CN117591901A - Insulator breakage detection method and device, storage medium and electronic equipment - Google Patents
Insulator breakage detection method and device, storage medium and electronic equipment Download PDFInfo
- Publication number
- CN117591901A CN117591901A CN202410065968.6A CN202410065968A CN117591901A CN 117591901 A CN117591901 A CN 117591901A CN 202410065968 A CN202410065968 A CN 202410065968A CN 117591901 A CN117591901 A CN 117591901A
- Authority
- CN
- China
- Prior art keywords
- insulator
- image
- matching model
- text matching
- text
- 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.)
- Granted
Links
- 239000012212 insulator Substances 0.000 title claims abstract description 326
- 238000001514 detection method Methods 0.000 title claims abstract description 80
- 238000003860 storage Methods 0.000 title claims abstract description 12
- 238000012549 training Methods 0.000 claims abstract description 112
- 230000006978 adaptation Effects 0.000 claims abstract description 39
- 230000006870 function Effects 0.000 claims abstract description 31
- 238000000605 extraction Methods 0.000 claims description 32
- 238000000034 method Methods 0.000 claims description 21
- 238000004140 cleaning Methods 0.000 claims description 18
- 238000004590 computer program Methods 0.000 claims description 9
- 230000009193 crawling Effects 0.000 claims description 9
- 238000004519 manufacturing process Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 description 19
- 230000005540 biological transmission Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 230000008014 freezing Effects 0.000 description 4
- 238000007710 freezing Methods 0.000 description 4
- 238000012795 verification Methods 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 238000003491 array Methods 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 208000025274 Lightning injury Diseases 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000010292 electrical insulation Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000009413 insulation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000004643 material aging Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000035882 stress Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/148—Segmentation of character regions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/19007—Matching; Proximity measures
- G06V30/19093—Proximity measures, i.e. similarity or distance measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/191—Design or setup of recognition systems or techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
- G06V30/1912—Selecting the most significant subset of features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/191—Design or setup of recognition systems or techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
- G06V30/19147—Obtaining sets of training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an insulator damage detection method, an insulator damage detection device, a storage medium and electronic equipment, wherein the insulator damage detection method comprises the following steps: acquiring an image to be detected; insulator detection is carried out on the image to be detected, and an insulator image is obtained; inputting the insulator image into a pre-trained insulator image text matching model to obtain insulator image characteristics; calculating a first similarity of the insulator sub-image features and the insulator breakage text features and a second similarity of the insulator sub-image features and the insulator normal text features; and determining whether the insulator in the picture to be detected is damaged or not according to the first similarity and the second similarity. According to the insulator image text matching model, the multi-mode insulator image text matching model is trained from texts and images, so that an efficient and accurate insulator breakage detection function is realized, the structure of a bit 3 image text matching network is improved through the plug-in adaptation module, and the limitation of a large model to a video memory is solved only by training the adaptation module.
Description
Technical Field
The present invention relates to the field of image detection technologies, and in particular, to a method and apparatus for detecting insulator breakage, a storage medium, and an electronic device.
Background
The insulator is a special insulation control and plays a key role in the field of power transmission and distribution. They not only provide electrical insulation for the wires, preventing short circuits or ground faults, but also have a mechanical support function, ensuring that the wires are subjected to the load of the external environment. The insulator is used for the connection part of the wire pole tower and the wire bearing part, and the substation framework and the line. Because the distribution of the power transmission lines in China is very wide, the geographical environment is complex, the power transmission cables are exposed in the field for a long time, and the insulator can be subjected to lightning stroke, material aging and the like to generate a series of problems of corrosion, damage, overvoltage breakdown and the like. The damage of the insulator is analyzed, and the insulator is mainly used for preventing the insulator from insulating failure caused by various electromechanical stresses caused by the change of the environment and the electric load conditions, thereby damaging the service life and the service life of the power line. However, the manual detection has the problems of low efficiency, low detection rate, low safety, high economic cost and the like, so that the intelligent inspection of the power transmission line increasingly shows the urgency for ensuring the reliability and safe power supply of the insulator. Therefore, an automatic, accurate and real-time monitoring manner is needed, and the damage condition of the insulator in the power supply line can be timely judged and timely processed.
In the related art, the tasks of detecting breakage of an insulator are generally divided into two types. The first method is to directly detect damaged insulators through a target detection algorithm, and the second method is to detect or classify the cut insulator small drawings in two stages on the basis of insulators detected in one stage. Neither of these two approaches circumvents one problem: the data collection difficulty of the damaged insulator is high. For insulator breakage detection tasks, there are serious problems in training data collection: the normal insulators account for the vast majority, the proportion of the normal insulators to the damaged insulators is seriously unbalanced, the data acquisition difficulty of the damaged insulators is high, and the recall rate is generally low by utilizing the traditional target detection or classification scheme. The reliability of the insulator is very important for safe power supply, an electric power company can timely process and replace the damaged insulator, and a model with high recall rate cannot be obtained by using limited damage data through a traditional insulator damage detection method.
Disclosure of Invention
The present invention aims to solve at least one of the technical problems in the related art to some extent. Therefore, an object of the present invention is to provide an insulator breakage detection method, an apparatus, a storage medium and an electronic device, in which the insulator image text matching model of the present invention guides multi-mode insulator image text matching model training from text and image, to realize efficient and accurate insulator breakage detection function, and the bit 3 image text matching network structure is improved by the plug-in adaptation module, only the adaptation module is trained, and the limit of the large model to the video memory is solved, so as to obtain the insulator breakage detection model with high recall rate.
To achieve the above object, an embodiment of a first aspect of the present invention provides a method for detecting breakage of an insulator, the method including: acquiring an image to be detected; insulator detection is carried out on the image to be detected, and an insulator image is obtained; inputting the insulator image into a pre-trained insulator image text matching model to obtain insulator image characteristics; calculating a first similarity between the insulator image features and insulator breakage text features and a second similarity between the insulator image features and insulator normal text features, wherein the insulator breakage text features are obtained by inputting breakage texts into an insulator image text matching model, and the insulator normal text features are obtained by inputting normal texts into the insulator image text matching model; and determining whether the insulator in the picture to be detected is damaged or not according to the first similarity and the second similarity.
