CN112347929A - Pointer instrument system and monitoring method - Google Patents
Pointer instrument system and monitoring method Download PDFInfo
- Publication number
- CN112347929A CN112347929A CN202011231714.5A CN202011231714A CN112347929A CN 112347929 A CN112347929 A CN 112347929A CN 202011231714 A CN202011231714 A CN 202011231714A CN 112347929 A CN112347929 A CN 112347929A
- Authority
- CN
- China
- Prior art keywords
- pointer instrument
- image
- network
- picture
- 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
- 238000012544 monitoring process Methods 0.000 title claims abstract description 27
- 238000000034 method Methods 0.000 title claims description 18
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 51
- 238000010606 normalization Methods 0.000 claims abstract description 24
- 230000009466 transformation Effects 0.000 claims abstract description 22
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 12
- 238000005457 optimization Methods 0.000 claims abstract description 8
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 30
- 230000000694 effects Effects 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 238000009826 distribution Methods 0.000 claims description 6
- 238000003062 neural network model Methods 0.000 claims description 5
- 239000003086 colorant Substances 0.000 claims description 4
- 238000005286 illumination Methods 0.000 claims description 4
- 230000003993 interaction Effects 0.000 claims description 4
- 210000002569 neuron Anatomy 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 230000002093 peripheral effect Effects 0.000 claims description 3
- 238000002360 preparation method Methods 0.000 claims description 3
- 238000003860 storage Methods 0.000 claims description 3
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical group [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 claims description 2
- 238000000605 extraction Methods 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 claims description 2
- 230000008569 process Effects 0.000 claims description 2
- 230000035945 sensitivity Effects 0.000 claims description 2
- 238000011426 transformation method Methods 0.000 claims description 2
- 230000006855 networking Effects 0.000 claims 1
- 238000012423 maintenance Methods 0.000 abstract description 3
- 230000002708 enhancing effect Effects 0.000 abstract 1
- 238000005259 measurement Methods 0.000 abstract 1
- 238000012545 processing Methods 0.000 description 6
- 238000001514 detection method Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The pointer type instrument occupies the main position of an electrical power instrument for a long time due to the advantages of simplicity, reliability, stability, strong anti-interference performance, convenience in maintenance and the like; the pointer instrument is simple, but is inconvenient to carry out real-time monitoring to measurement, and the RK3399 microcomputer is used for carrying out real-time intelligent monitoring on the pointer instrument through a camera. Firstly, using a histogram normalization transformation algorithm to carry out brightness optimization on an image and preprocessing for enhancing contrast; then, YOLOv3(You Only Look one 3) was usedrd) Detecting pointer instrument areas in pictures by using a feature recognition algorithm and extracting corresponding area images independentlyAnd storing, respectively reading and predicting the numerical values of the characteristic images by using a multilayer CNN (Convolutional Neural Networks), and finally uploading to a server.
Description
Technical Field
The application relates to the field of intelligent monitoring of pointer instruments, in particular to intelligent online reading and real-time monitoring of pointer instruments.
Background
The pointer instrument has the characteristics of simplicity, reliability, stability, cold and heat resistance, strong interference resistance, convenience in maintenance and the like, is widely applied to the traditional industries such as energy, industrial production, power transmission and the like, and occupies a major position for a long time. However, the pointer type meter does not reserve any digital interface, and most of the pointer type meters are read manually. The method has the defects of low precision, poor reliability, low efficiency, long time consumption and the like, and needs to consume a large amount of human resources. Therefore, the pointer position of the pointer type instrument is converted into a digital signal through the sensor, automatic and accurate meter reading is achieved, and the constructed method has very important significance in application of the unattended substation.
In order to solve this problem, many automatic recognition numerical value reading algorithms based on computer vision have appeared in recent years, and existing pointer instrument recognition algorithms can be divided into conventional algorithms based on digital image processing technology and modern algorithms based on machine learning and deep learning.
Disclosure of Invention
The utility model provides a pointer instrument system, includes calibrated scale, activity pointer, magnet, soft iron core, controller, computer, camera, coil and spring, the computer is the microcomputer of RK3399 hardware platform, and the camera uses the ordinary industrial grade USB CAMARE of a built-in image sensor of a section, and wherein CAMARE links to each other with the computer through the USB interface, and the system passes through WIFI module or ETHERNET net gape and connects the network, supplies power through TYPE-C interface, only need connect power supply can normal operating during the use.
