text vector set and a preset multi-modal knowledge graph;
FIG. 5 is a schematic flow chart diagram of a second embodiment of an industrial equipment fault detection method of the present application;
FIG. 6 is a schematic flow chart of a third embodiment of an industrial equipment fault detection method according to the present application;
FIG. 7 is a detailed flowchart of the step of obtaining enhanced image data by data enhancement of sample images in the training set according to the present application;
FIG. 8 is a detailed flowchart of the steps of training the network model by the enhanced image data and verifying the training result by the verification set to obtain a preset visual inspection model according to the present application;
FIG. 9 is a graph comparing the improved visual inspection model (HFaster) and the original network model (Faster) inspection accuracy of the present application;
FIG. 10 is a schematic diagram of the detection result of the visual detection model according to the present application;
FIG. 11 is a schematic diagram of an industrial equipment failure detection apparatus of the present application.
The achievement of the objects, functional features and advantages of the present application will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.
The industrial equipment fault detection method related to the embodiment of the application is mainly applied to a terminal, and the terminal can be a PC, a portable computer, a mobile terminal and other equipment with display and processing functions.
Referring to fig. 1, fig. 1 is a schematic diagram of a terminal structure according to an embodiment of the present application. The terminal of the embodiment of the application is industrial equipment fault detection equipment; in an embodiment of the present application, the terminal may include a processor 1001 (e.g., a CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Wherein the communication bus 1002 is used to enable connected communications between these components; the user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard); the network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface); the memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory, and the memory 1005 may alternatively be a storage device independent of the processor 1001.
Those skilled in the art will appreciate that the hardware configuration shown in fig. 1 does not constitute a limitation of the apparatus, and may include more or fewer components than shown, or may combine certain components, or may be arranged in different components.
With continued reference to FIG. 1, the memory 1005 of FIG. 1, which is a computer readable storage medium, may include an operating system, a network communication module, and an industrial equipment fault detection program.
In fig. 1, the network communication module is mainly used for connecting with a server and performing data communication with the server; and the processor 1001 may call the industrial equipment failure detection program stored in the memory 1005 and perform the following operations:
acquiring a detection image of industrial equipment;
inputting the detection image into a preset visual detection model to obtain a visual information feature vector of a target to be identified;
obtaining an inference result according to the visual information feature vector and a preset multi-mode knowledge graph;
and generating fault text according to the reasoning result.
Further, the processor 1001 may call an industrial equipment failure detection program stored in the memory 1005, and perform the following operations:
extracting keywords from the visual information feature vector to obtain a feature text vector;
sorting the characteristic text vectors with the confidence coefficient higher than a preset threshold value from high to low to obtain a text vector set;
and obtaining the reasoning result according to the text vector set and a preset multi-mode knowledge graph.
Further, the processor 1001 may call an industrial equipment failure detection program stored in the memory 1005, and perform the following operations:
sequentially performing cosine similarity matching on the text vectors in the text vector set and entities, attributes and relations in a preset multi-mode knowledge graph to obtain fault vectors with highest similarity corresponding to each text vector;
obtaining a fault triplet set according to the fault vector and the semantic matching template;
and adopting path question sentences and multi-hop question sentences to carry out the graph searching and the joint-by-joint strategy reasoning on the fault triplet set to generate the reasoning result.
Further, the processor 1001 may call an industrial equipment failure detection program stored in the memory 1005, and perform the following operations:
and inputting the reasoning result into a text generation model to generate a fault text.
Further, the processor 1001 may call an industrial equipment failure detection program stored in the memory 1005, and perform the following operations:
collecting sample images of industrial equipment, and dividing the sample images into a training set and a verification set;
carrying out data enhancement on the sample images in the training set to obtain enhanced image data;
and training the network model through the enhanced image data, and verifying the training result through the verification set to obtain a preset visual detection model.
Further, the processor 1001 may call an industrial equipment failure detection program stored in the memory 1005, and perform the following operations:
labeling industrial equipment in the sample images in the training set to obtain a labeling frame of the industrial equipment;
image enlargement is carried out on the sample image and the annotation frame through an enlargement formula, and an enlarged image is obtained;
and carrying out noise reduction, sharpening and brightening on the enlarged image to obtain the enhanced image data.