In addition, the insulator breakage detection method according to the embodiment of the present invention may further have the following additional technical features:
according to one embodiment of the invention, the insulator image text matching model is trained by the following method: manufacturing a first preset number of insulator image text pairs, wherein the insulator image text pairs comprise a second preset number of normal insulator image text pairs and a third preset number of damaged insulator image text pairs; and inputting the first preset number of the text pairs of the insulator images into a pre-trained first wait 3 image text matching model, and training to obtain a trained text matching model of the insulator images.
According to one embodiment of the invention, the first wait 3 image text matching model comprises a second wait 3 image text matching model, and the first wait 3 image text matching model is trained by the following method: acquiring a training data set, wherein the training data set comprises text data corresponding to a normal image and text data corresponding to a damaged image; training the second wait 3 image text matching model by using the training data set to obtain training weights of the second wait 3 image text matching model; and loading the training weight of the second wait 3 image text matching model, and training the first wait 3 image text matching model according to the training weight of the second wait 3 image text matching model to obtain a trained first wait 3 image text matching model.
According to one embodiment of the invention, the acquiring a training data set includes: crawling a preset type of picture set, wherein the picture set comprises damaged object pictures and normal object pictures; dividing the picture set according to a preset proportion to obtain a first sub-picture set and a second sub-picture set; screening the first sub-picture set, and removing noise pictures to obtain a third sub-picture set; cleaning the third sub-picture set by using the first wait 3 image text matching model to obtain a data cleaning model; and cleaning the second sub-picture set by using the data cleaning model, and taking the cleaned data as a training data set.
According to one embodiment of the present invention, the first wait 3 image text matching model further comprises an adaptation model, the output of the second wait 3 image text matching model is connected to the input of the adaptation model, and the training the first wait 3 image text matching model according to the training weights of the second wait 3 image text matching model comprises: training the adaptation model according to the training weight of the second wait 3 image text matching model to obtain the training weight of the adaptation module.
According to one embodiment of the present invention, the second wait 3 image text matching model includes an image feature extraction module, a text feature extraction module, and a first loss function module, wherein outputs of the image feature extraction module and the text feature extraction module are respectively connected to inputs of the first loss function module, and an output of the first loss function module is used as an output of the second wait 3 image text matching model.
According to one embodiment of the invention, the adaptation model comprises a downsampling module, a nonlinear layer module, an upsampling module, a residual module and a second loss function module which are sequentially connected, wherein the output of the second loss function module is used as the output of the first wait 3 image text matching model.
According to an embodiment of the present invention, the first similarity includes a first cosine similarity, the second similarity includes a second cosine similarity, and determining whether an insulator in the picture to be detected is broken according to the first similarity and the second similarity includes: and if the first cosine similarity is larger than the second cosine similarity, determining that the insulator in the picture to be detected is damaged.
According to the insulator damage detection method, a target detection algorithm and a multi-mode text image matching algorithm are selected according to requirements and limitation of damaged insulator data size, and insulator damage detection is achieved through two-stage combination. Firstly, all insulators are detected from an image (namely, one stage), then the insulators detected in one stage are cut off in an original image, a small image of the cut insulators is input into a multi-mode image feature extraction module, similarity comparison is carried out between the small image of the cut insulators and texts (brooken insulator and normal insulator) subjected to feature extraction in advance, and the small image of the cut insulators is matched with texts with large similarity so as to judge whether the insulators are damaged. The invention inserts additional modules into the pre-training model based on the wait 3 pre-training model, thereby conveniently fine-adjusting the model. And freezing the original network parameters of the wait 3, and only training an adapter module, thereby solving the limit of the large model on the video memory. According to the invention, a multi-mode text image matching algorithm is applied to insulator damage detection, and meanwhile, an additional module is finely adjusted in a pre-training model by utilizing limited insulator damage data, so that an insulator damage detection model with high recall rate is obtained.
To achieve the above object, an embodiment of a second aspect of the present invention provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements an insulator breakage detection method as described above.
To achieve the above object, an embodiment of a third aspect of the present invention provides an insulator breakage detection device, including: the acquisition module is used for acquiring the image to be detected; the detection module is used for carrying out insulator detection on the image to be detected to obtain an insulator image; the input module is used for inputting the insulator image into a pre-trained insulator image text matching model to obtain insulator image characteristics; the calculating module is used for calculating the first similarity between the insulator image features and the insulator breakage text features and the second similarity between the insulator image features and the insulator normal text features, wherein the insulator breakage text features are obtained by inputting breakage texts into the insulator image text matching model, and the insulator normal text features are obtained by inputting normal texts into the insulator image text matching model; and the determining module is used for determining whether the insulator in the picture to be detected is damaged or not according to the first similarity and the second similarity.
To achieve the above object, a fourth aspect of the present invention provides an electronic device, including a memory and a processor, where the memory stores a computer program, and the computer program implements the insulator breakage detection method as described above when executed by the processor.