The system scheme uses an RK3399 microcomputer to perform real-time monitoring through a camera. Firstly, a RK3399 microcomputer acquires a pointer instrument picture through a camera, performs brightness optimization and contrast enhancement preprocessing on the picture by using a histogram normalization transformation algorithm, detects a pointer instrument area in the picture by using a YOLOv3 feature recognition algorithm, extracts a corresponding area image for independent storage, respectively performs numerical reading prediction on the feature image by using a multilayer convolutional neural network, and finally synchronously uploads the feature image to a server through a WIFI wireless network for corresponding subsequent operation.
The RK3399 processor is a six-core 64-bits CPU with a structure of a double-core Cortex-A72 large core and a four-core Cortex-A53 small core big core and a LITTLE large core, has a main frequency as high as 1.8Ghz, has the operation performance close to that of a common computer platform CPU, supports WIFI and USB peripherals, can be networked and externally connected with a camera, supports the operation of a UBUNTU system, and is convenient to deploy PYTHON, OPENCV and TENSOFLOW environments.
The computer acquires a pointer instrument picture through a camera, performs brightness optimization and contrast enhancement pretreatment on the picture by using a histogram normalization transformation algorithm, detects a pointer instrument area in the picture by using a feature recognition algorithm, extracts a corresponding area image for independent storage, performs numerical reading prediction on the feature image by using a multilayer convolutional neural network, and finally uploads the feature image to a server synchronously through a network.
The intelligent monitoring system of the instrument is composed of early-stage model training and later-stage model calling; performing model training during early preparation, namely marking pointer instrument areas by using labelImg, filling corresponding labels, and storing the labels as corresponding xml files; meanwhile, histogram normalization transformation is carried out on the image, so that the regional characteristics of the pointer instrument are more obvious; and then training the image by using a Yolov3 feature recognition algorithm to obtain a model A.
Intercepting the pointer instrument area part in the picture by using an xml tag file, adjusting the size of the pointer instrument area part, then storing the pointer instrument area part in a classified manner, and reading and storing the numerical value of the intercepted picture; performing numerical reading training on a milliampere meter by establishing a multilayer neural network model to obtain a model B, performing numerical reading training on a 250-volt range voltage meter to obtain a model C, and performing numerical reading training on a 450-volt range voltage meter to obtain a model D; when the device is used, a frame of pointer instrument picture is obtained in real time through a camera by a RK3399 microcomputer, brightness optimization and contrast enhancement pretreatment are carried out on the picture by using histogram normalization algorithm transformation, then a model A is called to detect the pointer instrument area in the picture, corresponding area images are extracted and stored separately, then a multilayer convolutional neural network is used to carry out numerical reading prediction on characteristic images respectively, and finally the characteristic images are uploaded to a server through the network.
FIG. 1 is a flowchart of the RK3399 microcomputer work;
FIG. 2 is a diagram of an instrument intelligent monitoring system;
FIG. 3 is a diagram of the YOLOv3 Tiny network architecture;
FIG. 4 is a histogram comparison before and after normalization transformation;
FIG. 5 is a graph comparing the effect of histogram normalization transform;
FIG. 6 is a graph of feature recognition results;
FIG. 7 is a reading prediction chart of a novel scale identification method;
Detailed Description
The intelligent monitoring system for the instrument designed by the method consists of early-stage model training and later-stage model calling; performing model training during early preparation, namely firstly using label images (labelImg) to mark pointer instrument areas respectively, filling corresponding labels, and storing the labels as corresponding xml files; meanwhile, histogram normalization transformation is carried out on the image, so that the regional characteristics of the pointer instrument are more obvious; and then training the image by using a Yolov3 feature recognition algorithm to obtain a model A. Then, intercepting the pointer type instrument area part in the picture by using an xml tag file, adjusting the size of the pointer type instrument area part, storing the pointer type instrument area part in a classified manner, and reading and storing the numerical value of the intercepted picture; performing numerical reading training on a milliampere meter by establishing a multilayer neural network model to obtain a model B, performing numerical reading training on a 250-volt range voltage meter to obtain a model C, and performing numerical reading training on a 450-volt range voltage meter to obtain a model D; when the device is used, a frame of pointer instrument picture is obtained in real time through a camera by a RK3399 microcomputer, brightness optimization and contrast enhancement preprocessing are performed on the picture by using histogram normalization algorithm transformation, then a model A is called to detect a pointer instrument area in the picture, corresponding area images are extracted and stored separately, numerical value reading and prediction are performed on characteristic images by using a multilayer convolution neural network, and finally corresponding follow-up operations are performed through a WIFI wireless network uploading server, wherein the specific flow is shown in fig. 1.
