CN108921203B - Detection and identification method for pointer type water meter - Google Patents
Detection and identification method for pointer type water meter Download PDFInfo
- Publication number
- CN108921203B CN108921203B CN201810608129.9A CN201810608129A CN108921203B CN 108921203 B CN108921203 B CN 108921203B CN 201810608129 A CN201810608129 A CN 201810608129A CN 108921203 B CN108921203 B CN 108921203B
- Authority
- CN
- China
- Prior art keywords
- water meter
- pointer
- loss
- neural network
- image
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/02—Recognising information on displays, dials, clocks
Abstract
The invention discloses a detection and identification method of a pointer type water meter in the technical field of pattern recognition and artificial intelligence, which comprises a data acquisition terminal and an identification server, wherein the data acquisition terminal acquires images of the water meter and uploads the acquired images to the identification server, the identification server detects and identifies the reading of the water meter, and the processing steps of the identification server comprise: preprocessing data, making label of training data, constructing a deep convolution neural network, training a neural network model and the like. The invention adopts the camera to collect the water meter image, can realize the collection of the water meter data without changing the existing water meter metering equipment, has low cost and good expandability, is convenient for the access of the metering water meters with different specifications, and has better practical application value.
Description
Technical Field
The invention relates to the technical field of pattern recognition and artificial intelligence, in particular to a method for detecting and recognizing a pointer type water meter based on a deep convolutional neural network.
Background
With the development of artificial intelligence technology, especially the development of the application of the deep network model technology in the field of computer vision, the automatic detection and identification technology based on the deep network model becomes one of the current hot techniques. Especially, the recent rapid development of hardware technologies such as GPU solves the computation bottleneck of the deep network to a certain extent, so that the practical application of the deep neural network becomes possible.
Water affair data acquisition (meter reading) is a tedious and important work of water and energy operation units. At present, a manual meter reading mode is mainly adopted: on one hand, the manual meter reading mode requires great manpower and material resource investment; on the other hand, because the meter reading work relates to thousands of households, the data acquisition can not be completed in time, and a series of potential safety hazards can be brought. Another kind of data acquisition mode is to carry out the digital transformation to current water gauge, realizes data automatic acquisition, and the main problem that this mode exists is: (1) the gauge specifications are different, the gauge modification is large in related area, and a large amount of capital investment is needed; (2) the mechanical meter has the advantages that the mechanical meter can not be replaced by electronic equipment without a power supply, and is safe and reliable; (3) the novel digital meter has higher cost and higher maintenance difficulty. For example, the existing photoelectric direct-reading water meter has complex manufacturing and installation processes, and the cost is greatly improved.
The above-mentioned drawbacks are worth solving.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a detection and identification method of a pointer type water meter.
The technical scheme of the invention is as follows:
the utility model provides a detection and identification method of pointer water gauge, its characterized in that, includes data acquisition terminal and identification server, data acquisition terminal gathers the water gauge image to reach the image of gathering the identification server, identification server carries out the detection and the discernment of water gauge reading, identification server's processing step includes:
s1: data preprocessing: cutting and zooming the collected water meter image;
s2: preparation of labeled label for training data: calibrating monitoring information of a pre-collected water meter image;
s3: constructing a deep convolutional neural network: the detection and identification of the water meter adopt a model based on a deep convolutional network, and the constructed convolutional neural network completes the detection and reading identification of the pointer disk;
s4: training a neural network model: and inputting the marked water meter image and corresponding marking information into the constructed convolutional neural network model, and learning network parameters by adopting a random gradient descent algorithm.
The present invention according to the above aspect is characterized in that, in the step S1, the image scaling is 500 × 500.
The present invention according to the above aspect is characterized in that the step S2 specifically includes the following steps:
s21: marking all pointer disks in the water meter, and marking corresponding readings of the pointers as marking information of subsequent classification;
s22: storing the frame coordinates of each pointer disk and the reading of each pointer into a data file according to the reading sequence;
s23: and randomly dividing the acquired water meter image and the acquired marking data into a training set and a testing set.
