CN114638961A - Pointer dial plate identification method, system and computer storage medium - Google Patents
Pointer dial plate identification method, system and computer storage medium Download PDFInfo
- Publication number
- CN114638961A CN114638961A CN202210312517.9A CN202210312517A CN114638961A CN 114638961 A CN114638961 A CN 114638961A CN 202210312517 A CN202210312517 A CN 202210312517A CN 114638961 A CN114638961 A CN 114638961A
- Authority
- CN
- China
- Prior art keywords
- model
- training
- image
- identification
- pointer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000012549 training Methods 0.000 claims abstract description 79
- 238000013138 pruning Methods 0.000 claims abstract description 26
- 230000008569 process Effects 0.000 claims abstract description 12
- 238000010276 construction Methods 0.000 claims abstract description 7
- 230000006872 improvement Effects 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 2
- 230000009467 reduction Effects 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 6
- 238000012986 modification Methods 0.000 description 5
- 230000004048 modification Effects 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000013136 deep learning model Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method, a system and a computer storage medium for identifying a pointer dial plate, wherein the method comprises the following steps: acquiring a pointer dial image to be identified; inputting the pointer dial image to be identified into the identification model to obtain a corresponding identification result; the construction method of the identification model comprises the following steps: acquiring a template image of a pointer dial; generating a training image according to the template image; establishing an initial model by using a pyrrch framework, wherein the basic structure of the initial model adopts ResNet 50; training an initial model by using a training image, adjusting the hyper-parameters according to a loss function of the initial model in the training process, and obtaining a training model after the training is finished; and pruning the training model to obtain the recognition model. In an industrial scene, the method has high identification accuracy and high identification speed on the pointer dial plate, and has certain improvement on data diversity and data quality.
Description
Technical Field
The present invention relates to the field of image processing technologies, and in particular, to a method and a system for identifying a pointer dial, and a computer storage medium.
Background
The pointer instrument has the advantages of simple structure, high reliability, low price and the like, and is still widely applied in the modern industrial process. At present, the numerical value of a pointer instrument is generally read manually, the method has low efficiency, long time consumption and poor real-time performance, and the reading is easily influenced by factors such as the observation angle of a meter reader, fatigue and the like. However, for historical legacy reasons, there are still some meters that require manual entry, which is labor and time intensive, and many places where manual entry is prohibited. In contrast, it is more convenient to capture the instrument image according to real-time monitoring and read the instrument value by using the image recognition algorithm.
In recent years, leading researchers have attempted to implement automatic reading of pointer gauges using computer vision techniques. The angle method and the distance method are two most commonly used automatic reading methods of the traditional pointer meter at present. The reading method adopts the traditional angle method to read, firstly finds the angle corresponding to the zero scale mark and the maximum scale mark of the pointer meter, and then obtains the reading of the pointer meter through the angle relation, and the reading accuracy is easily influenced by the inclination of the pointer meter. If the distance method is adopted for reading, the method assumes that a pointer fitting straight line is parallel to an adjacent scale mark, and the algorithm has a certain degree of error.
At present, computer vision based on deep learning is greatly developed, more and more attention is paid to the dial plate recognition problem by using a neural network model, and the real-time performance of image recognition and algorithm operation under complex conditions is remarkably improved compared with the traditional computer vision. However, the use of deep learning training algorithm models to solve such problems necessarily faces some of its inherent drawbacks: 1. the generalization of the model is insufficient. Since deep learning depends on large-scale collected data to a great extent, generalization thereof depends on diversity of data to a great extent, and if a scene actually used by the model is relatively poor in correlation with data used in training, recognition performance of the model is greatly reduced. 2. The accuracy of model identification is not high enough. Compared with the traditional computer vision technology, the accuracy of the dial pointer task based on deep learning is improved greatly, however, in the actual production life, the accuracy of the pointer dial reading is related to the efficiency of equipment, even the safety of production personnel, so the high-precision identification accuracy is very important for an algorithm model. 3. The real-time performance of the model is low. In the prior art, the trained model is larger due to more parameters and high complexity of the deep learning model, and in the field of inspection, the model is usually required to be deployed on edge embedded equipment, so that the equipment has lower computing power and lower operating power, cannot support reasoning tasks of large-scale deep models, and can meet the requirement of real-time performance.
Disclosure of Invention
The embodiment of the invention provides a pointer dial identification method, a pointer dial identification system and a computer storage medium, which are used for solving the problems of insufficient generalization, low accuracy and low real-time performance of the pointer dial identification method in the prior art.