The instrument intelligent monitoring system designed by the text consists of a host and a camera; the host machine uses a microcomputer based on RK3399 hardware platform of Friendly Arm company, the RK3399 processor is a six-core 64-bits CPU of double-core Cortex-A72 large core + four-core Cortex-A53 small core big-small core architecture newly developed by Rockchip company, the main frequency is up to 1.8Ghz, the host machine has the operation performance close to the CPU of a common computer platform, supports WIFI and USB peripheral equipment, can be networked and externally connected with a camera, supports the operation of a UBUNTU system, and is convenient to deploy PYPYON, OPENCV and TENSORW environments; the camera uses a common industrial-grade USB CAMARE with a built-in SONY CMOS image sensor, the system is shown in figure 2, wherein the CAMARE is connected with a host through a USB interface, the system is connected with a network through a WIFI module or an ETHERNET network port and supplies power through a TYPE-C interface, and the system can normally run only by being connected with a power supply during use.
B. And (5) performing histogram normalization transformation.
In reality, places where the electrical power pointer type instruments are placed are often in special rooms, the ambient light environment condition is complex, the shooting effect quality of monitoring images is general, the contrast is low, and the image definition is poor; the intelligent monitoring system for the instrument designed by the invention uses a camera with a built-in SONY low-illumination CMOS image sensor, and because the image comprises a large number of pixels, the gray value distribution of the pixels accords with the probability statistical distribution rule, and the quality of the image can be roughly inferred according to the form of a histogram, the histogram normalization transformation pretreatment is firstly carried out on the obtained image, so that the balanced distribution of the gray of the image is realized, the contrast of the image is improved, and the image is optimizedThe image brightness has an obvious effect; let the input image I have a height h and a width w, and the minimum gray value appearing in I is denoted as IminAnd the maximum value of occurrence is denoted as ImaxIn order to make the gray scale range of the output image O be [ O ]min,Omax]We calculate using the following equation (1).
Equation (1) can be converted to:
wherein r is not less than 0<h,0≤c<w. The above process is called histogram normalization transform because
So O (r, c) is E [ Omin,Omax]Generally let Omin＝0，Omax255. Histogram normalization transform is a linear transformation method for automatically selecting a weight α and a deviation variable β, where α and β are shown in formula (3):
Omin＝α*I(r,c)+β (3)
formulae (4) to (5) can thus be obtained:
C. graying transformation of instrument dial
When the common pointer instrument dial is designed, the appearance color is single, and is not black or white, and in order to reduce the calculation amount, the captured picture is subjected to gray level conversion; according to the importance of the three primary colors and the sensitivity of human eyes to different colors, weighted averaging is carried out on the RGB components by different weights, and weighted averaging is carried out on the RGB components according to a formula (6) to obtain the gray level image of the instrument area.
f(i,j)＝0.299*R(i,j)+0.587*G(i,j)+0.114*B(i,j) (6)
Yolov3 feature recognition algorithm
The standard network structure of the YOLOv3 algorithm has 107 convolutional layers, wherein the first 74 convolutional layers are based on a darknet-53 network layer and are used as a main network structure for feature extraction, and 75-107 layers are feature interaction layers of the YOLO network, and local feature interaction is realized in a convolution kernel mode; however, because the target operation platform is an embedded platform with weak computing capability and the identification characteristics of the pointer instrument are obvious, a simplified version YOLOv3 Tiny of a YOLO network structure is used as the network structure; the Yolov3 Tiny network structure is shown in FIG. 3, where CONV is convolution operation, POOL is pooling operation, and UPSAMPLE is upsampling operation; in order to reduce the amount of calculation, the Input layer takes the RGB picture with 640 width and 640 height as the Input picture, and the YOLO layer as the output layer; although the YOLOv3 Tiny network only reserves 24 convolution layers, 2 YOLO network layers are still reserved, and the accuracy of model identification is still guaranteed on the premise of greatly reducing the operation amount.