Further, in step S22, the frame coordinates of the pointer disk are coordinates of two vertices of a rectangular frame, the coordinates of the top left vertex are recorded at a first position, the coordinates of the bottom right vertex are recorded at a second position, and then the reading value of the current pointer is recorded, the coordinates and the reading value are separated by commas, and the label of each pointer disk exclusively occupies a line of text.
The present invention according to the above aspect is characterized in that the step S3 specifically includes the following steps:
s31: constructing a convolutional neural network;
s32: setting training parameters;
s33: detecting a water meter pointer disc on the final convolution characteristic diagram in a sliding window mode;
s34: parameterizing the coordinates of the frame;
s35: regression;
s36: classifying;
s37: and initializing the network weight.
Further, the convolutional neural network comprises a feature extraction layer, a classifier and a regressor: the feature extraction layer is responsible for automatic learning and extraction of image features; the classifier is responsible for classifying each position and judging whether the position contains a water meter pointer disc or not; the regressor is responsible for adjusting the size and the position of the detection frame.
Further, the convolutional neural network model is:
Input(500×500)->conv1_1(3×3)->conv1_2(3×3)->pool1->conv2_1(3×3)->conv2_2(3×3)->pool2->conv3_1(3×3)->conv3_2(3×3)->conv3_3(3×3)->pool3->conv4_1(3×3)->conv4_2(3×3)->conv4_3(3×3)->pool4->conv5_1(3×3)->conv5_2(3×3)->conv5_3(3×3)->loss_cls/loss_bbox，
wherein conv represents a convolutional layer, pool represents a pooling layer, loss _ cls represents a classification loss layer, and loss _ bbox represents a regression loss layer.
Further, the training parameters include iteration number, update step size, initial learning rate, learning rate change coefficient, and weight attenuation coefficient, where the iteration number is 120000, the update step size is 50000, the initial learning rate is 0.001, the learning rate change coefficient is 0.1, and the weight attenuation coefficient is 0.0005.
The invention according to the above scheme is characterized in that in the step S4, an end-to-end training mode is adopted to train the convolutional neural network, and in the final output result, 128 results in the foreground and the background are sampled to calculate loss; when the number of foreground results is insufficient, 128 foreground results are supplemented by means of repeated sampling.
The invention according to the scheme has the advantages that:
(1) due to the adoption of the automatic learning algorithm of the deep network structure, effective image expression can be well learned from data, and the accuracy of detection and identification is improved.
(2) The invention adopts an end-to-end design, has high reading speed and higher accuracy compared with the traditional manual reading, simultaneously avoids subjective factors during manual reading, and can successfully solve the problem of reading difficulty caused by the non-intuitive pointer type water meter.
(3) The detection and identification method has the advantages of high accuracy, strong robustness and strong real-time property.
Drawings
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a flow chart of data acquisition and processing in the present invention.
Fig. 3 is a schematic structural diagram of a deep convolutional neural network in the present invention.
Fig. 4 is a flow chart of water meter detection and identification according to the present invention.
Fig. 5 is a schematic diagram of a detection and identification result of the water meter according to the embodiment of the invention.
Detailed Description
The invention is further described with reference to the following figures and embodiments:
as shown in fig. 1, a method for detecting and identifying a pointer-type water meter includes a data acquisition terminal and an identification server. The data acquisition terminal adopts a camera as an image acquisition terminal, is fixed on a water meter dial and acquires water meter images; and uploading the acquired images to an identification server in a wired or wireless mode, and detecting and identifying the reading of the water meter.
And the photographing terminal of the data acquisition terminal is fixedly arranged on the water surface plate. When the data acquisition terminal shoots the image, the camera is perpendicular to the surface of the water meter, so that the situation that the pointer is obliquely projected to wrong reading is prevented, and the shot water meter image has any rotation angle.
The processing of the recognition server comprises 4 steps of data preprocessing, label making of training data, deep convolution neural network construction, training of a neural network model and the like.
First, data preprocessing
As shown in fig. 2, the image is cropped to remove redundant areas in the photograph, and then the cropped image is scaled to 500 × 500.