Further, the processor 1001 may call an industrial equipment failure detection program stored in the memory 1005, and perform the following operations:
setting initialization weights and parameters of a training process;
inputting the enhanced image data into a network model for model training, updating and optimizing each group of parameters in the network model by using a random gradient descent method, inputting a verification set into the trained network model for calculating mAP values of the network model after each round of model training, and updating weights of the current optimal mAP values;
after training the preset number of rounds, selecting a network model corresponding to the weight of the optimal mAP value as a preset visual detection model.
Based on the hardware structure of the terminal, various embodiments of the industrial equipment fault detection method are provided.
The application provides an industrial equipment fault detection method.
Referring to fig. 2, in a first embodiment of the present application, the industrial equipment fault detection method includes the following steps:
s100, acquiring a detection image of industrial equipment;
specifically, in one embodiment, the detection image of the industrial device may be an image captured in real time by the data acquisition device, and when the terminal receives the fault detection instruction, the terminal collects the image of the industrial device captured in real time by the data acquisition device as the detection image. In another embodiment, a database may be provided in the terminal, the image of the industrial equipment captured by the data acquisition device is stored in the database, and when the industrial equipment needs to be detected by fault, the terminal may find the corresponding image in the database as a detection image according to a detection instruction input by a user. The data acquisition device can be a device with shooting function such as a camera, a scanner, a camera and the like.
S200, inputting the detection image into a preset visual detection model to obtain a visual information feature vector of a target to be identified;
specifically, before fault detection of industrial equipment is performed, training a visual detection model is required, and the trained visual detection model is used as a preset visual detection model. Inputting the detection image of the industrial equipment into a preset visual detection model to obtain entity information under the identification scene, and extracting text vectors in the entity information to serve as visual information feature vectors of the targets to be identified.
In an embodiment, the industrial device may be a substation device, and the object to be identified may include a body object including a transformer and a background object including a respirator, a radiator, an insulator, an electric pole, black, missing, yellow, a logo, and the like. The main device for fault judgment can be determined by identifying the main object, and the part generating fault and fault phenomenon of the main device can be obtained by identifying the background object. For example, the data quantity of the fault of oil leakage of the breather in the transformer is insufficient, the data quality is low, but the characteristic of black oil leakage is very common in other scenes or other data sets, so that the detection image can be identified by utilizing a preset visual detection model, the identification result of the main object is the transformer, and the identification result of the background object is the breather, black and the like.
S300, obtaining an inference result according to the visual information feature vector and a preset multi-mode knowledge graph;
in particular, with the continuous popularization of internet technology, information from different sources such as text, images, video and audio together characterize the same or related content, present complex and multi-level semantic relationships, and form multi-modal information. In this embodiment, information such as entities, attributes, relationships and the like is stored in a preset multi-mode knowledge graph, and after the feature vectors of the visual information of the target to be identified of the detection image are obtained, the vectors are mapped and matched with the information such as the entities, the attributes, the relationships and the like in the multi-mode knowledge graph, so that an inference result is obtained.
S400, generating a fault text according to the reasoning result.
Specifically, fault text can be correspondingly generated according to the reasoning result, so that the display is convenient.
In the technical scheme of the application, a detection image of industrial equipment is obtained; inputting the detection image into a preset visual detection model to obtain a visual information feature vector of the target to be identified; obtaining an reasoning result according to the visual information feature vector and a preset multi-mode knowledge graph; and generating a fault text according to the reasoning result. According to the technical scheme provided by the application, the visual information feature vector is obtained based on the preset visual detection model, the visual information feature vector is mapped with the multi-mode knowledge graph to obtain the reasoning result, and the reasoning result is processed to generate the fault text.
Further, referring to fig. 3, fig. 3 is a flowchart illustrating a step of obtaining an inference result according to the visual information feature vector and a preset multi-modal knowledge graph in another embodiment of the present application, and S300 includes:
S310, extracting keywords from the visual information feature vector to obtain a feature text vector;
specifically, after the visual information feature vector of the object to be identified is obtained, the visual information feature vector and the feature words in the knowledge graph cannot be well matched to determine the relevance existing between the visual information feature vector and the feature words in the knowledge graph due to the diversity and complexity of the identification scene, so that the keyword extraction is required to be carried out on the visual information feature vector so as to be convenient for matching the keyword and the feature words in the knowledge graph. In an embodiment, keyword extraction is performed on the feature vector of the visual information by using a preset visual detection model to obtain a feature text vector.