According to the insulator damage detection device, the storage medium and the electronic equipment, the target detection algorithm and the multi-mode text image matching algorithm are selected according to requirements and limitation of damaged insulator data quantity, and the insulator damage detection is achieved through two-stage combination. Firstly, all insulators are detected from an image (namely, one stage), then the insulators detected in one stage are cut off in an original image, a small image of the cut insulators is input into a multi-mode image feature extraction module, similarity comparison is carried out between the small image of the cut insulators and texts (brooken insulator and normal insulator) subjected to feature extraction in advance, and the small image of the cut insulators is matched with texts with large similarity so as to judge whether the insulators are damaged. The invention inserts additional modules into the pre-training model based on the wait 3 pre-training model, thereby conveniently fine-adjusting the model. And freezing the original network parameters of the wait 3, and only training an adapter module, thereby solving the limit of the large model on the video memory. According to the invention, a multi-mode text image matching algorithm is applied to insulator damage detection, and meanwhile, an additional module is finely adjusted in a pre-training model by utilizing limited insulator damage data, so that an insulator damage detection model with high recall rate is obtained.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
Fig. 1 is a flowchart of an insulator breakage detection method according to an embodiment of the present invention;
FIG. 2 is a training flow diagram of an insulator image text matching model in accordance with one embodiment of the present invention;
FIG. 3 is a training flow diagram of a first wait 3 image text matching model according to one embodiment of the invention;
FIG. 4 is a flow chart of acquiring a training data set in accordance with one embodiment of the present invention;
FIG. 5 is a schematic diagram of a second wait 3 image text matching model according to one embodiment of the invention;
FIG. 6 is a schematic diagram of a first wait 3 image text matching model according to one embodiment of the invention;
fig. 7 is a schematic view of an insulator breakage detection device according to an embodiment of the present invention;
fig. 8 is a schematic diagram of an electronic device according to an embodiment of the invention.
Detailed Description
Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are illustrative and intended to explain the present invention and should not be construed as limiting the invention.
The method, apparatus, storage medium and electronic device for detecting breakage of an insulator according to the embodiments of the present invention will be described in detail below with reference to the accompanying drawings and detailed description.
Fig. 1 is a flowchart of an insulator breakage detection method according to an embodiment of the present invention.
In one embodiment of the present invention, as shown in fig. 1, the insulator breakage detection method includes:
s1, acquiring an image to be detected.
Specifically, the image to be detected may be an image of a power transmission line captured randomly, and in order to ensure reliability and safe power supply of the insulator, the captured image of the power transmission line is detected to detect whether the insulator in the power transmission line is damaged.
S2, insulator detection is carried out on the image to be detected, and an insulator image is obtained.
Specifically, the collected images to be detected possibly comprise insulators, or possibly do not comprise insulators, or comprise a small part of insulators, if the collected images are not distinguished, the unified processing is directly carried out on all the collected images, so that the workload is greatly increased. The invention can use the target detection algorithm YOLO latest series YOLOV8 target detection algorithm to carry out target detection on an image to be detected, the algorithm principle is that an input image is subjected to characteristic extraction of a multi-layer convolution network and then is output through a multi-scale output layer, regression calculation of the target is directly carried out on a characteristic diagram, a target frame is obtained, and a final output result of the YOLOV8 algorithm is subjected to non-maximum value inhibition operation, so that a final target detection frame is obtained. And dividing the original image according to the insulator target detection result to obtain an insulator image.
Further specifically, insulator detection is performed on an image to be detected, an insulator target detection model is utilized, a data set is required to be manufactured in advance aiming at the insulator target detection model, and the insulator target detection model is trained. The image data may be annotated using a Labelme tool, and may be annotated in the upper left and lower right corners of the image. The data is normalized by dividing the image width and height values to generate a label, and the label format may be a class and rectangular center point and width and height (class, center, width, height). Because the resolution of the image of the unmanned aerial vehicle acquired data set is higher, 1280 x 1280 can be used as an input yolov8L target detection model, as an embodiment, an adopted yolov8 algorithm of an ultra-analysis open source, and finally an insulator test AP value on a test set is 0.984.
And detecting an insulator target by using an insulator detection model, cutting all insulator images from an original image, inputting the insulator images into a pre-trained insulator image text matching model after obtaining the insulator images, and judging whether the insulators in the images to be detected are damaged by using the insulator image text matching model.
And S3, inputting the insulator image into a pre-trained insulator image text matching model to obtain the insulator image characteristics.
Specifically, an insulator image is input into a pre-trained insulator image text matching model, the insulator image text matching model is utilized to extract the image characteristics of the insulator, the insulator image characteristics are compared with the damaged text and the normal text which are subjected to characteristic extraction in advance, and whether the insulator image is damaged or not is judged.
In one embodiment of the present invention, as shown in fig. 2, the insulator image text matching model is trained by the following method:
s31, manufacturing a first preset number of insulator sub-image text pairs, wherein the insulator sub-image text pairs comprise a second preset number of normal insulator sub-image text pairs and a third preset number of damaged insulator sub-image text pairs.
S32, inputting a first preset number of insulator image text pairs into a pre-trained first wait 3 image text matching model, and training to obtain a trained insulator image text matching model.
Specifically, the text matching model of the insulator image is obtained by inputting the insulator image into a pre-trained text matching model of a wait 3 image and training the text matching model of the insulator image. The pre-trained wait 3 image text matching model is a model that has learned the meaning of "broken" and is named the first wait 3 image text matching model. And inputting the pair of insulator image texts into an image text model which is learned to be in a broken meaning, so as to obtain the image text model which is learned to be in a broken meaning of the insulator.
Further specifically, first preset number of insulator image text pairs are manufactured, the detected insulators are scratched out of an original image by using the trained insulator detection model, two classifications are carried out on each scratched insulator small image, the normal insulator is classified into normal and damaged insulators, the normal insulator corresponds to the normal insulator, the damaged insulator corresponds to the rotor insulator, third preset number of damaged insulators and second preset number of normal insulators are finally collected, the third preset number can be 2w, and the second preset number can be 10 w. Inputting a first preset number of insulator image text pairs into a pre-trained first wait 3 image text matching model, and training to obtain a multi-mode insulator image text matching model.
The multi-modal text image matching model is mainly modeled for correlation between text and images. According to the method, an additional module is inserted into the pre-training model on the basis of the wait 3 pre-training model, so that a new wait 3 pre-training model is obtained, namely the first wait 3 image text matching model, and the model is conveniently fine-tuned. And extracting image and text features by using a transducer in the model. By learning the embedded representation of text and images in a common semantic space, the degree of matching between them can be determined.