E. Convolutional neural network feature identification
The convolutional neural network is a feedforward type neural network, has excellent performance in large-scale image processing, and is widely used in the fields of image classification, positioning and the like at present; the picture data of the pointer instrument is nonlinear separable data, and for processing the data, the intelligent monitoring system of the instrument designed by the method introduces a multilayer convolutional neural network; the multi-layer neural network performs classification by mapping the original data into a linearly separable high-dimensional space and then using a specific linear classifier. The test determines that the three-layer neural network model comprising an input layer, a hidden layer and an output layer is used for carrying out numerical reading training on the pointer instrument picture data, and the basic formula is as shown in formula (7).
If input x has m nodes, output fi(x) With n nodes, then the weight vector W is an n × m matrix, the input x is a vector of length m, the offset vector b is a vector of length n, act is the activation function, f (x) returns an n-dimensional vector.
The input layer-hidden layer is a fully connected network, that is, each input node is connected to all hidden layer nodes, that is, each hidden layer node is equivalent to a neuron and generates an output, so that the outputs of all nodes of the hidden layer become a vector; if the input layer is considered as a vector x, the hidden layer node h has a weight vector WhAnd an offset bhAnd the activation function uses the tanh function, the output formula of the hidden layer node is shown as formula (8).
The hidden layer-output layer is also a fully connected network and can be regarded as a neuron cascaded on the hidden layer, because of multi-classification, the activation function adopts a Softmax Regression function, and the output formula of the node of the output layer is as shown in a formula (9).
The paper adopts a specific network structure of a convolutional neural network as shown in table 1, wherein CONV is a convolutional layer, POOL is a pooling layer, FC is a fully connected dense layers (dense connected density layers), and an input picture is a gray image with a size of 80 × 80.
TABLE 1 CNN architecture
In the experiment, 3745 pointer instrument RGB color images are collected as a data set in total, the original size of the images is 1920-1080 in length, 374 images are randomly selected as verification data in the data set, and the rest images are used as training data; in order to ensure the final training effect, the data set comprises pictures of different conditions, such as different inclination angles, different rotation angles, different illumination degrees, different degrees of shading on the images and the like.
Transplanting a system model:
the system uses a RK3399 hardware platform as a bearing platform, a UBUNTU 18.04 operating system is installed on the hardware platform, PYTHON3.6, TENSORFLOW 1.14 and OPENCV2.0 operating environments are deployed, the following table 2 shows that the same operating environment is deployed on different platforms, and the comparison of the time length required by each algorithm required by the same photo operating system is carried out, so that although the operation performance of the RK3399 hardware platform is slightly inferior to that of a common computer platform, the operation requirement of the system is completely met, the integration level is higher, the size is far smaller than that of the common computer, the portability and the flexibility of the system are greatly improved, and the power consumption and the cost are also superior to those of the common PC platform.
TABLE 2 comparison of algorithms of the System at run times of different platforms
Histogram normalization conversion
Histogram normalization transformation is carried out on the image, the image brightness is automatically adjusted, and the contrast is enhanced, so that the regional characteristics of the pointer instrument are more obvious; fig. 4 is a comparison of image histograms before and after histogram normalization transformation, where the horizontal axis x represents gray scale and the vertical axis y represents the number of pixel elements per gray scale, with the unit being 105 pixs; the image processing effect before and after actual transformation is shown in fig. 5, the left side is an image before processing, and the right side is an effect image after processing, so that the image brightness is obviously optimized, and meanwhile, the contrast is increased, so that some details in a darker area are clearer.
The histogram normalization transformation of the pointer instrument image not only obviously optimizes the image display effect, but also improves the identification accuracy of a target identification algorithm because the regional characteristics of the pointer instrument are more obvious after transformation, and the table 3 is the actual comparison condition of the identification accuracy when the models are called before and after the histogram normalization transformation is used for the whole data set for prediction; the experimental result shows that the accuracy of feature recognition of the instrument image after transformation is increased.
Pointer instrument identification training
The common feature recognition algorithm is divided into two types, one type needs to generate a candidate region firstly, then carries out classification and position coordinate prediction on the RoI, the algorithm is called as a two-stage feature recognition algorithm, and the common Mask RCNN algorithm is just the algorithm; another single-stage detection algorithm which can generate the RoI and predict the object type and position coordinates simultaneously only by one network; the meter intelligent monitoring system designed in the invention adopts a YOLOv3 feature recognition algorithm to belong to the latter; compared with a two-stage feature recognition algorithm such as Mask RCNN, the YOLOv3 algorithm uses a single network structure, the object type and position can be predicted while the candidate region is generated, the detection task is completed without being divided into two stages, meanwhile, each real frame of the YOLOv3 algorithm only corresponds to a correct candidate region, and due to the characteristics, the YOLOv3 algorithm has less computation and higher detection speed, and is more suitable for being transplanted to an embedded system operation platform such as a microcomputer with general operation performance. The target image is called YOLOv3 to perform target prediction, and predicted values of target areas of the pointer meters are obtained, and examples of feature recognition results are shown in fig. 6.