S320, sorting the feature text vectors with the confidence coefficient higher than a preset threshold value from high to low according to the confidence coefficient to obtain a text vector set;
specifically, the feature vectors of the visual information identified by the visual detection model are generated randomly, after the feature vectors of the visual information are obtained by capturing keywords, the feature text vectors with the confidence coefficient higher than a preset threshold value in the feature text vectors are ranked from high to low according to the confidence coefficient, and a text vector set is obtained, namely the text vector set comprises a plurality of text vectors, and the text vectors are ranked from high to low according to the confidence coefficient. Wherein the confidence of each feature text vector can be obtained by referring to the prior art. Note that, since the amount of subject target data in the target to be identified is small, the detection accuracy obtained is not high, and therefore, the preset threshold value may be set to 30%. Thus, the running speed and accuracy can be improved by utilizing the text vector set comprising the feature vectors with high confidence.
S330, obtaining the reasoning result according to the text vector set and a preset multi-mode knowledge graph.
And sequentially placing the text vectors in the text vector set into a preset multi-mode knowledge graph according to the confidence level from high to low, and mapping the text vectors with entities, attributes and relations in the preset multi-mode knowledge graph to obtain an reasoning result. In another embodiment, the reasoning result can be obtained through reasoning by using a path question and a multi-hop question.
In this embodiment, unordered feature text vectors are ordered to form an ordered text vector set, so that the text vectors can be conveniently input into the multi-modal knowledge graph in sequence, and the efficiency and accuracy of mapping the feature text vectors and information in the knowledge graph are improved.
Further, referring to fig. 4, in still another embodiment of the present application, a flowchart detailing a step of obtaining the reasoning result according to the text vector set and a preset multi-modal knowledge-graph, S330 includes:
s331, sequentially performing cosine similarity matching on the text vectors in the text vector set and entities, attributes and relations in a preset multi-mode knowledge graph to obtain fault vectors with highest similarity corresponding to each text vector;
Specifically, in order to realize mapping between the text vectors in the text vector set and the multi-modal knowledge graph, the text vectors in the text vector set can be sequentially placed into a preset multi-modal knowledge graph according to the order of the confidence level from high to low, and cosine similarity calculation matching is performed on the text vectors and the entities, the attributes and the relations in the multi-modal knowledge graph, so that fault vectors with highest similarity corresponding to each text vector are obtained. The fault vector includes entity, relationship, attribute information. The cosine similarity calculation formula is as follows:wherein A is i And B i Representing the components of vector a and vector B, respectively.
S332, obtaining a fault triplet set according to the fault vector and the semantic matching template;
s333, adopting path question sentences and multi-jump question sentences to search according to the graph to generate the reasoning result by adopting the policy reasoning.
After the fault vector is obtained, the fault vector is subjected to semantic matching template to obtain a fault triplet set. The semantic analysis and verification of the fault vector can be carried out through semantic matching templates such as entity+attribute, entity+entity, attribute+attribute value and the like, namely, mixed relation reasoning is carried out through attribute or relation prediction, and the attribute and attribute value related to the fault vector are matched. The accuracy of the obtained triplet set can be higher through the semantic matching template.
In an embodiment, the semantic matching template may include sorting operations of the main device, the secondary device, the fault phenomenon, and the like, so that the obtained fault triplet set includes text vectors corresponding to the main device, the secondary device, the fault phenomenon, and the like. Table I is a knowledge graph matching table, and as can be seen from table I, when the input text vector is a transformer, the node attribute corresponding to the matching is a main device; when the input text vector is a radiator, the node attribute corresponding to the matching is a secondary device; when the input text vector is black, the node attribute corresponding to the matching is a fault phenomenon, and when the input text vector is an insulator, the node attribute corresponding to the matching is a secondary device; when the input text vector is missing, the node attribute corresponding to the match is a failure phenomenon.
TABLE 1 knowledge-graph matching
Further, referring to fig. 5, fig. 5 is a flow chart of a second embodiment of the present application, and S400 includes:
s410, inputting the reasoning result into a text generation model to generate a fault text.
Specifically, after the inference result is obtained, the inference result may be input into a text generation model, and an interpretable fault text may be generated. According to the embodiment, fault text generation is performed through the text generation model, and the visual detection model, the multi-mode knowledge graph and the text generation model are effectively combined. The reasoning among the real relations of the visual information feature vectors is realized through the multi-modal knowledge graph, the interpretable fault diagnosis information is generated from the reasoning result through the text generation model, and the stability of generating the final fault text by utilizing the text generation model is ensured by utilizing the richness and complementarity of the multi-modal information.