According to the invention, a multi-mode text image matching algorithm is applied to insulator damage detection, and meanwhile, an additional module is finely adjusted in a pre-training model by utilizing limited insulator damage data, so that an insulator damage detection model with high recall rate is obtained.
In one embodiment of the present invention, as shown in fig. 3, the first wait 3 image text matching model includes a second wait 3 image text matching model, and the first wait 3 image text matching model is trained by the following method:
s321, acquiring a training data set, wherein the training data set comprises text data corresponding to a normal image and text data corresponding to a damaged image.
Specifically, the invention improves the structure of the fit 3 image text matching network through externally hanging an adaptation model adapter, the improved fit 3 image text matching model is a first fit 3 image text matching model, namely the first fit 3 image text matching model comprises a second fit 3 image text matching model and an adaptation model, and the first fit 3 image text matching model is trained to obtain an image text matching model with a learned 'broken' meaning.
Further specifically, training a first bit 3 image text matching model, firstly acquiring a training data set, using a crawler to crawl image text pairs, training the image-text pairs by using less part of data through artificial verification, performing data cleaning on the rest image-text pairs by using the first bit 3 image text matching model to obtain training weights of a second bit 3 image text matching model, and finally performing finishing on the cleaned data on the second bit 3 image text matching model (namely loading the training weights of the second bit 3 image text matching model and performing parameter fine adjustment on the adaptation model only), thereby obtaining the bit 3 image text matching model for learning 'breakage'.
In one embodiment of the present invention, as shown in FIG. 4, acquiring the training data set includes:
s3211, crawling a preset type of picture set, wherein the picture set comprises damaged object pictures and normal object pictures.
S3212, dividing the picture sets according to a preset proportion to obtain a first sub-picture set and a second sub-picture set.
S3213, screening the first sub-picture set, and removing the noise picture to obtain a third sub-picture set.
S3214, cleaning the third sub-picture set by using the first wait 3 image text matching model to obtain a data cleaning model.
S3215, cleaning the second sub-picture set by using the data cleaning model, and taking the cleaned data as a training data set.
Specifically, the training data is an image text matching model with the meaning of breakage of a training society, so that the training data set not only comprises breakage and normal images of insulators, but also comprises breakage and normal images of various types, a crawler can be utilized to perform keyword crawling on a preset type of picture set, and the picture set comprises broken object pictures and normal object pictures. After crawling the pictures, cleaning the crawled pictures, and using a method combining artificial verification and model filtering to clean the picture set, so that noise images in the picture set can be manually removed to obtain noiseless images, then cleaning the noiseless images by using a first wait 3 image text matching model, and taking the cleaned data as a training data set.
As an example, the preset categories may be 150 categories, and the search keywords 'brooken_xx_', '_ xx_', such as brooken sofa, brooken wall, brooken bowl, etc., each object is regarded as a category collecting 150 categories of objects, and each damaged object crawls 10500 pictures with normal objects, each category being 21000 pictures. And then, data cleaning is carried out on the crawling data, each type of data is firstly separated according to a preset proportion, wherein the preset proportion can be 1000:20000, namely 21000 pictures are separated according to the proportion of 1000:20000, a first sub-picture set and a second sub-picture set are obtained, the first sub-picture set comprises 1000 pictures, and the second sub-picture set comprises 20000 pictures. The first sub-picture set can be manually screened, 1000 crawling pictures (whether normal pictures are normal or not and whether damaged pictures exist or not) are manually checked, irrelevant pictures are rejected, about 12 ten thousand noiseless pictures are obtained after manual checking, the noiseless pictures are a third sub-picture set, the third sub-picture set is manufactured into image text matching training data (normal image corresponding text: normal_xx_e.g. normal bootl, damaged image corresponding text: brookfen_xx_e.g. brookfen bootl), image text data manufactured by the third sub-picture set (namely about 12 ten thousand noiseless images) are cleaned by using the first wait 3 image text matching model, a training data cleaning model of the society 'damaged' is obtained after training, the second sub-picture set (namely 150 classes of text is cleaned by using the data cleaning model), finally about 118w image text pairs are obtained through manual checking and finally the image text pairs are cleaned as training data, and the image pairs are manufactured as training data pairs. The invention adopts the crawling of the damaged and normal image text pairs, and cleans the crawling data through artificial verification and model filtration to obtain high-quality pre-training data.
After the training data set is obtained, training a second bit 3 image text matching model in the first bit 3 image text matching model by using the training data set to obtain the training weight of the second bit 3 image text matching model, so that in the subsequent model training, the training weight of the second bit 3 image text matching model is kept unchanged, only parameters of an adaptation model in the first bit 3 image text matching model are required to be adjusted, and the limit of a large model on the video memory is solved.
S322, training the second wait 3 image text matching model by using the training data set to obtain training weights of the second wait 3 image text matching model.
S323, loading training weights of the second wait 3 image text matching model, and training the first wait 3 image text matching model according to the training weights of the second wait 3 image text matching model to obtain a trained first wait 3 image text matching model.
Specifically, the cleaned image text pair is used as a training data set, a second wait 3 image text matching model is trained to obtain training weights of the second wait 3 image text matching model, the training weights of the second wait 3 image text matching model are loaded, the training weights of the second wait 3 image text matching model are kept unchanged, the whole first wait 3 image text matching model is trained, parameters of an adaptation model in the first wait 3 image text matching model are adjusted to obtain a trained first wait 3 image text matching model, namely an image text matching model with a learned "broken" meaning is obtained, and then the insulator image data is input into the image text matching model with the learned "broken" meaning to obtain the image text matching model with the learned insulator "broken" meaning.
In one embodiment of the present invention, the first wait 3 image text matching model further comprises an adaptation model, the output of the second wait 3 image text matching model is connected to the input of the adaptation model, the training of the first wait 3 image text matching model according to the training weights of the second wait 3 image text matching model comprises: and training the adaptation model according to the training weight of the second wait 3 image text matching model to obtain the training weight of the adaptation module.