Second, label making of training data
The detection and identification model adopts a supervision mode to train and learn model parameters, and needs a manual mode to calibrate supervision information of the pre-collected water meter image. The water meter images collected in the practical application are not less than 10000, and various pointer water meters are covered. The specific treatment process comprises the following steps:
1. and marking all pointer disks in the water meter by using specially designed marking software, and marking the corresponding readings of the pointers as marking information of subsequent classification.
2. And storing the frame coordinates of each pointer disk and the reading of each pointer into a data file according to the reading sequence.
Preferably, the frame coordinates of the pointer disk are two vertex coordinates of a rectangular frame, the coordinates of the top left vertex are recorded at a first position, the coordinates of the bottom right vertex are recorded at a second position, then the reading value of the current pointer is recorded, the coordinate values and the reading value are separated by commas, and the label of each pointer disk exclusively occupies a line of text.
3. And randomly dividing the collected water meter image and the label data into a training set and a testing set according to the proportion of 80% to 20%. In practical application, the training set is not less than 8000 images, and the test set is not less than 2000 images.
Thirdly, constructing a deep convolution neural network
3-4, the detection and identification of the water meter adopts a model based on a deep convolutional network, and the constructed convolutional neural network mainly completes the detection and reading identification of the pointer disk. The method specifically comprises the following steps:
1. constructing a convolutional neural network:
Input(500×500)->conv1_1(3×3)->conv1_2(3×3)->pool1->conv2_1(3×3)->conv2_2(3×3)->pool2->conv3_1(3×3)->conv3_2(3×3)->conv3_3(3×3)->pool3->conv4_1(3×3)->conv4_2(3×3)->conv4_3(3×3)->pool4->conv5_1(3×3)->conv5_2(3×3)->conv5_3(3×3)->loss_cls/loss_bbox
wherein conv represents a convolutional layer, pool represents a pooling layer, loss _ cls represents a classification loss layer, and loss _ bbox represents a regression loss layer.
The convolutional neural network comprises a feature extraction layer, a classifier and a regressor. The conv1-conv5 convolutional layers are stacked together to construct the feature extraction part of the network, and are responsible for automatic learning and extraction of image features. The classifier is realized by a convolution network with 1 multiplied by 1 convolution kernel and is responsible for classifying each position and judging whether the position contains a water meter pointer disc or not, and the classifier is optimized according to loss generated by loss _ cls. The regressor is realized by another convolution network with 1 multiplied by 1 convolution kernel and is responsible for adjusting the size and the position of the detection frame, and the regressor is optimized according to the loss generated by the loss _ bbox.
2. Setting training parameters:
iteration times iters are set according to the performance test result, and in practical application, iters is 120000. Network optimization is carried out by adopting a random gradient descent algorithm, and the following strategy is adopted for updating the learning rate: the update step size is 50000, the initial learning rate is 0.001, the learning rate change coefficient is 0.1, and weight _ decay (weight decay coefficient) is 0.0005.
The loss function uses the following formula:
wherein the content of the first and second substances,
is the molecular weight of the label,
is a regression target; p is a radical ofiIs the predicted confidence, tiIs the predicted regression result; n is a radical ofclsAnd NregIs the batch size of the classification and regression, λ is the weighted value of regression loss; l isclsAnd LregRepresenting the loss of classification and regression, respectively.
3. Detecting a water meter pointer disc:
on the final convolution characteristic diagram, a water meter pointer disc is detected in a sliding window mode, 9 anchors are preset at each sliding position, and the sizes of the anchors are 1282,2562,5122The aspect ratios are set to 1:1,1:2 and 2:1, respectively.
4. Parameterizing coordinates of the frame:
tx＝(x-xa)/wa,ty＝(y-ya)/ha
tw＝log(w/wa),th＝log(h/ha)
tx *＝(x*-xa)/wa,ty *＝(y*-ya)/ha
tw *＝log(w*/wa),th *＝log(h*/ha)
wherein, x, y, w and h represent the center coordinate and width and height of the frame; x, xaX is the central x coordinate from the prediction box, the anchor box, the ground channel box, y, w, h, respectively, similarly.