The text generation model is preferably a BART text generation model, and the BART is a noise reduction self-encoder constructed by adopting a sequence-to-sequence model and is suitable for various text generation tasks. The method uses a neural machine translation architecture based on a standard transducer, destroys the text through a noise function, and learns a sequence-to-sequence model to reconstruct the original text, so that the model can flexibly process the original input text and learn to reconstruct the text efficiently.
Taking fault detection of substation equipment as an example, the industrial equipment fault detection method of the scheme is elaborated, visual recognition is carried out on detection images of the substation equipment to obtain visual information feature vectors, keyword grabbing is carried out on the visual information feature text vectors to obtain feature text vectors, data with confidence higher than 30% in the feature text vectors are taken out, the data are ordered from high to low according to the confidence to obtain a text vector set, the text vectors in the text vector set are mapped into a multi-mode knowledge graph from high to low according to the confidence, semantic matching is carried out on the text vectors and node name texts in the multi-mode knowledge graph, and pairing operations of main equipment, secondary equipment, fault phenomena and the like are carried out according to the obtained node attributes, so that ordered structure text vectors are obtained, the structure text vectors form a fault triplet set, the reasoning result is generated according to a graph searching strategy of a path question sentence and a multi-question-skip sentence, and text result is generated by utilizing a BART text generation reasoning model to obtain a multi-mode knowledge graph, and the fault text is obtained.
Table II is a fault text generation table for substation equipment, from which it can be seen that the text vector sets are "black (99%), heat sink (59%), transformer (49%)"; the sequencing results of main equipment, secondary equipment, fault phenomena and the like are 'transformer, radiator and black' after the semantic matching template, and the triads obtained by multi-mode knowledge graph reasoning are 'transformer, radiator and oil leakage'; the fault text is "radiator oil leak on transformer".
Table II failure text generation table for substation equipment
Further, referring to fig. 6, fig. 6 is a flow chart of a third embodiment of the present application, and S100 includes:
s500, collecting sample images of industrial equipment, and dividing the sample images into a training set and a verification set;
s600, carrying out data enhancement on the sample images in the training set to obtain enhanced image data;
and S700, training the network model through the enhanced image data, and verifying the training result through the verification set to obtain a preset visual detection model.
Specifically, before identifying the detected image of the industrial equipment, training and improvement are required to be performed on the basic network model, wherein the basic network model to be trained in the embodiment is a fast-RCNN model, and the loss function is as follows: Wherein p is i To detect probability of frame target, p i * To detect the probability of frame non-target, L cls (p i ,p i * ) Is the logarithmic loss, t i To predict the offset, t i * Is the actual offset from the actual frame coordinates.
The sample image of the industrial equipment is acquired and divided into a training set and a verification set, and can be acquired by a user or can be from a stored public data set. The basic network model comprises a feature extraction module, an RPN (image region extraction) module, a RoI Pooling module and an RCNN (deep learning) module, in this embodiment, a data enhancement module may be added to the basic network model, the data enhancement module performs data enhancement on a sample image in a training set to obtain enhanced image data, and then trains the network model through the enhanced image data, and verifies a training result through the verification set to obtain a preset visual detection model. It should be noted that, the preset visual inspection model is an improved visual inspection model obtained by improving the basic network model. Before the sample images in the training set are input into the main body network model for training, the sample images are subjected to image enhancement processing, so that the data quality of the sample images can be improved, and the recognition capability of the improved visual detection model on small objects is improved.
Further, referring to fig. 7, fig. 7 is a flowchart illustrating a step of performing data enhancement on a sample image in the training set to obtain enhanced image data in an embodiment of the present application, and S600 includes:
s610, marking the industrial equipment in the sample images in the training set to obtain a marking frame of the industrial equipment;
s620, performing image enlargement on the sample image and the annotation frame through an enlargement formula to obtain an enlarged image;
and S630, denoising, sharpening and brightening the increased image to obtain enhanced image data.
In one embodiment, the data enhancement can be performed on the sample image by using the PEFT (parameter efficient fine tuning) concept, specifically, before performing model training, industrial equipment in the sample image is labeled in advance to obtain a labeling frame of the industrial equipment, the sample image and the labeling frame are subjected to image enhancement according to an enhancement formula to obtain an enhancement image, and then the enhancement image is subjected to noise reduction, sharpening and brightening to obtain enhanced image data. Wherein, the increase formula is:B W representing sample images or annotation boxes, B H Representing the length of the enlarged sample image or the label frame, B w Representing the broadness of the enlarged sample image or the label frame, B h The length of the sample image or the label frame is represented, H represents the height value of the image after the image is increased, H represents the height value of the image when the image is not increased, and H/H represents a preset multiple super-parameter.