Specifically, the first wait 3 image text matching model comprises a second wait 3 image text matching model and an adaptation model, the input of the second wait 3 image text matching model is used as the input of the first wait 3 image text matching model, the output of the second wait 3 image text matching model is connected with the input of the adaptation model, and the output of the adaptation model is used as the output of the first wait 3 image text matching model.
Further specifically, training the second fit 3 image text matching model by using the training data set to obtain training weights of the second fit 3 image text matching model, training the first fit 3 image text matching model according to the training weights of the second fit 3 image text matching model, namely loading the training weights of the second fit 3 image text matching model, keeping the training weights of the second fit 3 image text matching model unchanged, and adjusting the adaptation model in the whole first fit 3 image text matching model to obtain the training weights of the adaptation module. And the two weights are combined to obtain an image text matching model for learning the meaning of breakage, and then the data of the insulator image is input into the image text matching model for learning the meaning of breakage to obtain the image text matching model for learning the meaning of breakage of the insulator.
In one embodiment of the present invention, the second wait 3 image text matching model includes an image feature extraction module, a text feature extraction module, and a first loss function module, wherein outputs of the image feature extraction module and the text feature extraction module are respectively connected to inputs of the first loss function module, and outputs of the first loss function module are used as outputs of the second wait 3 image text matching model.
Specifically, the second wait 3 image text matching model comprises an image feature extraction module, a text feature extraction module and a first loss function module, wherein the image feature extraction module is used for extracting image features, the text feature extraction module is used for extracting text features, and the image feature extraction module and the text feature extraction module share a Multi-Head Self-Attention part. The output of the image feature extraction module and the output of the text feature extraction module are respectively connected with the input of the first loss function module, and the first loss function module calculates the loss function of the extracted image features and text features.
As an example, as shown in fig. 5, fig. 5 shows that the second wait 3 is an image text matching structure, and the image end in the image feature extraction module enters the n layer transformer layer after passing through a patch (blocking) and an ebeddings (coding) (the invention can use a backup based on ViT-giant, so n can be 40), and the image feature is extracted. Text in the text feature extraction module passes through the empeddings and enters the n layers transformer layer to extract text features. The image and text share Multi-Head Self-Attention mechanism Multi-Head Self-Attention part in transformer layer (a neural network structure), the image and text have respective feedforward neural networks FeedForward neural Network, and finally the extracted image features and text features are input to a first loss function module to be itc loss (image-text contrastive loss). The ITC loss function is shown in the following equation,
Wherein,for the ith text feature>For the ith visual feature +.>、/>Represents a positive sample pair (features from the same pair of image and text, should have high similarity),>、/>representing negative pairs of samples (features from different pairs of images and text, should have low similarity),>representation->,/>Inner product between two features, ++>Is a temperature parameter, ++>Is the number of samples. The ITC loss function ensures that the model learns consistency between the image and text from two directions through log. The first encouragement model maps text features to spaces that are close to the visual features corresponding to the positive sample, and the second encouragement model maps visual features to spaces that are close to the text features corresponding to the positive sample. Consistent training of these two directions helps to enhance the robustness of the modelThe swiping and generalization ability ensures good performance in multi-modal understanding tasks.
In one embodiment of the invention, the adaptation model comprises a downsampling module, a nonlinear layer module, an upsampling module, a residual module and a second loss function module which are sequentially connected, wherein the output of the second loss function module is used as the output of the first wait 3 image text matching model.
Specifically, the output of the second wait 3 for image text matching is connected with the input of an adaptation model, the adaptation model comprises a downsampling module, a nonlinear layer module, an upsampling module, a residual module and a second loss function module which are sequentially connected, namely, the input of the downsampling module is used as the input of the adaptation model, and the output of the second loss function module is used as the output of the adaptation model and is also the output of the first wait 3 image text matching model.
As an example, as shown in FIG. 6, the invention improves the structure of the bit 3 image text matching network by externally hanging an adaptation model adapter, the bit 3 uses a backup based on ViT-giant, and the invention uses a mode of adding the externally hanging adaptation to the pre-training model based on the bit 3 due to the restriction of calculation force and video memory, and freezes the original model parameters of the bit 3 during the finishing, so that only the parameters in the adaptation module are trained. The structure of the adaptation module is shown on the right side of fig. 6, when the image and the text extract the features through the respective Feedforward neural network Feedforward neural Network, the downsampling module feed down-project is first performed, then two nonlinear layers (mix+swish) can be performed through the nonlinear module, finally the upsampling module feed up-project is performed to the dimension of the input, finally the residual edges are added with the input of the adaptation module adaptation, and the final image features and text features are obtained after transformer layer of 40 layers. The same final input is to the second loss function module for ITC loss calculation. The ITC loss calculation is described above and will not be described in detail here.
And comparing the insulator image with the insulator damage text characteristics and the insulator normal text characteristics obtained by training in advance after extracting the insulator image characteristics through the pre-trained insulator image text matching model so as to judge whether the insulator image to be detected currently is a damaged insulator.
S4, calculating a first similarity between the insulator image features and the insulator breakage text features and a second similarity between the insulator image features and the insulator normal text features, wherein the insulator breakage text features are obtained by inputting breakage texts into an insulator image text matching model, and the insulator normal text features are obtained by inputting normal texts into the insulator image text matching model.
Specifically, the similarity can be used as a measurement scale, and the first similarity between the insulator image features and the insulator breakage text features and the second similarity between the insulator image features and the insulator normal text features are calculated to judge whether the insulator in the image to be detected is a breakage insulator or not through comparison of the first similarity and the second similarity.
As an example, the similarity may be cosine similarity, and then a first cosine similarity between the image feature of the insulator and the damaged text feature of the insulator and a second cosine similarity between the image feature of the insulator and the normal text feature of the insulator are calculated, and whether the insulator in the image to be detected is a damaged insulator is determined by comparing the first cosine similarity with the second cosine similarity.