In order to test the performance superiority of the algorithm, two-stage feature recognition algorithms of fast RCNN and Mask RCNN are transplanted on a RK3399 hardware platform; in order to reduce the amount of computation as much as possible, fast RCNN uses Vgg16 as a feature extractor (as the feature extractor), and Mask RCNN uses ResNet-18 as a feature extractor (as the feature extractor); in contrast to YOLOv3 algorithm using YOLOv3 tiny network structure, The whole volume data set is tested and averaged, and The actual test effect is shown in table x, where The training time (The training time) is The number of hours it takes to model train a total of 3743 pictures on The same server platform. As can be seen from The table, compared with Fast RCNN algorithm of Vgg16 network structure, The operation time (The operation time) of YOLOv3 algorithm using YOLOv3 tiny network structure is shortened by 1.7 seconds, The recognition accuracy is improved by 0.67%, and The hours consumed by model training is shortened by 19.6 hours; compared with a Mask RCNN algorithm of a ResNet-18 network structure, The operation time (The operation time) is shortened by 3.4 seconds, The recognition accuracy is improved by 0.46 percent, and The hours consumed by model training is shortened by 30.2 hours. The experimental result shows that in comparison with the Master RCNN algorithm and the fast RCNN algorithm, the YOLOv3 algorithm is slightly improved in recognition accuracy rate, is also superior in model calling time, and is more suitable for being transplanted to an embedded platform; and the time length consumed by training the model is greatly shortened, the model training is more advantageous, and the actual feature recognition result is shown in fig. 6.
Table 3 comparison of our algorithm data with other algorithm data
Pointer type ammeter reading training
Compared with the traditional machine vision algorithm, the multilayer neural network has the advantages of high universality, wide application range, high prediction accuracy and the like in recognition application, and particularly can highlight the advantages in scenes with complex environmental illumination conditions; before the multilayer neural network is used, reading prediction is performed by using other target recognition algorithms, but the recognition accuracy is not high in some occasions with complex light rays, or when the inclination angle of an image is changed, no matter the image rotates left and right or inclines forwards and backwards by a small angle, the recognition accuracy is gradually reduced, and the algorithm has high requirements on the resolution ratio of the image; table 4 is a comparison of the verification results of reading prediction with reference to other data sets, and the actual effect is shown in fig. 7; as can be seen from the table, the algorithm used in the present application is time-consuming but the accuracy is significantly increased compared to other algorithms.
TABLE 4 comparison of reading predictions
The intelligent monitoring system for the instrument designed in the invention has the advantages that the recognition rate of the pointer instrument panel reaches 97.71%, the numerical value reading accuracy rate reaches 96.2%, the area of the pointer instrument can be accurately positioned, the corresponding numerical value reading indication is carried out, and the requirement on real-time detection of the pointer instrument is met; the identification result is synchronously uploaded to the server, and is intelligently read on line and monitored, so that the metering problem existing on the site can be conveniently found in real time, the emergency can be timely dealt with, and meanwhile, the historical data can be conveniently collected and analyzed for subsequent operations such as system optimization and the like; the system has simple running condition, small volume and simple use, does not need to modify the pointer instrument, has the advantages of simplicity, reliability, stability, strong anti-interference performance, convenient maintenance and the like of the pointer instrument, and has higher feasibility and practical value.
Although the present invention has been described with reference to the preferred embodiments, it is not intended to limit the present invention, and those skilled in the art can make possible variations and modifications of the present invention without departing from the spirit and scope of the present invention by using the methods and technical contents disclosed above, and therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention shall fall within the protection scope of the present invention.
Claims (10)
1. A pointer instrument system characterized in that: the intelligent electronic device comprises a dial, a movable pointer, a magnet, a soft iron core, a controller, a computer, a camera, a coil and a spring, wherein the computer is a microcomputer with a RK3399 hardware platform, the camera uses a common industrial USB camera with a built-in image sensor, the camera is connected with the computer through a USB interface, the system is connected with a network through a WIFI module or an ETHERNET network port, power is supplied through a TYPE-C interface, and the intelligent electronic device can normally run only by being connected with a power supply during use.