Specifically, the sample image is scaled correspondingly according to the ratio of the shortest side length in the same patch (image block) to the preset value, if the enlarged long side is insufficient to correspond to the longest side of the image in the same patch after being enlarged according to the scale, the sample image is filled with the number 0, so that when the image is input into the feature extraction module to carry out the operation of convolutionally extracting the feature image, the enlarged data does not influence the finally generated feature image. In addition, in order to avoid the situation that the pre-marked marking frame is not matched with the amplified image in the sample image amplifying process, the pre-marked marking frame can be scaled in the same proportion, so that the target to be identified cannot exceed the range of the marking frame, and the training target cannot be lost. The sample image and the marking frame are amplified in the same proportion, and noise reduction, sharpening and brightening operations are carried out on the amplified image, so that the duty ratio range of the target to be identified is larger, the definition is higher, the obtained characteristics are more obvious, and the accuracy of fault detection of industrial equipment is improved.
Further, referring to fig. 8, fig. 8 is a flowchart illustrating a training process of the network model by the enhanced image data and verifying the training result by the verification set to obtain a preset visual detection model in the embodiment of the present application, and S700 includes:
s710, setting initialization weights and parameters of a training process;
s720, inputting the image enhancement data into a network model for model training, updating and optimizing each group of parameters in the network model by using a random gradient descent method, inputting a verification set into the trained network model for calculating mAP values of the network model after each round of model training, and updating weights of the current optimal mAP values;
and S730, after training the preset number of rounds, selecting a network model corresponding to the weight of the optimal mAP value as a preset visual detection model.
The method comprises the steps of presetting initialization weights and parameters in a basic network model, inputting image enhancement data in a training set into the basic network model for model training, updating and optimizing each group of parameters in the network model by using a random gradient descent method (SGD), wherein the random gradient descent method is a gradient optimization algorithm for updating parameters of a deep neural network, and is characterized in that in each iteration, a small batch of samples are randomly selected to calculate gradients of a loss function, and the gradients are used for updating the parameters; after each round of model training, inputting the verification set into a trained network model to calculate the mAP value of the network model, and when the calculated mAP is the optimal mAP value, storing and updating the weight corresponding to the optimal mAP value; and after training the preset number of rounds, taking a network model corresponding to the weight of the optimal mAP value as a preset visual detection model. The preset visual detection model is obtained after the basic network model is trained, so that the detection and identification accuracy of the visual detection model can be improved.
As shown in fig. 9, fig. 9 is a comparison diagram of the improved visual inspection model (HFaster) and the original network model (fast), and it can be seen from fig. 9 that the improved visual inspection model has an average rise of 1.2 percentage points for each category of object recognition accuracy. As shown in fig. 10, fig. 10 is a schematic diagram of the detection result corresponding to the improved visual detection model. After the improved visual detection model is used for detection, the detection result of the application obtains more detection frame results, and the recognition rate and the accuracy of the improved visual detection model on small targets are obviously enhanced compared with the original network model.
In order to easily understand the technical scheme provided by the embodiment of the application, the industrial equipment fault detection method provided by the embodiment of the application is briefly described by a complete industrial equipment fault detection process:
collecting sample images of industrial equipment, and dividing the sample images into a training set and a verification set;
labeling industrial equipment in the sample images in the training set to obtain a labeling frame of the industrial equipment;
image enlargement is carried out on the sample image and the annotation frame through an enlargement formula, and an enlarged image is obtained;
Noise reduction, sharpening and brightening are carried out on the enlarged image to obtain enhanced image data;
setting initialization weights and parameters of a training process;
inputting the enhanced image data into a network model for model training, updating and optimizing each group of parameters in the network model by using a random gradient descent method, inputting a verification set into the trained network model for calculating mAP values of the network model after each round of model training, and updating the weight of the current optimal mAP value;
after training the preset number of rounds, selecting a network model corresponding to the weight of the optimal mAP value as a preset visual detection model;
acquiring a detection image of industrial equipment;
inputting the detection image into a preset visual detection model to obtain a visual information feature vector of a target to be identified;
extracting keywords from the visual information feature vector to obtain a feature text vector;
sorting the characteristic text vectors with the confidence coefficient higher than a preset threshold value from high to low to obtain a text vector set;