In one aspect, an embodiment of the present invention provides a method for identifying a dial of a pointer, including:
acquiring a pointer dial image to be identified;
inputting the pointer dial image to be identified into the identification model to obtain a corresponding identification result;
the construction method of the identification model comprises the following steps:
acquiring a template image of a pointer dial;
generating a training image according to the template image;
establishing an initial model by using a pyrrch framework, wherein the basic structure of the initial model adopts ResNet 50;
training an initial model by using a training image, adjusting the hyper-parameters according to a loss function of the initial model in the training process, and obtaining a training model after the training is finished;
and pruning the training model to obtain the recognition model.
In another aspect, an embodiment of the present invention provides a pointer dial identification system, including:
the image acquisition module is used for acquiring a dial plate image of the pointer to be identified;
the identification module is used for inputting the pointer dial image to be identified into the identification model to obtain a corresponding identification result;
the construction method of the identification model comprises the following steps:
acquiring a template image of a pointer dial;
generating a training image according to the template image;
establishing an initial model by using a pyrrch frame, wherein the basic structure of the initial model is ResNet 50;
training an initial model by using a training image, adjusting the hyper-parameters according to a loss function of the initial model in the training process, and obtaining a training model after the training is finished;
and pruning the training model to obtain the recognition model.
In another aspect, an embodiment of the present invention provides a computer storage medium having a plurality of computer instructions stored therein, where the computer instructions are used to make a computer execute the method described above.
The pointer dial plate identification method, the pointer dial plate identification system and the computer storage medium have the following advantages:
in an industrial scene, the pointer dial plate recognition accuracy rate is high, the recognition speed is high, and the data diversity and the data quality are improved to a certain extent.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a flowchart of a method for identifying a dial plate of a pointer according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a template image provided by an embodiment of the invention;
FIG. 3 is a schematic diagram of a training image provided by an embodiment of the present invention;
FIG. 4 is a block structure schematic diagram before and after the improvement provided by the embodiment of the present invention;
fig. 5 is a schematic diagram of a network before and after residual connection is removed according to an embodiment of the present invention;
FIG. 6 is a schematic view of a pruning operation provided by an embodiment of the present invention;
fig. 7 is a schematic diagram of an identification result provided in the embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a flowchart of a method for identifying a dial plate of a pointer according to an embodiment of the present invention. The embodiment of the invention provides a pointer dial plate identification method, which comprises the following steps:
and S100, acquiring an image of the dial plate of the pointer to be identified.
And S110, inputting the dial plate image of the pointer to be recognized into the recognition model to obtain a corresponding recognition result.
The construction method of the identification model comprises the following steps:
and S111, acquiring a template image of the pointer dial.
In practical production life, as the types of dials vary and scales and ranges of dial pointers are different, in order to obtain an image suitable for training a model, a template image of one dial needs to be acquired first, and a dial image at each scale, i.e., a training image, can be generated according to the template image, where fig. 2 is a template image of the dial.
And S112, generating a training image according to the template image.
Exemplarily, S112 specifically includes: s1120, identifying the scale number in the template image; and S1121, generating training images with the number matched with the number of the scales according to the scales in the template image.
It can be seen from fig. 2 that the dial has a range of 1-10, which is divided into 200 small scales, so that the data generation software is used to generate more than 2000 training images, as shown in fig. 3, which need to include the dial images of all readings, and the name of each training image is the reading of the dial in the picture, i.e. the training label of the picture.
S113, establishing an initial model by using a pyrrch framework, wherein the basic structure of the initial model is ResNet 50.
Illustratively, after sufficient data is obtained, the initial model building operation can be performed. The invention uses the pytorech framework to build an initial model whose basic structure employs ResNet50, the structure of which is shown in the 50-layer column in Table 1 below.