2. The pointer instrument system of claim 1, wherein: the RK3399 processor is a six-core 64-bits CPU with a structure of a double-core Cortex-A72 large core and a four-core Cortex-A53 small core big core and a LITTLE large core, the main frequency is up to 1.8Ghz, WIFI and USB peripherals are supported, networking and external camera can be realized, a UBUNTU system is supported to run, and PYTHON, OPENCV and TENSOFLOW environments are convenient to deploy.
3. A pointer instrument monitoring method is characterized in that: the computer acquires a pointer instrument picture through a camera, performs brightness optimization and contrast enhancement pretreatment on the picture by using a histogram normalization transformation algorithm, detects a pointer instrument area in the picture by using a feature recognition algorithm, extracts a corresponding area image for independent storage, performs numerical reading prediction on the feature image by using a multilayer convolutional neural network, and finally uploads the feature image to a server synchronously through a network.
4. The pointer instrument monitoring method of claim 3, wherein: the intelligent monitoring system of the instrument is composed of early-stage model training and later-stage model calling; performing model training during early preparation, namely marking pointer instrument areas by using labelImg, filling corresponding labels, and storing the labels as corresponding xml files; meanwhile, histogram normalization transformation is carried out on the image, so that the regional characteristics of the pointer instrument are more obvious; and then training the image by using a Yolov3 feature recognition algorithm to obtain a model A.
5. The pointer instrument monitoring method of claim 4, wherein: intercepting the pointer instrument area part in the picture by using an xml tag file, adjusting the size of the pointer instrument area part, then storing the pointer instrument area part in a classified manner, and reading and storing the numerical value of the intercepted picture; performing numerical reading training on a milliampere meter by establishing a multilayer neural network model to obtain a model B, performing numerical reading training on a 250-volt range voltage meter to obtain a model C, and performing numerical reading training on a 450-volt range voltage meter to obtain a model D; when the device is used, a frame of pointer instrument picture is obtained in real time through a camera by a RK3399 microcomputer, brightness optimization and contrast enhancement pretreatment are carried out on the picture by using histogram normalization algorithm transformation, then a model A is called to detect the pointer instrument area in the picture, corresponding area images are extracted and stored separately, then a multilayer convolutional neural network is used to carry out numerical reading prediction on characteristic images respectively, and finally the characteristic images are uploaded to a server through the network.
6. The pointer instrument monitoring method of claim 5, wherein: the intelligent monitoring system of the instrument uses a camera with a built-in low-illumination CMOS image sensor, and because an image contains a large number of pixels, the gray value distribution of the pixels accords with the probability statistical distribution rule, and the quality of the image can be roughly inferred according to the form of a histogram, the acquired image is subjected to histogram normalization transformation preprocessing firstly, so that the balanced distribution of the gray of the image is realized, and the intelligent monitoring system has obvious effects on improving the contrast of the image and optimizing the brightness of the image; let the input image I have a height h and a width w, and the minimum gray value appearing in I is denoted as IminAnd the maximum value of occurrence is denoted as ImaxIn order to make the gray scale range of the output image O be [ O ]min,Omax]We calculate using the following equation (1):
equation (1) can be converted to:
wherein r is not less than 0<h,0≤c<w. The above process is called histogram normalization transform because
So O (r, c) is E [ Omin,Omax]Generally let Omin＝0，Omax255. Histogram normalization transform is a linear transformation method for automatically selecting a weight α and a deviation variable β, where α and β are shown in formula (3):
Omin＝α*I(r,c)+β (3)
formulae (4) to (5) can thus be obtained:
7. the pointer instrument monitoring method of claim 6, wherein: according to the importance of three primary colors and the sensitivity of human eyes to different colors, the three components of RGB are weighted and averaged by different weights, the three components of RGB are weighted and averaged according to the following formula (6) to obtain the gray level image of the instrument area,
f(i,j)＝0.299*R(i,j)+0.587*G(i,j)+0.114*B(i,j) (6)。
8. the pointer instrument monitoring method of claim 7, wherein: the standard network structure of the YOLOv3 algorithm has 107 convolutional layers, wherein the first 74 convolutional layers are based on a darknet-53 network layer and are used as a main network structure for feature extraction, and 75-107 layers are feature interaction layers of the YOLO network, and local feature interaction is realized in a convolution kernel mode; however, because the target operation platform is an embedded platform with weak computing capability and the identification characteristics of the pointer instrument are obvious, a simplified version YOLOv3 Tiny of a YOLO network structure is used as the network structure; in order to reduce the amount of calculation, the Input layer takes the RGB picture with 640 width and 640 height as the Input picture, and the YOLO layer as the output layer; although the YOLOv3 Tiny network only reserves 24 convolution layers, 2 YOLO network layers are still reserved, and the accuracy of model identification is still guaranteed on the premise of greatly reducing the operation amount.