sequentially performing cosine similarity matching on the text vectors in the text vector set and entities, attributes and relations in a preset multi-mode knowledge graph to obtain fault vectors with highest similarity corresponding to each text vector;
Reasoning the fault vector through a semantic matching template to obtain a fault triplet set;
according to the entity and attribute value in the multi-mode knowledge graph corresponding to the entity text vector obtained after the cosine similarity is matched, the entity (transformer) of the main equipment, the entity (radiator) of the secondary equipment and the attribute value (black) of the fault phenomenon in the knowledge graph are utilized to obtain a fault triplet set, and the semantic matching template is utilized to determine the corresponding fault entity (oil leakage) when the attribute value (black) of the secondary equipment (radiator) is determined by adopting the map-wise policy reasoning of the path question sentence and the multi-jump question sentence. And adopting path question sentences and multi-hop question sentences to carry out the graph searching and the joint-by-joint strategy reasoning on the fault triplet set to generate the reasoning result. Specifically, describing in detail the fault detection of substation equipment as an example, table III isA fault text generation table of substation equipment sets a first preset threshold to 30%; performing keyword grabbing on the visual feature vectors identified by the improved visual detection model to obtain feature text vectors, discarding the results with the confidence level lower than 30% in the feature text vectors, and sorting the feature text vectors with the confidence level higher than 30% from high to low to obtain a text vector set, wherein the text vector set is black (99%), missing (86%), insulator (67%), radiator (59%), and transformer (49%); sequentially placing text vectors in a text vector set into a multi-mode knowledge graph from high to low, carrying out vector cosine similarity calculation on the text vectors and information such as entities, attributes, relations and the like in the multi-mode knowledge graph, obtaining an inference result by a semantic matching template through a path question sentence and a graph-based question sentence strategy of a multi-jump question sentence, wherein the inference result is a transformer, a radiator, oil leakage, a transformer, an insulator and damage, namely a first group of triples with a rear mark of 1 in a fault triplet set is a transformer, a radiator and oil leakage, and a second group of triples with a rear mark of 2 is a transformer, a radiator and a damage; and finally generating a text of the reasoning result by utilizing the BART model to obtain an interpretable fault text, wherein the fault text is 'oil leakage of a radiator on the transformer, and breakage of an insulator on the transformer'. The text similarity calculation is carried out on the obtained interpretable fault text and the correct fault text by using the Euclidean distance formula, and the accuracy of fault detection of the scheme can reach 98%. The Euclidean distance formula is: Wherein x is 1 、x 2 、y 1 、y 2 Is the coordinate value of the vector point.
Table III failure text accuracy table of substation equipment
In addition, referring to fig. 11, the present application further provides an industrial equipment fault detection device 10, where the industrial equipment fault detection device 10 includes:
an acquisition module 20 for acquiring a detection image of the industrial equipment;
a first obtaining module 30, configured to input the detection image into a preset visual detection model to obtain a visual information feature vector of the target to be identified;
the second obtaining module 40 is configured to obtain an inference result according to the visual information feature vector and a preset multi-mode knowledge graph;
and the generating module 50 is used for generating fault text according to the reasoning result.
Further, the second obtaining module 40 is further configured to: extracting keywords from the visual information feature vector to obtain a feature text vector; sorting the characteristic text vectors with the confidence coefficient higher than a preset threshold value from high to low to obtain a text vector set; and obtaining the reasoning result according to the text vector set and a preset multi-mode knowledge graph.
Further, the second obtaining module 40 is further configured to: sequentially performing cosine similarity matching on the text vectors in the text vector set and entities, attributes and relations in a preset multi-mode knowledge graph to obtain fault vectors with highest similarity corresponding to each text vector; obtaining a fault triplet set according to the fault vector and the semantic matching template; and adopting path question sentences and multi-hop question sentences to carry out the graph searching and the joint-by-joint strategy reasoning on the fault triplet set to generate the reasoning result.
Further, the generating module 50 is further configured to: and inputting the fault triplet set into a text generation model to generate a fault text.
Further, the industrial equipment failure detection apparatus 10 includes:
an acquisition module (not shown) for acquiring a sample image of the industrial equipment, dividing the sample image into a training set and a validation set;
a data enhancement module (not shown) for performing data enhancement on the sample images in the training set to obtain enhanced image data;
and the training module (not shown) is used for training the network model through the enhanced image data, and verifying the training result through the verification set to obtain a preset visual detection model.
Further, the data enhancement module is further used for labeling industrial equipment in the sample images in the training set to obtain a labeling frame of the industrial equipment;
image enlargement is carried out on the sample image and the annotation frame through an enlargement formula, and an enlarged image is obtained;
and denoising, sharpening and brightening the enlarged image to obtain enhanced image data.