TABLE 1 ResNet architecture
On the basis, the structure of the initial model is improved, the number proportion of blocks (blocks) in the conv2_ x to conv5_ x layers is changed from [3,4,6,3] to [3,3,9,3], then the size of a convolution kernel is changed to 4x4, and the step distance is changed from 2 to 4. Since the trade-off of resenxt is superior to resenet 50, some reference has been made to mainly use packet convolution. The guiding criteria of resenext is "divide more groups, widen the width of each layer", so the deep separable convolution (Depthwise Conv) is used directly, i.e. the number of groups equals the number of input channels. The 3x3 convolution in the hourglass configuration (bottleeck) was replaced with a depth separable convolution, and the network width was raised from 64 to 96. All activation functions in ResNet50 were changed from ReLU to GeLU, and all Batch Normalization (Batch Normalization) operations were changed to Linear Normalization (Liner Normalization). The downsampling layer of standard ResNet is typically a 3x3 convolution with a step size of 2, and for blocks with residual structure a 1x1 convolution with a step size of 2 is used in the short-circuit connection, which keeps the downsampling layer of CNN substantially similar to other layers in the computation strategy. Referring to the design of individual downsampling layers in the network structure of Swin transform, the present invention was simulated with a 2 × 2 convolution with a step size of 2. The design of the single blocks in the resulting network structure is shown in the right diagram of fig. 4. After the improvement of the initial model is completed, the initial model may be trained, and the training images generated in S112 are input into the initial model in a batch (batch) form, where the batch size (batch size) is generally set to 12, and the size may be adjusted according to the video memory of the GPU.
And S114, training the initial model by adopting a training image, adjusting the hyper-parameter according to the loss function of the initial model in the training process, and obtaining the training model after the training is finished.
Illustratively, the above-mentioned hyper-parameter is specifically a learning rate. The learning rate (learning rate) super parameter needs to be adjusted during training, and the initial learning rate can be set to 0.0001, and then amplified or reduced according to the loss function of the initial model. The number of training rounds was adjusted according to the curve of the model loss function, and was initially set to 200 rounds (epoch).
And S115, pruning the training model to obtain the recognition model.
Before pruning the training model, the method further includes: s1150, residual connection of the training model is removed.
After the training model is obtained, it needs to be further processed, i.e. pruning and fine tuning. Because the deep learning model contains a large number of parameters, the real-time performance is not ideal in practical use, in order to reduce the volume of the model (namely, reduce the parameter number of the model), pruning is needed to adjust some relatively redundant network structures of the model, two operations of Reserving (Reserving) and fusing (gathering) are adopted in the invention, residual connection is removed on the basis of ResBlock, and then a high-rate pruning operation is used to simplify the model. Compared with ResNet, the network after residual error removal is better in speed precision and friendly to high-ratio pruning operation. Assuming that the number of input channels is 4, when residual connection is removed, the same number of channels is inserted into Conv1, and the input features are preserved through a Dirac initialized convolution kernel. The Dirac initialized weights are then convolved with the concatenation to convolution kernels, which equates to the substitution residuals being concatenated. The whole operation process is shown in fig. 5. There is no residual concatenation after both the preservation and fusion operations are completed, so it is more friendly to filter pruning.
S115 specifically comprises: s1151, pruning is carried out on the training model by adopting network reduction operation, and an identification model is obtained.
The invention employs Network pruning to perform pruning operations on the training model because it is simple and efficient. Fig. 6 shows the overall process of pruning.
After pruning processing is carried out on the training model, the method further comprises the following steps: and S1152, fine adjustment is carried out on the training model after pruning.
After pruning, a fine tuning operation needs to be performed on the model, i.e. the model is retrained.
After obtaining the recognition model, the recognition model is also deployed in the required platform. Generally, the model building and training process is responsible for training and structure exploration of various SOTA (State of the art) models, and the deployment process is responsible for applying the SOTA models to the ground to enable the industry. In CV scenarios, how to achieve a fast landing of the model is crucial to enable industrial applications. The model deployment generally does not need to consider how to modify the training mode or modify the network structure to improve the model accuracy, and more, the scene and the deployment mode (central service or local terminal deployment) need to be explicitly deployed. The problem of the deployment scene mainly comes from a common mode of differential cloud deployment of a central server cloud deployment mode and an edge deployment mode, wherein a model is deployed in a cloud server, a user sends a request to the cloud server in a webpage access mode or an API (application programming interface) interface calling mode, and the cloud receives the request and then processes and returns a result. The edge deployment is mainly used for embedded equipment, the model is packaged and packaged into the SDK and is integrated into the embedded equipment, and the data processing and the model reasoning are executed on the terminal equipment. There are many deployment modes of the model, and for the above two scenarios, there are two different deployment schemes, Service deployment and SDK deployment. Service deployment: the method is mainly used for cloud deployment of the central server, and generally takes a trained engine library as an inference service mode directly. And (3) SDK deployment: the method is mainly used for an embedded end deployment scene, and a set of efficient front-back processing and reasoning engine library is realized in languages such as C + + and the like.
Taking the SDK deployment mode of the marginalized device as an example, the model is converted into an engine format file by using a model conversion library provided by NVIDIA, and the engine format file can be deployed in a jeston series embedded module of NVIDIA. After deployment is completed, an image can be input to perform a dial pointer identification task, for example, fig. 7 shows an identification result of an input picture, and the left and right corners display the pointer reading (num) and the discrimination confidence (con).