9. The pointer instrument monitoring method of claim 8, wherein: the multilayer neural network maps the original data into a linearly separable high-dimensional space and finishes classification by using a specific linear classifier; the test determines that the three-layer neural network model comprising an input layer, a hidden layer and an output layer is used for carrying out numerical reading training on the pointer instrument picture data, and the basic formula is as the formula (7):
if input x has m nodes, output fi(x) With n nodes, then the weight vector W is an n × m matrix, the input x is a vector of length m, the offset vector b is a vector of length n, act is the activation function, f (x) returns an n-dimensional vector.
10. The pointer instrument monitoring method of claim 9, wherein: the input layer-hidden layer is a fully connected network, that is, each input node is connected to all hidden layer nodes, that is, each hidden layer node is equivalent to a neuron and generates an output, so that the outputs of all nodes of the hidden layer become a vector; if the input layer is considered as a vector x, the hidden layer node h has a weight vector WhAnd an offset bhAnd the activation function uses a tanh function, the output formula of the hidden layer node is as shown in formula (8):
the hidden layer-output layer is also a fully connected network, and can be regarded as a neuron cascaded on the hidden layer, because of multi-classification, the activation function adopts a Softmax Regression function, and the output formula of the node of the output layer is as shown in formula (9):
the input picture is a grayscale image with dimensions 80 x 80.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011231714.5A CN112347929A (en) | 2020-11-06 | 2020-11-06 | Pointer instrument system and monitoring method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011231714.5A CN112347929A (en) | 2020-11-06 | 2020-11-06 | Pointer instrument system and monitoring method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112347929A true CN112347929A (en) | 2021-02-09 |
Family
ID=74429905
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011231714.5A Pending CN112347929A (en) | 2020-11-06 | 2020-11-06 | Pointer instrument system and monitoring method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112347929A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113657385A (en) * | 2021-10-20 | 2021-11-16 | 山东摄云信息技术有限公司 | Data detection method and device of electronic metering device and electronic equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101593425A (en) * | 2009-05-06 | 2009-12-02 | 深圳市汉华安道科技有限责任公司 | A kind of fatigue driving monitoring method and system based on machine vision |
| CN105571608A (en) * | 2015-12-22 | 2016-05-11 | 苏州佳世达光电有限公司 | Navigation system, vehicle and navigation map transmission method |
| CN107229930A (en) * | 2017-04-28 | 2017-10-03 | 北京化工大学 | A kind of pointer instrument numerical value intelligent identification Method and device |
| CN109815950A (en) * | 2018-12-28 | 2019-05-28 | 汕头大学 | A kind of reinforcing bar end face recognition methods based on depth convolutional neural networks |
| CN110929723A (en) * | 2019-11-20 | 2020-03-27 | 汕头大学 | Identification method of transformer substation pointer instrument based on convolutional neural network |
| CN111461121A (en) * | 2020-05-18 | 2020-07-28 | 江苏电力信息技术有限公司 | Electric meter number identification method based on YO L OV3 network |
-
2020
- 2020-11-06 CN CN202011231714.5A patent/CN112347929A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101593425A (en) * | 2009-05-06 | 2009-12-02 | 深圳市汉华安道科技有限责任公司 | A kind of fatigue driving monitoring method and system based on machine vision |
| CN105571608A (en) * | 2015-12-22 | 2016-05-11 | 苏州佳世达光电有限公司 | Navigation system, vehicle and navigation map transmission method |
| CN107229930A (en) * | 2017-04-28 | 2017-10-03 | 北京化工大学 | A kind of pointer instrument numerical value intelligent identification Method and device |
| CN109815950A (en) * | 2018-12-28 | 2019-05-28 | 汕头大学 | A kind of reinforcing bar end face recognition methods based on depth convolutional neural networks |
| CN110929723A (en) * | 2019-11-20 | 2020-03-27 | 汕头大学 | Identification method of transformer substation pointer instrument based on convolutional neural network |