Further, the training module is further configured to:
setting initialization weights and parameters of a training process;
Inputting the enhanced image data into a network model for model training, updating and optimizing each group of parameters in the network model by using a random gradient descent method, inputting a verification set into the trained network model for calculating mAP values of the network model after each round of model training, and updating weights of the current optimal mAP values;
after training the preset number of rounds, selecting a network model corresponding to the weight of the optimal mAP value as a preset visual detection model.
The modules in the industrial equipment fault detection apparatus 10 correspond to the steps in the industrial equipment fault detection method embodiment, and the functions and implementation processes thereof are not described herein in detail.
Furthermore, the application also provides a computer readable storage medium.
The computer readable storage medium of the present application stores an industrial equipment fault detection program, wherein the industrial equipment fault detection program, when executed by a processor, implements the steps of the industrial equipment fault detection method as described above.
The method implemented when the industrial equipment fault detection program is executed may refer to various embodiments of the industrial equipment fault detection method of the present application, which are not described herein.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It should be noted that in the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The application may be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by one and the same item of hardware. The use of the words first, second, third, etc. do not denote any order. These words may be interpreted as names.
While preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiments and all such alterations and modifications as fall within the scope of the application.
The foregoing description is only of the preferred embodiments of the present application and is not intended to limit the scope of the application, and all equivalent structural changes made by the description of the present application and the accompanying drawings or direct/indirect application in other related technical fields are included in the scope of the application.
Claims (10)
1. An industrial equipment fault detection method, characterized in that the industrial equipment fault detection method comprises:
acquiring a detection image of industrial equipment;
inputting the detection image into a preset visual detection model to obtain a visual information feature vector of a target to be identified;
obtaining an inference result according to the visual information feature vector and a preset multi-mode knowledge graph;
and generating fault text according to the reasoning result.
2. The industrial equipment fault detection method according to claim 1, wherein the step of obtaining an inference result according to the visual information feature vector and a preset multi-modal knowledge-graph comprises:
Extracting keywords from the visual information feature vector to obtain a feature text vector;
sorting the characteristic text vectors with the confidence coefficient higher than a preset threshold value from high to low to obtain a text vector set;
and obtaining the reasoning result according to the text vector set and a preset multi-mode knowledge graph.
3. The industrial equipment fault detection method according to claim 2, wherein the step of obtaining the inference result according to the text vector set and a preset multi-modal knowledge-graph comprises:
sequentially performing cosine similarity matching on the text vectors in the text vector set and entities, attributes and relations in a preset multi-mode knowledge graph to obtain fault vectors with highest similarity corresponding to each text vector;
obtaining a fault triplet set according to the fault vector and the semantic matching template;
and adopting path question sentences and multi-hop question sentences to carry out the graph searching and the joint-by-joint strategy reasoning on the fault triplet set to generate the reasoning result.
And adopting a path question and a multi-hop question for the fault triplet set, namely, when one entity and a plurality of relations are involved, starting from a first-level entity, finding a second-level entity through a first-level relation, and finding a fault entity through a fault attribute value to obtain a result. The method is characterized by comprising the steps of starting from the obtained first-layer entity according to a drawing policy, traversing the graph reasoning and generating a result.
4. A method of fault detection of an industrial device as claimed in claim 3 wherein the step of generating fault text from the inference results comprises:
and inputting the reasoning result into a text generation model to generate a fault text.
5. The industrial equipment failure detection method according to any one of claims 1 to 4, wherein the step of acquiring the detection image of the industrial equipment is preceded by:
collecting sample images of industrial equipment, and dividing the sample images into a training set and a verification set;
carrying out data enhancement on the sample images in the training set to obtain enhanced image data;
and training the network model through the enhanced image data, and verifying the training result through the verification set to obtain a preset visual detection model.
6. The industrial equipment fault detection method of claim 5, wherein the step of data enhancing the sample images in the training set to obtain enhanced image data comprises:
labeling industrial equipment in the sample images in the training set to obtain a labeling frame of the industrial equipment;
image enlargement is carried out on the sample image and the annotation frame through an enlargement formula, and an enlarged image is obtained;
And carrying out noise reduction, sharpening and brightening on the enlarged image to obtain the enhanced image data.