The embodiment of the invention also provides a pointer dial plate identification system, which is characterized by comprising the following components:
the image acquisition module is used for acquiring a dial plate image of the pointer to be identified;
the identification module is used for inputting the pointer dial image to be identified into the identification model to obtain a corresponding identification result;
the construction method of the identification model comprises the following steps:
acquiring a template image of a pointer dial;
generating a training image according to the template image;
establishing an initial model by using a pyrrch framework, wherein the basic structure of the initial model adopts ResNet 50;
training an initial model by using a training image, adjusting the hyper-parameters according to a loss function of the initial model in the training process, and obtaining a training model after the training is finished;
and pruning the training model to obtain the recognition model.
Embodiments of the present invention also provide a computer storage medium, in which a plurality of computer instructions are stored, and the computer instructions are used to enable a computer to execute the above method.
While preferred embodiments of the present invention 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. Therefore, it is intended that the appended claims be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the invention.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.
Claims (9)
1. A pointer dial identification method, comprising:
acquiring a pointer dial image to be identified;
inputting the pointer dial plate image to be identified into an identification model to obtain a corresponding identification result;
the construction method of the identification model comprises the following steps:
acquiring a template image of a pointer dial;
generating a training image according to the template image;
establishing an initial model by using a pyrrch framework, wherein the basic structure of the initial model adopts ResNet 50;
training the initial model by using the training image, adjusting the hyper-parameters according to the loss function of the initial model in the training process, and obtaining a training model after the training is finished;
and pruning the training model to obtain the recognition model.
2. The method of claim 1, wherein said generating a training image from said template image comprises:
identifying the number of scales in the template image;
and generating training images with the number matched with the number of the scales according to the scales in the template image.
3. The method of claim 1, wherein after the initial model is established, the structure of the initial model is further improved, and the improvement of the structure of the initial model comprises: changing the proportion of the number of blocks in the conv2_ x to conv5_ x layers to [3,3,9,3], changing the size of a convolution kernel to 4x4 and changing the step pitch to 4; the 3x3 convolution in the hourglass configuration was changed to a depth separable convolution, and the model width was changed to 96; all activation functions in ResNet50 are changed to GeLU and all batch normalization operations are changed to linear normalization.
4. The method of claim 1, further comprising, prior to pruning the training model:
removing residual connections of the training model.
5. The method according to claim 1, wherein the pruning the training model to obtain the recognition model comprises:
and pruning the training model by adopting network reduction operation to obtain the recognition model.
6. The method of claim 1, further comprising, after pruning the training model:
and carrying out fine adjustment on the training model after pruning.
7. The method of claim 1, wherein after obtaining the recognition model, the recognition model is also deployed in a desired platform.
8. A pointer dial identification system, comprising:
the image acquisition module is used for acquiring a dial plate image of the pointer to be identified;
the identification module is used for inputting the pointer dial image to be identified into an identification model to obtain a corresponding identification result;
the construction method of the identification model comprises the following steps:
acquiring a template image of a pointer dial;
generating a training image according to the template image;
establishing an initial model by using a pyrrch framework, wherein the basic structure of the initial model adopts ResNet 50;
training the initial model by using the training image, adjusting the hyper-parameters according to the loss function of the initial model in the training process, and obtaining a training model after the training is finished;
and pruning the training model to obtain the recognition model.