| CN111461121A (en) * | 2020-05-18 | 2020-07-28 | 江苏电力信息技术有限公司 | Electric meter number identification method based on YO L OV3 network |
Non-Patent Citations (7)
| Title |
|---|
| KAI WEN等: "Lightning Arrester Monitor Pointer Meter and Digits Reading Recognition Based on Image Processing", 《2018 IEEE 3RD ADVANCED INFORMATION TECHNOLOGY, ELECTRONIC AND AUTOMATION CONTROL CONFERENCE(IAEAC 2018)》, pages 759 - 764 * |
| YUE LIN等: "A Pointer Type Instrument Intelligent Reading System Design Based on Convolutional Neural Networks", 《FRONTIERS IN PHYSICS》, vol. 8, pages 1 - 9 * |
| 周勇等: "核电厂应急柴油机表盘数字化识别技术研究", 《自动化与仪表》, vol. 33, no. 12, pages 1 * |
| 我要飞升: "图像增强：直方图正规化、直方图均衡 (python实现)", pages 1 - 3, Retrieved from the Internet <URL:《 https://blog.csdn.net/chris_xy/article/details/93225309》> * |
| 我要飞升: "图像增强：直方图正规化、直方图均衡(python 实现)", pages 1 - 3, Retrieved from the Internet <URL:《https://blog.csdn.net/chris_xy/article/details/93225309》> * |
| 荪荪: "基于电力行业的智能读表系统--基于RK3399 嵌入式设备部署", pages 1, Retrieved from the Internet <URL:《https://blog.csdn.net/SMF0504/article/details/109456604》> * |
| 荪荪: "基于电力行业的智能读表系统--基于RK3399嵌入式设备部署", pages 1 - 2, Retrieved from the Internet <URL:《https://blog.csdn.net/SMF0504/article/details/109456604》> * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113657385A (en) * | 2021-10-20 | 2021-11-16 | 山东摄云信息技术有限公司 | Data detection method and device of electronic metering device and electronic equipment |
| CN113657385B (en) * | 2021-10-20 | 2022-01-25 | 山东摄云信息技术有限公司 | Data detection method and device of electronic metering device and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110021033B (en) | Target tracking method based on pyramid twin network | |
| CN111652852B (en) | Product surface defect detection method, device and equipment | |
| CN104050443B (en) | It is pre-processed using the posture of the video flowing of Face Detection | |
| CN112163496B (en) | Embedded terminal reservoir water level early warning method based on semantic segmentation | |
| CN109961097B (en) | Image classification scheduling method based on edge calculation in embedded scene | |
| CN111444939A (en) | Small-scale equipment component detection method based on weak supervision cooperative learning in open scene of power field | |
| CN110400315A (en) | A kind of defect inspection method, apparatus and system | |
| CN116448019B (en) | Intelligent detection device and method for quality flatness of building energy-saving engineering | |
| CN112040198A (en) | Intelligent water meter reading identification system and method based on image processing | |
| CN114677339A (en) | Power transmission line bolt out-of-stock defect detection method introducing attention mechanism | |
| CN112347929A (en) | Pointer instrument system and monitoring method | |
| CN114627461A (en) | Method and system for high-precision identification of water gauge data based on artificial intelligence | |
| Yanmin et al. | Research on ear recognition based on SSD_MobileNet_v1 network | |
| CN112699824B (en) | Method and device for detecting constant of electric energy meter and storage medium | |
| CN112906602B (en) | Automatic identification device and identification method for electric quantity meter of power distribution cabinet based on image processing | |
| CN108073873A (en) | Human face detection and tracing system based on high-definition intelligent video camera | |
| Chen et al. | Port ship detection in complex environments | |
| CN109034033B (en) | Smoke discharge video detection method based on improved VGG16 convolutional network | |
| CN117173223A (en) | Standard template generation method, device, equipment and medium for ammeter code-breaking screen | |
| CN111950500A (en) | Real-time pedestrian detection method based on improved YOLOv3-tiny in factory environment | |
| CN110930393A (en) | Chip material pipe counting method, device and system based on machine vision | |
| CN109740509A (en) | A kind of method of the video human face detection improving stability of improvement | |
| CN112435296B (en) | Image matching method for VSLAM indoor high-precision positioning | |
| JP7060107B2 (en) | Environment similarity display device, environment similarity display method and environment similarity display algorithm | |
| Jiang et al. | Remote Sensing Image Target Recognition System of Tennis Sports based on CAMSHIFT Algorithm |
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 |