5. And (3) regression:
and according to the parameterization mode of the previous step, optimizing the network by using the smoothed L1 norm as a loss function, and finally achieving the effect of predicting the pointer coordinate of the water meter.
6. And (4) classification:
since the pointer reads 0-9, the classifier is set to 10 classifiers, corresponding to 10 readings. The respective pointer reading values are finally identified by these classifiers.
7. Initializing a network weight:
the weight of the shared convolution layer adopts Imagenet classification task data set training model, and uses the training result to initialize the weight of the shared convolution layer, and the other network layers adopt zero mean value Gaussian distribution to initialize the weight.
Training of four-step neural network model
And inputting the marked water meter image and corresponding marking information into the constructed convolutional neural network model, and learning network parameters by adopting a random gradient descent algorithm.
And training the convolutional neural network by adopting an end-to-end training mode, and sampling 128 results in the foreground and the background in the final output result to calculate loss. When the number of foreground results is insufficient, 128 foreground results are supplemented by means of repeated sampling.
As shown in fig. 5, an example of the detection and recognition result of the magic water meter is shown. The method is based on the deep learning model, realizes automatic detection and identification of the pointer type water meter, has high identification accuracy and short processing time, can be widely applied to automatic meter reading, flow monitoring and the like of water meter reading, and has better practical application value.
In addition, the invention adopts the camera to collect the water meter image, and the collection of the water meter data can be realized without changing the existing water meter metering equipment. The intelligent water meter access system has the advantages of low cost, good expandability and convenience for accessing the water meters with different specifications, thereby having better practical application value.
It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.
The invention is described above with reference to the accompanying drawings, which are illustrative, and it is obvious that the implementation of the invention is not limited in the above manner, and it is within the scope of the invention to adopt various modifications of the inventive method concept and technical solution, or to apply the inventive concept and technical solution to other fields without modification.
Claims (5)
1. The utility model provides a detection and identification method of pointer water gauge, its characterized in that, includes data acquisition terminal and identification server, data acquisition terminal gathers the water gauge image to reach the image of gathering the identification server, identification server carries out the detection and the discernment of water gauge reading, identification server's processing step includes:
s1: data preprocessing: cutting and zooming the collected water meter image;
s2: preparation of labeled label for training data: the method for calibrating the supervision information of the pre-collected water meter image specifically comprises the following steps: s21: marking all pointer disks in the water meter, and marking corresponding readings of the pointers as marking information of subsequent classification; s22: storing the frame coordinates of each pointer disk and the reading of each pointer into a data file according to the reading sequence; s23: randomly dividing the collected water meter image and the marking data into a training set and a testing set according to the proportion of 80% to 20%;
s3: constructing a deep convolutional neural network: the detection and identification of the water meter adopt a model based on a deep convolutional network, and the constructed convolutional neural network completes the detection and reading identification of the pointer disk, and the method specifically comprises the following steps: s31: constructing a convolutional neural network; s32: setting training parameters; s33: detecting a water meter pointer disc on the final convolution characteristic diagram in a sliding window mode; s34: parameterizing the coordinates of the frame; s35: regression; s36: classifying; s37: initializing a network weight; the convolutional neural network model is as follows:
Input(500×500)->conv1_1(3×3)->conv1_2(3×3)->pool1->conv2_1(3×3)->conv2_2(3×3)->pool2->conv3_1(3×3)->conv3_2(3×3)->conv3_3(3×3)->pool3->conv4_1(3×3)->conv4_2(3×3)->conv4_3(3×3)->pool4->conv5_1(3×3)->conv5_2(3×3)->conv5_3(3×3)->loss_cls/loss_bbox，
wherein conv represents a convolutional layer, pool represents a pooling layer, loss _ cls is a classification loss layer, and loss _ bbox is a regression loss layer, and the convolutional neural network comprises a feature extraction layer, a classifier and a regressor: the feature extraction layer is responsible for automatic learning and extraction of image features; the classifier is responsible for classifying each position and judging whether the position contains a water meter pointer disc or not; the regressor is responsible for adjusting the size and the position of the detection frame, the regressor is optimized according to the loss generated by the loss _ bbox, and the loss function adopts the following formula:
wherein the content of the first and second substances,
is the molecular weight of the label,
is a regression target; p is a radical ofiIs the predicted confidence, tiIs the predicted regression result; n is a radical ofclsAnd NregIs the batch size of the classification and regression, λ is the weighted value of regression loss; l isclsAnd LregLoss representing classification and regression, respectively;
s4: training a neural network model: inputting the marked water meter image and corresponding marking information into the constructed convolutional neural network model, and learning network parameters by adopting a random gradient descent algorithm;
when the data acquisition terminal shoots the image, the camera is perpendicular to the surface of the water meter, so that the situation that the pointer is obliquely projected to wrong reading is prevented, and the shot water meter image has any rotation angle.