7. The industrial equipment fault detection method of claim 5, wherein the step of training the network model with the enhanced image data and verifying the training result with the verification set, obtaining a preset visual detection model comprises:
setting initialization weights and parameters of a training process;
inputting the enhanced image data into a network model for model training, updating and optimizing each group of parameters in the network model by using a random gradient descent method, inputting a verification set into the trained network model for calculating mAP values of the network model after each round of model training, and updating weights of the current optimal mAP values;
after training the preset number of rounds, selecting a network model corresponding to the weight of the optimal mAP value as a preset visual detection model.
8. An industrial equipment fault detection device, characterized in that the industrial equipment fault detection device comprises:
the acquisition module is used for acquiring a detection image of the industrial equipment;
the first obtaining module is used for inputting the detection image into a preset visual detection model to obtain a visual information feature vector of the target to be identified;
The second obtaining module is used for obtaining an reasoning result according to the visual information feature vector and a preset multi-mode knowledge graph;
and the generating module is used for generating a fault text according to the reasoning result.
9. An industrial equipment fault detection device comprising a processor, a memory, and an industrial equipment fault detection program stored on the memory that is executable by the processor, wherein the industrial equipment fault detection program, when executed by the processor, implements the steps of the industrial equipment fault detection method according to any one of claims 1 to 7.
10. A computer-readable storage medium, wherein an industrial equipment failure detection program is stored on the computer-readable storage medium, wherein the industrial equipment failure detection program, when executed by a processor, implements the steps of the industrial equipment failure detection method according to any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310721247.1A CN116630749A (en) | 2023-06-16 | 2023-06-16 | Industrial equipment fault detection method, device, equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310721247.1A CN116630749A (en) | 2023-06-16 | 2023-06-16 | Industrial equipment fault detection method, device, equipment and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116630749A true CN116630749A (en) | 2023-08-22 |
Family
ID=87609900
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310721247.1A Pending CN116630749A (en) | 2023-06-16 | 2023-06-16 | Industrial equipment fault detection method, device, equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116630749A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117556920A (en) * | 2023-10-23 | 2024-02-13 | 星环信息科技(上海)股份有限公司 | Large model illusion treatment method, device, equipment and storage medium |
-
2023
- 2023-06-16 CN CN202310721247.1A patent/CN116630749A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117556920A (en) * | 2023-10-23 | 2024-02-13 | 星环信息科技(上海)股份有限公司 | Large model illusion treatment method, device, equipment and storage medium |
CN117556920B (en) * | 2023-10-23 | 2024-05-31 | 星环信息科技(上海)股份有限公司 | Large model illusion treatment method, device, equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110610166B (en) | Text region detection model training method and device, electronic equipment and storage medium | |
CN112348117B (en) | Scene recognition method, device, computer equipment and storage medium | |
CN110717534B (en) | Target classification and positioning method based on network supervision | |
WO2019015246A1 (en) | Image feature acquisition | |
CN111198964B (en) | Image retrieval method and system | |
CN110795527B (en) | Candidate entity ordering method, training method and related device | |
CN109635135A (en) | Image index generation method, device, terminal and storage medium | |
CN114241505B (en) | Method and device for extracting chemical structure image, storage medium and electronic equipment | |
CN116630749A (en) | Industrial equipment fault detection method, device, equipment and storage medium | |
CN113127667A (en) | Image processing method and device, and image classification method and device | |
CN111582358A (en) | Training method and device for house type recognition model and house type weight judging method and device | |
CN114882267A (en) | Small sample image classification method and system based on relevant region | |
CN110852263B (en) | Mobile phone photographing garbage classification recognition method based on artificial intelligence | |
CN114445691A (en) | Model training method and device, electronic equipment and storage medium | |
CN112200862B (en) | Training method of target detection model, target detection method and device | |
CN107193979B (en) | Method for searching homologous images | |
CN111191065B (en) | Homologous image determining method and device | |
CN111126617B (en) | Method, device and equipment for selecting fusion model weight parameters | |
CN110727762B (en) | Method, device, storage medium and electronic equipment for determining similar texts | |
CN111428612A (en) | Pedestrian re-identification method, terminal, device and storage medium | |
CN116363469A (en) | Method, device and system for detecting infrared target with few samples | |
CN110929644A (en) | Heuristic algorithm-based multi-model fusion face recognition method and device, computer system and readable medium | |
CN113282781B (en) | Image retrieval method and device | |
CN113297411B (en) | Method, device and equipment for measuring similarity of wheel-shaped atlas and storage medium | |
CN112465002B (en) | Intelligent recognition method and device for copper wire weld mark on fire scene |
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 |