9. A computer storage medium having stored thereon a plurality of computer instructions for causing a computer to perform the method of any one of claims 1-7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210312517.9A CN114638961A (en) | 2022-03-28 | 2022-03-28 | Pointer dial plate identification method, system and computer storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210312517.9A CN114638961A (en) | 2022-03-28 | 2022-03-28 | Pointer dial plate identification method, system and computer storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114638961A true CN114638961A (en) | 2022-06-17 |
Family
ID=81949804
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210312517.9A Pending CN114638961A (en) | 2022-03-28 | 2022-03-28 | Pointer dial plate identification method, system and computer storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114638961A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117744057A (en) * | 2023-11-30 | 2024-03-22 | 广州熠数信息技术有限公司 | Clock image verification code identification method, system, computer equipment and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110543878A (en) * | 2019-08-07 | 2019-12-06 | 华南理工大学 | pointer instrument reading identification method based on neural network |
WO2020248495A1 (en) * | 2019-06-14 | 2020-12-17 | 平安科技(深圳)有限公司 | Model training method and apparatus, and computer-readable storage medium |
CN112232349A (en) * | 2020-09-23 | 2021-01-15 | 成都佳华物链云科技有限公司 | Model training method, image segmentation method and device |
CN112508098A (en) * | 2020-12-08 | 2021-03-16 | 南京理工大学 | Dial plate positioning and automatic reading pointer type meter value identification method and system |
CN112861867A (en) * | 2021-02-01 | 2021-05-28 | 北京大学 | Pointer type instrument panel identification method, system and storage medium |
CN113344089A (en) * | 2021-06-17 | 2021-09-03 | 北京百度网讯科技有限公司 | Model training method and device and electronic equipment |
-
2022
- 2022-03-28 CN CN202210312517.9A patent/CN114638961A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020248495A1 (en) * | 2019-06-14 | 2020-12-17 | 平安科技(深圳)有限公司 | Model training method and apparatus, and computer-readable storage medium |
CN110543878A (en) * | 2019-08-07 | 2019-12-06 | 华南理工大学 | pointer instrument reading identification method based on neural network |
CN112232349A (en) * | 2020-09-23 | 2021-01-15 | 成都佳华物链云科技有限公司 | Model training method, image segmentation method and device |
CN112508098A (en) * | 2020-12-08 | 2021-03-16 | 南京理工大学 | Dial plate positioning and automatic reading pointer type meter value identification method and system |
CN112861867A (en) * | 2021-02-01 | 2021-05-28 | 北京大学 | Pointer type instrument panel identification method, system and storage medium |
CN113344089A (en) * | 2021-06-17 | 2021-09-03 | 北京百度网讯科技有限公司 | Model training method and device and electronic equipment |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117744057A (en) * | 2023-11-30 | 2024-03-22 | 广州熠数信息技术有限公司 | Clock image verification code identification method, system, computer equipment and storage medium |
CN117744057B (en) * | 2023-11-30 | 2024-09-06 | 广州熠数信息技术有限公司 | Clock image verification code identification method, system, computer equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Li et al. | A leaf segmentation and phenotypic feature extraction framework for multiview stereo plant point clouds | |
CN111368825B (en) | Pointer positioning method based on semantic segmentation | |
CN108549873A (en) | Three-dimensional face identification method and three-dimensional face recognition system | |
CN106295613A (en) | A kind of unmanned plane target localization method and system | |
CN108182433A (en) | A kind of meter reading recognition methods and system | |
CN111259957A (en) | Visibility monitoring and model training method, device, terminal and medium based on deep learning | |
CN110287806A (en) | A kind of traffic sign recognition method based on improvement SSD network | |
CN110689118A (en) | Improved target detection method based on YOLO V3-tiny | |
CN110059765B (en) | Intelligent mineral identification and classification system and method | |
CN113129284B (en) | Appearance detection method based on 5G cloud edge cooperation and implementation system | |
CN114638961A (en) | Pointer dial plate identification method, system and computer storage medium | |
CN111159451B (en) | Power line point cloud dynamic monomer method based on spatial database | |
CN112164065A (en) | Real-time image semantic segmentation method based on lightweight convolutional neural network | |
CN117765083A (en) | Equipment positioning method and device, electronic equipment and storage medium | |
CN116721345A (en) | Morphology index nondestructive measurement method for pinus massoniana seedlings | |
CN109657907B (en) | Quality control method and device for geographical national condition monitoring data and terminal equipment | |
CN114565511B (en) | Lightweight image registration method, system and device based on global homography estimation | |
CN113256567B (en) | Banana leaf area index detection method and system | |
CN115267762A (en) | Low-altitude slow-speed small target tracking method integrating millimeter wave radar and visual sensor | |
CN112927304B (en) | Fish-eye lens calibration method based on convolutional neural network | |
CN113450364B (en) | Tree-shaped structure center line extraction method based on three-dimensional flux model | |
CN115165363A (en) | CNN-based light bearing fault diagnosis method and system | |
CN103996044B (en) | The method and apparatus that target is extracted using remote sensing images | |
CN114037993A (en) | Substation pointer instrument reading method and device, storage medium and electronic equipment | |
US20240005639A1 (en) | Instrument recognition method based on improved u2 network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
CB03 | Change of inventor or designer information |
Inventor after: Wang Hao Inventor after: Zhang Huiqi Inventor after: He Zinan Inventor before: Wang Hao Inventor before: Gao Jianwen Inventor before: He Zinan |
|
CB03 | Change of inventor or designer information |