2. The method for detecting and identifying a pointer-type water meter as recited in claim 1, wherein in said step S1, the image is scaled to 500 x 500.
3. A method for detecting and identifying a pointer-type water meter as claimed in claim 1, wherein in step S22, the frame coordinates of the pointer disk are the coordinates of two vertices of a rectangular frame, the coordinates of the top left vertex are recorded at a first position, the coordinates of the bottom right vertex are recorded at a second position, and then the reading value of the current pointer is recorded, each coordinate value and reading value are separated by a comma, and the label of each pointer disk is exclusive of a line of text.
4. The method of detecting and identifying a pointer-type water meter as recited in claim 1, wherein the training parameters include iteration number, update step length, initial learning rate, learning rate change coefficient and weight attenuation coefficient, wherein the iteration number is 120000, the update step length is 50000, the initial learning rate is 0.001, the learning rate change coefficient is 0.1, and the weight attenuation coefficient is 0.0005.
5. The method for detecting and identifying a pointer type water meter as claimed in claim 1, wherein in the step S4, an end-to-end training mode is adopted to train the convolutional neural network, and in the final output result, 128 results in the foreground and the background are sampled to calculate loss; when the number of foreground results is insufficient, 128 foreground results are supplemented by means of repeated sampling.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810608129.9A CN108921203B (en) | 2018-06-13 | 2018-06-13 | Detection and identification method for pointer type water meter |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810608129.9A CN108921203B (en) | 2018-06-13 | 2018-06-13 | Detection and identification method for pointer type water meter |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108921203A CN108921203A (en) | 2018-11-30 |
| CN108921203B true CN108921203B (en) | 2022-04-01 |
Family
ID=64419295
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810608129.9A Active CN108921203B (en) | 2018-06-13 | 2018-06-13 | Detection and identification method for pointer type water meter |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108921203B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109871754A (en) * | 2019-01-08 | 2019-06-11 | 深圳禾思众成科技有限公司 | A kind of instrument read method, equipment and computer readable storage medium |
| CN109840497A (en) * | 2019-01-30 | 2019-06-04 | 华南理工大学 | A kind of pointer-type water meter reading detection method based on deep learning |
| CN109816041A (en) * | 2019-01-31 | 2019-05-28 | 南京旷云科技有限公司 | Commodity detect camera, commodity detection method and device |
| CN110263790A (en) * | 2019-04-18 | 2019-09-20 | 汕头大学 | A kind of power plant's ammeter character locating and recognition methods based on convolutional neural networks |
| CN110188662A (en) * | 2019-05-28 | 2019-08-30 | 唐山海森电子股份有限公司 | A kind of AI intelligent identification Method of water meter number |
| CN110543880A (en) * | 2019-08-21 | 2019-12-06 | 上海小猬科技有限公司 | many pointers dial plate automatic reading system based on artificial intelligence |
| CN111062354A (en) * | 2019-12-26 | 2020-04-24 | 广州物道水务科技有限公司 | Automatic picture identification method for photographing meter reading |
| CN111950396A (en) * | 2020-07-27 | 2020-11-17 | 江苏大学 | Instrument reading neural network identification method |
| CN111738229B (en) * | 2020-08-05 | 2020-11-24 | 江西小马机器人有限公司 | Automatic reading method for scale of pointer dial |
| CN113221959A (en) * | 2021-04-20 | 2021-08-06 | 上海济辰水数字科技有限公司 | Intelligent water meter image recognition training method, intelligent water meter recognition system and intelligent water meter recognition method |
| CN113159172A (en) * | 2021-04-20 | 2021-07-23 | 上海济辰水数字科技有限公司 | Intelligent water meter image positioning training method, intelligent water meter identification system and method |
| CN113283419B (en) * | 2021-04-29 | 2022-07-05 | 国网浙江省电力有限公司湖州供电公司 | Convolutional neural network pointer instrument image reading identification method based on attention |