S5, determining whether the insulator in the picture to be detected is damaged or not according to the first similarity and the second similarity.
In one embodiment of the present invention, the first similarity includes a first cosine similarity, the second similarity includes a second cosine similarity, and determining whether an insulator in the picture to be detected is broken according to the first similarity and the second similarity includes: if the first cosine similarity is larger than the second cosine similarity, determining that the insulator in the picture to be detected is damaged.
Specifically, if the first similarity between the insulator sub-image feature and the insulator breakage text feature is greater than the second similarity between the insulator sub-image feature and the insulator normal text feature, that is, the insulator sub-image feature is more attached to the breakage insulator feature, the image to be detected corresponding to the insulator image includes a breakage insulator.
As an example, the similarity may be cosine similarity, and if the first cosine similarity between the insulator sub-image feature and the insulator damaged text feature is greater than the second cosine similarity between the insulator sub-image feature and the insulator normal text feature, that is, the insulator sub-image feature is more attached to the damaged insulator feature, the damaged insulator is included in the image to be detected corresponding to the insulator image.
According to the damage detection method provided by the embodiment of the invention, the target detection algorithm and the multi-mode text image matching algorithm are selected according to the requirements and the limitation of the data quantity of the damaged insulator, and the damage detection of the insulator is realized through the combination of two stages. Firstly, all insulators are detected from an image (namely, one stage), then the insulators detected in one stage are cut off in an original image, a small image of the cut insulators is input into a multi-mode image feature extraction module, similarity comparison is carried out between the small image of the cut insulators and texts (brooken insulator and normal insulator) subjected to feature extraction in advance, and the small image of the cut insulators is matched with texts with large similarity so as to judge whether the insulators are damaged. The invention inserts additional modules into the pre-training model based on the wait 3 pre-training model, thereby conveniently fine-adjusting the model. And freezing the original network parameters of the wait 3, and only training an adapter module, thereby solving the limit of the large model on the video memory. According to the invention, a multi-mode text image matching algorithm is applied to insulator damage detection, and meanwhile, an additional module is finely adjusted in a pre-training model by utilizing limited insulator damage data, so that an insulator damage detection model with high recall rate is obtained.
The invention also proposes a computer readable storage medium.
In one embodiment of the present invention, a computer program is stored on a computer readable storage medium, which when executed by a processor, implements the insulator breakage detection method as described above.
The invention further provides an insulator breakage detection device.
In one embodiment of the present invention, as shown in fig. 7, an insulator breakage detection device 100 includes: an acquisition module 10, configured to acquire an image to be detected; the detection module 20 is used for carrying out insulator detection on the image to be detected to obtain an insulator image; the input module 30 is configured to input an insulator image to a pre-trained text matching model of the insulator image, so as to obtain features of the insulator image; a calculating module 40, configured to calculate a first similarity between an insulator image feature and an insulator damaged text feature and a second similarity between the insulator image feature and an insulator normal text feature, where the insulator damaged text feature is obtained by inputting a damaged text into an insulator image text matching model, and the insulator normal text feature is obtained by inputting a normal text into the insulator image text matching model; the determining module 50 is configured to determine whether the insulator in the picture to be detected is damaged according to the first similarity and the second similarity.
The invention further provides electronic equipment.
In one embodiment of the present invention, as shown in fig. 8, the electronic device 200 includes a memory 60 and a processor 70, and the memory 60 stores a computer program, which when executed by the processor 70, implements the insulator breakage detection method as described above.
According to the insulator damage detection device, the storage medium and the electronic equipment, through the insulator damage detection method, a target detection algorithm and a multi-mode text image matching algorithm are selected according to requirements and limitation of damaged insulator data quantity, and the insulator damage detection is achieved in a two-stage combination mode. Firstly, all insulators are detected from an image (namely, one stage), then the insulators detected in one stage are cut off in an original image, a small image of the cut insulators is input into a multi-mode image feature extraction module, similarity comparison is carried out between the small image of the cut insulators and texts (brooken insulator and normal insulator) subjected to feature extraction in advance, and the small image of the cut insulators is matched with texts with large similarity so as to judge whether the insulators are damaged. The invention inserts additional modules into the pre-training model based on the wait 3 pre-training model, thereby conveniently fine-adjusting the model. And freezing the original network parameters of the wait 3, and only training an adapter module, thereby solving the limit of the large model on the video memory. According to the invention, a multi-mode text image matching algorithm is applied to insulator damage detection, and meanwhile, an additional module is finely adjusted in a pre-training model by utilizing limited insulator damage data, so that an insulator damage detection model with high recall rate is obtained.
It should be noted that the logic and/or steps represented in the flowcharts or otherwise described herein, for example, may be considered as a ordered listing of executable instructions for implementing logical functions, and may be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program is printed, as the program may be electronically captured, via, for instance, optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.
It is to be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, may be implemented using any one or combination of the following techniques, as is well known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.
In the description of the present specification, a description referring to terms "one embodiment," "some 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 invention. 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.
In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings are merely for convenience in describing the present invention and simplifying the description, and do not indicate or imply that the device or element being referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present invention, the meaning of "plurality" means at least two, for example, two, three, etc., unless specifically defined otherwise.
In the present invention, unless explicitly specified and limited otherwise, the terms "mounted," "connected," "secured," and the like are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally formed; can be mechanically or electrically connected; either directly or indirectly, through intermediaries, or both, may be in communication with each other or in interaction with each other, unless expressly defined otherwise. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances.
In the present invention, unless expressly stated or limited otherwise, a first feature "up" or "down" a second feature may be the first and second features in direct contact, or the first and second features in indirect contact via an intervening medium. Moreover, a first feature being "above," "over" and "on" a second feature may be a first feature being directly above or obliquely above the second feature, or simply indicating that the first feature is level higher than the second feature. The first feature being "under", "below" and "beneath" the second feature may be the first feature being directly under or obliquely below the second feature, or simply indicating that the first feature is less level than the second feature.
While embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the invention, and that variations, modifications, alternatives and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the invention.
Claims (11)
1. A method of detecting breakage of an insulator, the method comprising:
acquiring an image to be detected;
insulator detection is carried out on the image to be detected, and an insulator image is obtained;
inputting the insulator image into a pre-trained insulator image text matching model to obtain insulator image characteristics;
calculating a first similarity between the insulator image features and insulator breakage text features and a second similarity between the insulator image features and insulator normal text features, wherein the insulator breakage text features are obtained by inputting breakage texts into an insulator image text matching model, and the insulator normal text features are obtained by inputting normal texts into the insulator image text matching model;
and determining whether the insulator in the picture to be detected is damaged or not according to the first similarity and the second similarity.
2. The insulator breakage detection method according to claim 1, wherein the insulator image text matching model is trained by the following method:
manufacturing a first preset number of insulator image text pairs, wherein the insulator image text pairs comprise a second preset number of normal insulator image text pairs and a third preset number of damaged insulator image text pairs;
and inputting the first preset number of the text pairs of the insulator images into a pre-trained first wait 3 image text matching model, and training to obtain a trained text matching model of the insulator images.
3. The insulator breakage detection method according to claim 2, wherein the first wait 3 image text matching model comprises a second wait 3 image text matching model, and the first wait 3 image text matching model is trained by:
acquiring a training data set, wherein the training data set comprises text data corresponding to a normal image and text data corresponding to a damaged image;
training the second wait 3 image text matching model by using the training data set to obtain training weights of the second wait 3 image text matching model;
And loading the training weight of the second wait 3 image text matching model, and training the first wait 3 image text matching model according to the training weight of the second wait 3 image text matching model to obtain a trained first wait 3 image text matching model.
4. The method for detecting breakage of an insulator according to claim 3, wherein said acquiring a training data set includes:
crawling a preset type of picture set, wherein the picture set comprises damaged object pictures and normal object pictures;
dividing the picture set according to a preset proportion to obtain a first sub-picture set and a second sub-picture set;
screening the first sub-picture set, and removing noise pictures to obtain a third sub-picture set;
cleaning the third sub-picture set by using the first wait 3 image text matching model to obtain a data cleaning model;
and cleaning the second sub-picture set by using the data cleaning model, and taking the cleaned data as the training data set.
5. The insulator breakage detection method of claim 3, wherein the first bit 3 image text matching model further comprises an adaptation model, an output of the second bit 3 image text matching model is connected to an input of the adaptation model, and training the first bit 3 image text matching model according to training weights of the second bit 3 image text matching model comprises:
Training the adaptation model according to the training weight of the second wait 3 image text matching model to obtain the training weight of the adaptation module.
6. The insulator breakage detection method according to claim 5, wherein the second bit 3 image text matching model includes an image feature extraction module, a text feature extraction module, and a first loss function module, outputs of the image feature extraction module and the text feature extraction module are respectively connected to inputs of the first loss function module, and an output of the first loss function module is used as an output of the second bit 3 image text matching model.
7. The method for detecting breakage of an insulator according to claim 5, wherein the adaptation model includes a downsampling module, a nonlinear layer module, an upsampling module, a residual module, and a second loss function module connected in sequence, and an output of the second loss function module is used as an output of the first fit 3 image text matching model.
8. The method for detecting breakage of an insulator according to claim 1, wherein the first similarity includes a first cosine similarity, the second similarity includes a second cosine similarity, and the determining whether the insulator in the picture to be detected is broken according to the first similarity and the second similarity includes:
And if the first cosine similarity is larger than the second cosine similarity, determining that the insulator in the picture to be detected is damaged.
9. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the insulator breakage detection method according to any one of claims 1 to 8.
10. An insulator breakage detection device, characterized in that the device comprises:
the acquisition module is used for acquiring the image to be detected;
the detection module is used for carrying out insulator detection on the image to be detected to obtain an insulator image;
the input module is used for inputting the insulator image into a pre-trained insulator image text matching model to obtain insulator image characteristics;
the calculating module is used for calculating the first similarity between the insulator image features and the insulator breakage text features and the second similarity between the insulator image features and the insulator normal text features, wherein the insulator breakage text features are obtained by inputting breakage texts into the insulator image text matching model, and the insulator normal text features are obtained by inputting normal texts into the insulator image text matching model;
And the determining module is used for determining whether the insulator in the picture to be detected is damaged or not according to the first similarity and the second similarity.