| CN113610161A (en) * | 2021-08-09 | 2021-11-05 | 东南数字经济发展研究院 | Target detection data labeling method based on image classification technology |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP4293257B2 (en) * | 2007-03-27 | 2009-07-08 | 株式会社デンソー | Display device |
| CN105809179B (en) * | 2014-12-31 | 2019-10-25 | 中国科学院深圳先进技术研究院 | A kind of Recognition of Reading method and device of pointer instrument |
| CN106682664A (en) * | 2016-12-07 | 2017-05-17 | 华南理工大学 | Water meter disc area detection method based on full convolution recurrent neural network |
| CN106951900B (en) * | 2017-04-13 | 2019-10-22 | 杭州申昊科技股份有限公司 | A kind of automatic identifying method of arrester meter reading |
| CN107229930B (en) * | 2017-04-28 | 2021-01-19 | 北京化工大学 | Intelligent identification method for numerical value of pointer instrument |
-
2018
- 2018-06-13 CN CN201810608129.9A patent/CN108921203B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| A coarse-to-fine approach for industrial meter detection and its application;Li Fang et al.;《2016 IEEE Workshop on Advanced Robotics and its Social Impacts (ARSO)》;20161107;214-220 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108921203A (en) | 2018-11-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108921203B (en) | Detection and identification method for pointer type water meter | |
| CN109147254B (en) | Video field fire smoke real-time detection method based on convolutional neural network | |
| CN107229930B (en) | Intelligent identification method for numerical value of pointer instrument | |
| CN109635806B (en) | Ammeter value identification method based on residual error network | |
| CN110263705A (en) | Towards two phase of remote sensing technology field high-resolution remote sensing image change detecting method | |
| CN108038846A (en) | Transmission line equipment image defect detection method and system based on multilayer convolutional neural networks | |
| CN109344786A (en) | Target identification method, device and computer readable storage medium | |
| CN104239899B (en) | A kind of power transmission line spacer recognition methods for unmanned plane inspection | |
| CN103605971B (en) | Method and device for capturing face images | |
| CN112183667A (en) | Insulator fault detection method in cooperation with deep learning | |
| CN107978110A (en) | Fence intelligence identifying system in place and recognition methods based on images match | |
| CN108921201A (en) | Dam defect identification and classification method based on feature combination with CNN | |
| CN109766873A (en) | A kind of pedestrian mixing deformable convolution recognition methods again | |
| CN111191610A (en) | People flow detection and processing method in video monitoring | |
| CN113435282A (en) | Unmanned aerial vehicle image ear recognition method based on deep learning | |
| CN115880260A (en) | Method, device and equipment for detecting base station construction and computer readable storage medium | |
| CN107194432A (en) | A kind of refrigerator door recognition methods and system based on depth convolutional neural networks | |
| CN105160285A (en) | Method and system for recognizing human body tumble automatically based on stereoscopic vision | |
| CN113888737A (en) | Hot user room temperature measuring method and system based on temperature measurement image shot by intelligent equipment | |
| CN113902792A (en) | Building height detection method and system based on improved RetinaNet network and electronic equipment | |
| CN113378672A (en) | Multi-target detection method for defects of power transmission line based on improved YOLOv3 | |
| CN112085725A (en) | Residual film residual quantity detection method and early warning system based on heuristic iterative algorithm | |
| CN109934151B (en) | Face detection method based on movidius computing chip and Yolo face | |
| CN115690614A (en) | Test method for improving embankment piping data acquisition precision | |
| CN110518881A (en) | Hot spot monitoring device and prediction technique based on environment meteorological factor |
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 |