11. An electronic device comprising a memory and a processor, the memory having stored thereon a computer program, wherein the computer program, when executed by the processor, implements the insulator breakage detection method of any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410065968.6A CN117591901B (en) | 2024-01-17 | 2024-01-17 | Insulator breakage detection method and device, storage medium and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410065968.6A CN117591901B (en) | 2024-01-17 | 2024-01-17 | Insulator breakage detection method and device, storage medium and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117591901A true CN117591901A (en) | 2024-02-23 |
| CN117591901B CN117591901B (en) | 2024-05-03 |
Family
ID=89910249
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410065968.6A Active CN117591901B (en) | 2024-01-17 | 2024-01-17 | Insulator breakage detection method and device, storage medium and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117591901B (en) |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190156202A1 (en) * | 2016-05-02 | 2019-05-23 | Scopito Aps | Model construction in a neural network for object detection |
| CN110210387A (en) * | 2019-05-31 | 2019-09-06 | 华北电力大学（保定） | Insulator object detection method, system, the device of knowledge based map |
| US20200019807A1 (en) * | 2017-09-12 | 2020-01-16 | Tencent Technology (Shenzhen) Company Limited | Training method of image-text matching model, bi-directional search method, and relevant apparatus |
| CN112801555A (en) * | 2021-04-01 | 2021-05-14 | 江苏海平面数据科技有限公司 | Vehicle dynamic property comprehensive evaluation method based on Internet of vehicles big data |
| US20210182615A1 (en) * | 2019-12-16 | 2021-06-17 | Yongchuan Power Supply Branch, State Grid Chongqing Electric Power Company | Alexnet-based insulator self-explosion recognition method |
| CN115238050A (en) * | 2022-06-27 | 2022-10-25 | 北京爱医声科技有限公司 | Intelligent dialogue method and device based on text matching and intention recognition fusion processing |
| CN115471487A (en) * | 2022-09-29 | 2022-12-13 | 北京星闪世图科技有限公司 | Insulator defect detection model construction and insulator defect detection method and device |
| US20230306723A1 (en) * | 2022-03-25 | 2023-09-28 | Arizona Board Of Regents On Behalf Of Arizona State University | Systems, methods, and apparatuses for implementing self-supervised domain-adaptive pre-training via a transformer for use with medical image classification |
| CN116977763A (en) * | 2022-12-28 | 2023-10-31 | 腾讯科技（深圳）有限公司 | Model training method, device, computer readable storage medium and computer equipment |
| CN116994021A (en) * | 2022-11-18 | 2023-11-03 | 腾讯科技（深圳）有限公司 | Image detection method, device, computer readable medium and electronic equipment |
| US20240010227A1 (en) * | 2022-07-11 | 2024-01-11 | Zenseact Ab | Method and system for in-vehicle self-supervised training of perception functions for an automated driving system |
-
2024
- 2024-01-17 CN CN202410065968.6A patent/CN117591901B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190156202A1 (en) * | 2016-05-02 | 2019-05-23 | Scopito Aps | Model construction in a neural network for object detection |
| US20200019807A1 (en) * | 2017-09-12 | 2020-01-16 | Tencent Technology (Shenzhen) Company Limited | Training method of image-text matching model, bi-directional search method, and relevant apparatus |
| CN110210387A (en) * | 2019-05-31 | 2019-09-06 | 华北电力大学（保定） | Insulator object detection method, system, the device of knowledge based map |
| US20210182615A1 (en) * | 2019-12-16 | 2021-06-17 | Yongchuan Power Supply Branch, State Grid Chongqing Electric Power Company | Alexnet-based insulator self-explosion recognition method |
| CN112801555A (en) * | 2021-04-01 | 2021-05-14 | 江苏海平面数据科技有限公司 | Vehicle dynamic property comprehensive evaluation method based on Internet of vehicles big data |
| US20230306723A1 (en) * | 2022-03-25 | 2023-09-28 | Arizona Board Of Regents On Behalf Of Arizona State University | Systems, methods, and apparatuses for implementing self-supervised domain-adaptive pre-training via a transformer for use with medical image classification |
| CN115238050A (en) * | 2022-06-27 | 2022-10-25 | 北京爱医声科技有限公司 | Intelligent dialogue method and device based on text matching and intention recognition fusion processing |
| US20240010227A1 (en) * | 2022-07-11 | 2024-01-11 | Zenseact Ab | Method and system for in-vehicle self-supervised training of perception functions for an automated driving system |
| CN115471487A (en) * | 2022-09-29 | 2022-12-13 | 北京星闪世图科技有限公司 | Insulator defect detection model construction and insulator defect detection method and device |
| CN116994021A (en) * | 2022-11-18 | 2023-11-03 | 腾讯科技（深圳）有限公司 | Image detection method, device, computer readable medium and electronic equipment |
| CN116977763A (en) * | 2022-12-28 | 2023-10-31 | 腾讯科技（深圳）有限公司 | Model training method, device, computer readable storage medium and computer equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117591901B (en) | 2024-05-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114463594A (en) | Multi-mode deep learning power generation equipment abnormity integrated identification method and equipment | |
| CN110827251A (en) | Power transmission line locking pin defect detection method based on aerial image | |
| CN111337789A (en) | Method and system for detecting fault electrical element in high-voltage transmission line | |
| CN115471487A (en) | Insulator defect detection model construction and insulator defect detection method and device | |
| CN113780401A (en) | Composite insulator fault detection method and system based on principal component analysis method | |
| CN110599458A (en) | Underground pipe network detection and evaluation cloud system based on convolutional neural network | |
| CN116453056A (en) | Target detection model construction method and transformer substation foreign matter intrusion detection method | |
| CN113724219A (en) | Building surface disease detection method and system based on convolutional neural network | |
| CN115619778A (en) | Power equipment defect identification method and system, readable storage medium and equipment | |
| CN116977907A (en) | Image recognition-based power inspection fault diagnosis method and system | |
| CN117131110A (en) | Method and system for monitoring dielectric loss of capacitive equipment based on correlation analysis | |
| CN112950576A (en) | Power transmission line defect intelligent identification method and system based on deep learning | |
| CN115035328A (en) | Converter image increment automatic machine learning system and establishment training method thereof | |
| CN112419243B (en) | Power distribution room equipment fault identification method based on infrared image analysis | |
| CN117591901B (en) | Insulator breakage detection method and device, storage medium and electronic equipment | |
| CN113012107B (en) | Power grid defect detection method and system | |
| CN116951328B (en) | Intelligent drainage pipeline operation monitoring system based on big data | |
| CN111899138A (en) | Teaching behavior analysis system and method based on big data | |
| CN116028893A (en) | Intelligent data monitoring method, system and electronic equipment | |
| CN111093140A (en) | Method, device, equipment and storage medium for detecting defects of microphone and earphone dust screen | |
| CN116363501A (en) | Foreign matter intrusion detection method and device for power transmission line and computer equipment | |
| CN115187880A (en) | Communication optical cable defect detection method and system based on image recognition and storage medium | |
| CN115880472A (en) | Intelligent diagnosis and analysis system for electric power infrared image data | |
| CN113033489A (en) | Power transmission line insulator identification and positioning method based on lightweight deep learning algorithm | |
| CN114492636A (en) | Transformer winding state signal acquisition system |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |