CN116645329A - Abnormality monitoring method for instrument and meter cabinet - Google Patents
Abnormality monitoring method for instrument and meter cabinet Download PDFInfo
- Publication number
- CN116645329A CN116645329A CN202310460460.1A CN202310460460A CN116645329A CN 116645329 A CN116645329 A CN 116645329A CN 202310460460 A CN202310460460 A CN 202310460460A CN 116645329 A CN116645329 A CN 116645329A
- Authority
- CN
- China
- Prior art keywords
- image
- network
- abnormal
- data
- 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 74
- 230000005856 abnormality Effects 0.000 title claims abstract description 33
- 238000012544 monitoring process Methods 0.000 title claims abstract description 24
- 230000002159 abnormal effect Effects 0.000 claims abstract description 117
- 238000012549 training Methods 0.000 claims description 64
- 230000007547 defect Effects 0.000 claims description 43
- 238000013507 mapping Methods 0.000 claims description 23
- 230000008859 change Effects 0.000 claims description 19
- 101100268665 Caenorhabditis elegans acc-1 gene Proteins 0.000 claims description 15
- 238000001514 detection method Methods 0.000 claims description 15
- 235000009508 confectionery Nutrition 0.000 claims description 13
- 238000013135 deep learning Methods 0.000 claims description 12
- 101100268668 Caenorhabditis elegans acc-2 gene Proteins 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 8
- 238000004458 analytical method Methods 0.000 claims description 7
- 101100083337 Schizosaccharomyces pombe (strain 972 / ATCC 24843) pic1 gene Proteins 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000003672 processing method Methods 0.000 claims description 6
- 230000009467 reduction Effects 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 5
- 238000013527 convolutional neural network Methods 0.000 claims description 4
- 238000009825 accumulation Methods 0.000 claims description 3
- 230000003321 amplification Effects 0.000 claims description 3
- 230000002950 deficient Effects 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 230000000873 masking effect Effects 0.000 claims description 3
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 3
- 101150034941 AURKB gene Proteins 0.000 claims description 2
- 238000012423 maintenance Methods 0.000 abstract description 3
- 238000003745 diagnosis Methods 0.000 abstract description 2
- 238000012795 verification Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
- G06T1/0042—Fragile watermarking, e.g. so as to detect tampering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2201/00—General purpose image data processing
- G06T2201/005—Image watermarking
- G06T2201/0065—Extraction of an embedded watermark; Reliable detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30232—Surveillance
-
- 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
- G06V2201/06—Recognition of objects for industrial automation
-
- 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
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an abnormality monitoring method of an instrument cabinet, which comprises the steps of collecting instrument cabinet data to obtain an extended image data set, uniformly marking target image data and position signals in the extended image data set to form a marking file to form a targetDetecting a network, storing mAP values and weight files of the network, reading real-time video, and calling network weight w 1 And selecting the target frame, and determining the abnormal type, coordinates of the abnormal image and confidence information according to the data result. The invention not only can realize the monitoring and fault diagnosis of equipment and provide convenience for remote operation and maintenance protection, but also can provide basis and premise for future abnormal trend and even abnormal prediction by forming data characteristics for a long time.
Description
Technical Field
The invention belongs to the field of control system monitoring, and particularly relates to an abnormality monitoring method of an instrument cabinet.
Background
The instrument cabinet is used for equipment installation and control in many fields, and is one of central control elements indispensable to systems such as subways, electric power, railways, buildings and the like, so that the monitoring of faults of the instrument cabinet is also a necessary requirement. At present, most of intelligent technical means adopted by many systems only collect and analyze data of a sensor, such as obtaining a numerical value of the instrument, alarming when the numerical value of the instrument is too high, or alarming when an indicator lamp fault is detected, but continuous analysis for a continuous period of time is lacking, an analysis model and an analysis method are also lacking, rules are further summarized for the continuous analysis period of time, and summarized means are also lacking, so that the intelligent means are one of the important technical means for avoiding on-site operation and maintenance for a long time, and the intelligent means cannot really meet the requirement of replacing manual work and cannot replace the original manual work by a machine.
Meanwhile, the fault conditions of equipment are different from equipment to equipment, most of equipment fault rates are not very high, the prior art is mainly periodic, short-term and fixed-point detection, resource waste is easily caused by repeated inspection of mass healthy cabinets, and abnormal conditions of the cabinets cannot be analyzed and predicted.
Disclosure of Invention
In order to solve the problems, the invention provides an abnormality monitoring method for an instrument cabinet, which not only can identify the defects of the instrument cabinet, but also can intelligently analyze the abnormality type, thereby achieving the aim of safety protection.
In order to achieve the technical purpose, the invention adopts the following technical scheme:
an anomaly monitoring method of an instrument cabinet comprises the following steps:
(1) Collecting instrument cabinet data and obtaining an extended image dataset;
(2) Uniformly marking target image data and position signals in the extended image data set to form a marking file, forming a target detection network, storing mAP values and weight files of the network, not adjusting the architecture and model of the network when the mAP values reach the availability network, and storing the network weight w 1 ;
(3) Reading real-time video and calling network weight w 1 And selecting the target frame, and determining the abnormal type, coordinates of the abnormal image and confidence information according to the data result.
The method for determining the anomaly type comprises the following steps:
the target position i is expressed as f (i) = (X) i ,Y i ,W i ,H i ,c i ,p i ) Wherein X is i For the abscissa of the target position, Y i For the ordinate of the target position, W i Is of width, H i Is of height, c i Class labels for targets, p i Probability of being considered to be the target by the systemThe method comprises the steps of carrying out a first treatment on the surface of the When the aspect ratio W i /H i Greater than threshold beta 1 Calling a digital identification network to obtain a numerical result, and judging whether an abnormality exists according to the numerical value; the step of the digital identification network judging numerical value result is as follows: the individual digital figures are collected to form a data set, a digital identification network is trained and the weights w thereof are saved 2 Graying and gray projection are carried out on the image to be detected, the processed image is segmented into n pieces, and when i is less than or equal to n, w is called 2 Obtaining the numbers of n pictures, and recombining to obtain a numerical result;
aspect ratio W i /H i Not greater than threshold beta 1 And when the image is in an abnormal state, the image data is input into an abnormal judgment network according to the area, and whether the image is in the abnormal state or not is obtained, wherein the image comprises a pointer, a button, an indicator light, a handle and a defect.
The algorithm of the anomaly judgment network adopts a single training network method, a fusion training network method or an image processing method;
the individual training network method comprises the steps of: respectively collecting and training a pointer, a button, an indicator lamp, a handle and a normal map and an abnormal map of a defect to form a pointer network, a signal lamp network and a defect network, carrying out data enhancement, calculating the weight of each network, calling different network weights according to an image form, and obtaining whether the network is abnormal or not, wherein the specific identification method comprises the following steps: obtaining normal and abnormal samples, building a network model, and storing the weight of the network model when the accuracy of the network model reaches more than 99%; detecting a video image in real time, calling the weight of a stored network model after obtaining image data, and judging whether the image data is abnormal or not;
the fusion training network method comprises the following steps: forming normal and abnormal figures of the pointer, the button, the indicator light, the handle and the defect into a comprehensive data set, carrying out data enhancement, corresponding to the normal label and the abnormal label, training a network to store weights, and calling the weights to the target position image to obtain whether the target position image is abnormal or not;
the image processing method includes the steps of:
preprocessing and thresholding images of the pointer, the handle and the button, highlighting the height of the button, fitting the slopes of the pointer and the handle, and judging whether the slopes of the real-time pointer and the handle images or the height of the button are abnormal or not;
performing color analysis on the signal lamp image, and judging whether the signal lamp image is abnormal or not according to the real-time signal lamp color;
preprocessing a defect image, calculating the defect area, and judging whether the defect is a defect or not according to the defect area of the real-time defect image;
the image preprocessing comprises graying and binarization.
The specific steps of the anomaly judgment network training are as follows:
the rest of the original background removing target is marked as Q, the obtained abnormal data image is respectively processed by 11 methods of rotation, brightness increase, brightness reduction, filtering, noise addition, clipping, blurring, masking processing, binarization, LBP change and gabor change, the processed 11 result images are respectively input into n abnormal judging models, each model outputs 11 results which are considered to be abnormal, when more than n/2 models are considered to be abnormal, the data image is considered to be abnormal data, the results of the 11 abnormal data are combined with a plurality of original background images, and the data image is expanded to an image data set as a new sample.
The method for acquiring the abnormality of the extended image data set comprises the following steps:
detecting original data in real time, if abnormality exists, taking the abnormal image as one of accumulation numbers, when certain data is accumulated, putting the data into a discrimination network, determining that the accumulated data are all abnormal data, and recording the abnormal data as B; assuming that the original sample library data set is A, the initial index is the accuracy a obtained through training, B is uploaded and expanded to the original data set to form a new data set A+B, training is carried out on the data set A+ B, A, B, the obtained accuracy values are respectively marked as a1, a2 and a3, if a1 and a3 are larger than a2, the data set of the original sample library is updated to be A+B, the index is updated to be the maximum value of a1 and a3, and otherwise, the index is not updated.
The data expansion of the expanded image data set is carried out by adopting a data enhancement method, and model training is carried out by adopting a semi-supervision method, wherein the data enhancement method comprises the following steps:
(1) Acquiring the width ww and the height hh of an original image pic1, defining an original target in the original image as (xx 0, yy0, ww0, hh 0), wherein four variables respectively represent the abscissa, the ordinate, the width and the height of the original target;
(2) Randomly executing scaling operation on the original image and the original target and repeating for a plurality of times to realize data enhancement processing operation;
the zoom-out operation includes the steps of:
taking a random number t0 smaller than 1 as a reduction ratio, reducing the original target into a new image pic2 with width w0 x t0 and hh0 x t0, wherein the abscissa range of pic2 is [ xx0 x t0: xx 0/t0+w0/t 0] and the ordinate range is [ yy0/t0: yy0/t0+hh0/t0]; generating an image pic3 with width and height of ww0 and hh0 respectively, placing pic2 at the positions (ee 1, ee 2), wherein ee1 is smaller than ww-ww0, ee2 is smaller than hh-hh0, and in pic3, updating the position of the target reduced image pic2 to be (ee 1+ xx0 x t0, ee2+ yy0 x t 0); repeating the steps for a plurality of times to respectively obtain different enhanced images, and putting the enhanced images into a database;
the amplifying operation is as follows:
taking one of which is more than 1 and less thanAs an amplification ratio, amplifying an original target into a new image pic4 with the width and the height of ww0 t1 and hh0 t1 respectively, and randomly generating initial positions (ee 3 and ee 4) in the original image pic1 to place pic4; repeating the steps for a plurality of times to respectively obtain different enhanced images, and putting the enhanced images into a database;
the training method of the semi-supervision method comprises the following steps:
(1) Constructing two channels for training, wherein the first is an original target, and the second is an image generated by a shrinking operation;
(1) After training an original target, obtaining a feature map fe1, wherein the abscissa range of the fe1 position is [ xx0:xx0+ww0], and the ordinate range is [ yy0:y0+hh0 ];
(3) Training the image of which the original target is reduced to obtain a feature image fe2;
(4) Obtaining the coordinate position of fe1 by adopting the Roi-Align method, and obtaining aim1 for the target by assuming the target position to be positioned
(5) According to the above description, the position of the object in the original graph is updated to be the position (ee 1+xx0 t0, ee2+yy0 t 0) of pic2, and the object position of the feature graph fe2 is obtained by a roi-Align method, so as to obtain aim2 for the object;
(6) Aim1, aim are subjected to an immesize operation to fix the size, the similarity loss of an aim1.Aim2 is added to the loss function of the whole network, cross-correlation values are defined or the image is flattened into row vectors, and the absolute error sum between the row vectors is calculated;
(7) And (5) training is completed according to the steps.
Drawing an abnormal trend curve according to the abnormality comprises the following steps:
for abnormal numbers, 0 represents high values, 1 represents low values, 2 represents normal values, and a continuous curve is formulated according to the above, so that a graph in a digital form is obtained;
for the image form being defective, the historical data is plotted as a curve in the form of a defect of 0 and a defect of 1;
for other anomalies, 6 anomaly forms are defined, and the data are plotted as a curve by using 0 to represent only one display lamp fault, 1 to represent multiple display lamp faults, 2 to represent handle position anomalies, 3 to represent 0 and 2 to coexist, 4 to represent 1 and 2 to coexist, and 5 to represent logic anomalies.
Preferably, a safe encryption scheme is adopted for the abnormal image, a watermark is embedded into the abnormal image, the watermark image is selected according to pixel values of the abnormality and CANDY change, then a deep learning network is constructed according to the embedded watermark image, the identification of the embedded watermark frame is realized, and a watermark extraction algorithm is adopted for watermark extraction.
The secure encryption scheme includes the steps of:
encryption process:
(11) Collecting an abnormal image as a training sample;
(12) Five types of anomalies, namely a pointer, a button, an indicator light, a handle and a defect, are numbered 1,2,3,4 and 5;
(13) Collecting a large number of watermark images, respectively solving a graph after candy change, solving pixel sums for the graph after candy processing, dividing the watermark images into 5 groups according to the pixel sums, and marking the watermark images as A, B, C, D and E according to the sequence from large to small;
(14) For A, B, C, D and E, constructing a mapping relation between the sequence of the A, B, C, D and E and the abnormal marks 1,2,3,4 and 5, wherein the mapping relation is non-sequence mapping;
(15) When a video detected in real time is detected, if an abnormality exists in a certain frame, selecting an image group corresponding to the embedded watermark according to the abnormal sequence number, selecting two images from the corresponding group watermark, and embedding the video watermark by adopting an LSB algorithm, wherein the central positions of the two images are random positions near the central coordinates of the abnormal image;
the decryption process adopts a deep learning algorithm and LSB to realize the extraction of video watermark and the acquisition of information:
(21) Selecting a plurality of images which are embedded with watermarks and contain abnormal scenes, selecting a plurality of images which are not embedded with watermarks and do not contain abnormal scenes, and forming a training set by the three parts;
(22) Building a convolutional neural network for the training set of the first part, training the three parts by adopting a vgg deep learning algorithm, obtaining network performance through a LOSS curve and an accuracy curve, and storing a weight file;
(23) When the video is decrypted, converting the video into frames, verifying each frame, calling the weight of the steps, and extracting the watermark by adopting an LSB algorithm if the frame is judged by a network to be embedded with the watermark and contains an image of an abnormal scene;
(24) Solving a candy changed graph after extracting the watermark, solving pixel sums of the candy processed graph, and judging which type of A, B, C, D and E the image belongs to according to the size of the pixel sums;
(25) Obtaining an abnormal label according to the mapping relation;
(26) Embedding the abnormal images with watermarks according to the mapping relation to realize the expansion of an original training set, and carrying out target detection on the anomalies to obtain network weights;
(27) If the obtained abnormal number is consistent with the step 25 when the current detection frame is input to the network weight of the step 26, the decoding work is completed; if the watermark is inconsistent or not extracted, judging the unencrypted video;
preferably, the step of determining that the pointer image is abnormal in form includes:
(1) The network for target detection collects pointer images, marks the pointer positions to obtain a data set, and performs training by adopting a fasterRCNN;
(2) When the real-time image is input, calling the network in the step 1 to obtain the coordinates of the pointer, and independently storing the pointer image into an image;
(3) Constructing a data set of an independent pointer graph and pointer degrees, constructing an alexnet network for the data set, and realizing the mapping of the independent pointer graph and the pointer degrees;
(4) For the graph obtained in the step 2, calling the network in the step 3 to obtain the pointer degree P1;
(5) For the pointer graph obtained in step 2, performing hough change operation to obtain a maximum line segment, solving the slope corresponding to the maximum line segment, and obtaining a pointer degree p2 according to the mapping relation between the slope and the degree;
(6) Defining the accuracy of the pointer degree obtained by deep learning as acc1, and the accuracy obtained by hough as acc2;
(7) And finally, the pointer degree is p1 (acc 1/acc1+acc 2) +p2 (acc 1/acc1+acc 2).
The invention has the advantages that: the full-automatic all-weather fault monitoring process is designed, real-time monitoring is carried out on images in different forms (such as a pointer, a signal lamp, a defect, a numerical dial, a handle and the like), an abnormal data set is formed by collecting images of normal display and fault display, each numerical value is quantized by adopting a deep learning model, a network model is built, the defect of a sample in a real environment is made up, abnormal real-time detection is carried out more accurately, and automatic protection is effectively realized; and the original real video is tamper-proof and protected, so that the video data is restored after being watermarked, and whether the video is tamper-proof or not is verified.
The new data expansion network based on model verification is provided, the data is input into a plurality of node models for verification after self-enhancement, and the data is expanded to an abnormal data set after multiparty authentication belongs to abnormality
The invention not only can realize the monitoring and fault diagnosis of equipment and provide convenience for remote operation and maintenance protection, but also can provide basis and premise for future abnormal trend and even abnormal prediction by forming data characteristics for a long time.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a flow chart of the method for establishing a digital identification network according to the embodiment 1 of the present invention;
FIG. 3 is a flowchart of the method for creating an extended image dataset according to embodiment 1 of the present invention;
FIG. 4 is a flowchart of the method for establishing an anomaly determination network according to embodiment 1 of the present invention;
fig. 5 is a flowchart of anomaly determination network training in embodiment 1 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 accompanying drawings in the embodiments of the present invention.
Example 1
The embodiment discloses an anomaly monitoring method of an instrument cabinet, which comprises the following steps:
(1) A large amount of instrument cabinet data is collected and subjected to data enhancement, such as noise adding, noise removing and brightness change. Contrast variation, rotation, cropping, mixup, binarization, color space variation, image lbp, gabor variation, etc., to obtain a new extended image dataset.
(2) Uniformly marking target image data in the extended image data set to form a json file of the mark, constructing a target detection network, wherein a backbone network can adopt an alexnet, vgg, googlenet, yolo series detection network, ssd network, fastercnn network and the likeThe network can also be built by itself or a semi-supervised network can be set, mAP values and weight files of the network are stored, when the mAP values reach the availability network, the framework and the model of the network are not regulated any more, and the network weight w is stored 1 。
(3) Reading real-time video and calling network weight w 1 And selecting the target frame, and storing the target formed jpg image independently in a cache area for further detection.
(4) Aiming at the result obtained by the network, obtaining the interested area, further obtaining the values of the horizontal and vertical coordinates and the length and width, and calling the digital display identification network according to the form of digital display when the length and width ratio is large, so as to obtain a digital result, and judging whether the abnormality exists according to the values.
The following discloses a self-configurable identification mode: the target position i is expressed as f (i) = (X) i ,Y i ,W i ,H i ,c i ,p i ) Wherein X is i For the abscissa of the target position, Y i For the ordinate of the target position, W i Is of width, H i Is of height, c i Class labels for targets, p i Probability of being considered the target by the system; when the aspect ratio W i /H i Greater than threshold beta 1 Calling a digital identification network to obtain a numerical result, and judging whether an abnormality exists according to the numerical value; aspect ratio W i /H i Not greater than threshold beta 1 And when the image is in an abnormal state, the image data is input into an abnormal judgment network according to the area, and whether the image is in the abnormal state or not is obtained, wherein the image comprises a pointer, a button, an indicator light, a handle and a defect.
The distinction of image forms in this embodiment is defined as:
if area W i ×H i Satisfy W i ×H i >β 1 The image is in the form of a pointer;
if area W i ×H i Satisfy beta 1 ≤W i ×H i <β 2 The image is in the form of a signal lamp;
if area W i ×H i Satisfy beta 2 ≤W i ×H i ≤β 3 The image is in the form of a defect;
otherwise, the method is marked as other anomalies.
The digital recognition network judges the numerical value result as follows: the individual digital patterns are collected to form a dataset, a digital identification network such as BP, SVM, CNN is trained and its weights w are saved 2 Graying and gray projection are carried out on the image to be detected, the processed image is segmented into n pieces, and when i is less than or equal to n, w is called 2 And obtaining the numbers of the n pictures, and recombining to obtain a numerical result.
The algorithm of the anomaly determination network can take three methods: and (3) independently training a network, fusing the training network, and adopting an image processing method.
The method adopts a single training network method, and comprises the following steps: respectively collecting and training a pointer, a button, an indicator lamp, a handle and a normal map and an abnormal map of defects to form a pointer network, a signal lamp network and a defect network, and carrying out data enhancement, for example, noise adding, noise removing, brightness change, contrast change, rotation, cutting, mixup, binarization, color space change, image lbp, gabor change and the like to obtain a new data set, calculating the weight of each network, calling different network weights according to an image form to obtain whether the network is abnormal, wherein the specific identification method comprises the following steps: obtaining normal and abnormal samples, building a network model, and storing the weight of the network model when the accuracy of the network model reaches more than 99%; and detecting the video image in real time, calling the weight of the stored network model after obtaining the image data, and judging whether the image data is abnormal or not.
The fusion training network method comprises the following steps: and forming normal and abnormal figures of the pointer, the button, the indicator light, the handle and the defect into a comprehensive data set, carrying out data enhancement, corresponding to the normal label and the abnormal label, training a network to store weights, and calling the weights to the target position image to obtain whether the target position image is abnormal or not.
The image processing method comprises the following steps:
preprocessing and thresholding (image preprocessing comprises graying and binarizing) the images of the pointer, the handle and the button, highlighting the height of the button, fitting the slopes of the pointer and the handle, and judging whether the slopes of the real-time pointer and the handle images or the heights of the buttons are abnormal or not;
performing color analysis on the signal lamp image, and judging whether the signal lamp image is abnormal or not according to the real-time signal lamp color;
and preprocessing the defect image, calculating the defect area, and judging whether the defect is a defect according to the defect area of the real-time defect image.
The specific steps of the anomaly judgment network training are as follows:
and (3) marking the rest part of the original background removal target as Q, performing enhancement processing on the obtained abnormal data image by adopting 11 methods of rotation, brightness increase, brightness reduction, filtering, noise addition, clipping, blurring, masking processing, binarization, LBP change and gabor change, respectively inputting the processed 11 result images into n abnormal judgment models, wherein each model outputs 11 results which are considered to be abnormal when the model outputs more than n/2 models, and the data image is considered to be abnormal data, and expanding the 11 results as new samples to the image data set.
The method for acquiring the abnormal image data in the extended image data set comprises the following steps:
detecting original data in real time, if abnormality exists, taking the abnormal image as one of accumulation numbers, when certain data is accumulated, putting the data into a discrimination network, determining that the accumulated data are all abnormal data, and recording the abnormal data as B; assuming that the original sample library data set is A, the initial index is the accuracy a obtained through training, B is uploaded and expanded to the original data set to form a new data set A+B, training is carried out on the data set A+ B, A, B, the obtained accuracy values are respectively marked as a1, a2 and a3, if a1 and a3 are larger than a2, the data set of the original sample library is updated to be A+B, the index is updated to be the maximum value of a1 and a3, and otherwise, the index is not updated.
The data expansion of the expanded image data set is carried out by adopting a data enhancement method, the model training is carried out by adopting a semi-supervision method, and the data enhancement method comprises the following steps:
acquiring the width ww and the height hh of an original image pic1, defining an original target in the original image as (xx 0, yy0, ww0, hh 0), wherein four variables respectively represent the abscissa, the ordinate, the width and the height of the original target;
and randomly performing scaling operation on the original image and the original target and repeating for a plurality of times to realize data enhancement processing operation.
The zoom-out operation includes the steps of:
taking a random number t0 smaller than 1 as a reduction ratio, reducing the original target into a new image pic2 with width w0 x t0 and height hh0 x t0, wherein the abscissa range of pic2 is [ xx0 x t0: xx0 x0+ ww0 x t0] and the ordinate range is [ yy0 x t0: yy0 x t0+ hh0 x t0]; generating an image pic3 with width and height of ww0 and hh0 respectively, placing pic2 at the positions (ee 1, ee 2), wherein ee1 is smaller than ww-ww0, ee2 is smaller than hh-hh0, and in pic3, updating the position of the target reduced image pic2 to be (ee 1+ xx0 x t0, ee2+ yy0 x t 0); repeating the steps for 5 times to obtain five different enhancement patterns, and solving the problem of low accuracy of small target identification.
The amplifying operation is as follows:
taking one of which is more than 1 and less thanAs an amplification ratio, amplifying an original target into a new image pic4 with the width and the height of ww0 t1 and hh0 t1 respectively, and randomly generating initial positions (ee 3 and ee 4) in the original image pic1 to place pic4; repeating the steps for 5 times to obtain different data results, and solving the problem of inconsistent size of the targets.
The training method of the semi-supervision method comprises the following steps:
constructing two channels for training, wherein the first is an original target, and the second is an image generated by a shrinking operation;
after training an original target, obtaining a feature map fe1, wherein the abscissa range of the fe1 position is [ xx0:xx0+ww0], and the ordinate range is [ yy0:y0+hh0 ];
training the image of which the original target is reduced to obtain a feature image fe2;
obtaining the coordinate position of fe1 by adopting the Roi-Align method, and obtaining aim1 for the target by assuming the target position to be positioned
According to the above description, the position of the target in the original graph is updated to be the position (ee 1+xx0 t0, ee2+yy0 t 0) of pic2, and the target position of the feature graph fe2 is obtained by the method of Roi-Align, so as to obtain the corresponding target aim2;
for aim, aim, adopting an immesize operation to fix the size, adding a similarity loss of aim1.Aim2 to a loss function of the whole network, defining a cross-correlation value or flattening an image into row vectors, and calculating an absolute error sum between the row vectors;
and (5) training is completed according to the steps.
Example 2
The invention also discloses a monitoring means based on the embodiment 1, which can draw an abnormal trend curve according to the abnormality, and specifically comprises the following steps:
for abnormal numbers, 0 represents high values, 1 represents low values, 2 represents normal values, and a continuous curve is formulated according to the above, so that a graph in a digital form is obtained;
for the image form being defective, the historical data is plotted as a curve in the form of a defect of 0 and a defect of 1;
for other anomalies, 6 anomaly forms are defined, and the data are plotted as a curve by using 0 to represent only one display lamp fault, 1 to represent multiple display lamp faults, 2 to represent handle position anomalies, 3 to represent 0 and 2 to coexist, 4 to represent 1 and 2 to coexist, and 5 to represent logic anomalies.
Example 3
The invention also discloses an encryption algorithm, which can adopt a safe encryption scheme for the abnormal image when the digital embodiments 1 and 2 detect the abnormality, embed the watermark into the abnormal image, select the watermark image according to the pixel values of the abnormality and the CANDY change, then construct a deep learning network according to the embedded watermark image, realize the judgment of the embedded watermark frame, and adopt a watermark extraction algorithm to extract the watermark.
The secure encryption scheme includes the steps of:
encryption process:
(1) Collecting an abnormal image as a training sample;
(2) Five types of anomalies, namely a pointer, a button, an indicator light, a handle and a defect, are numbered 1,2,3,4 and 5;
(3) Collecting 100 watermark images, solving a candy-changed image of the 100 watermark images, solving pixel sums of the candy-processed image, dividing the watermark images into 5 groups according to the pixel sums, and marking the watermark images as A, B, C, D and E according to the sizes from large to small;
(4) For A, B, C, D, E, and anomaly labels 1,2,3,4,5, mapping relation is built, for safety, non-sequential mapping is adopted, namely mapping other than 1 and A, 2 and B and the like is adopted, for example, mapping methods of 2 and A, 3 and B, 4 and C, 5 and D, 1 and E can be adopted, or other mapping can be adopted as keys transferred by the two;
(5) When detecting video in real time, if a frame is found to be abnormal, selecting image groups corresponding to the embedded watermarks according to the abnormal sequence numbers, selecting two images from the corresponding group watermarks, embedding the video watermarks by adopting an LSB algorithm, wherein the center positions of the two images are random positions near the center coordinates of the abnormal images, and the decryption can be facilitated.
The decryption process adopts a deep learning algorithm and LSB to realize the extraction of video watermark and the acquisition of information:
(1) Selecting a plurality of images which are embedded with watermarks and contain abnormal scenes, selecting a plurality of images which are not embedded with watermarks and do not contain abnormal scenes, and forming a training set by the three parts;
(2) Building a convolutional neural network for the training set of the first part, training the three parts by adopting deep learning algorithms such as lenet, alexnet and vgg, obtaining network performance through a LOSS curve and an accuracy curve, and storing a weight file;
(3) When the video is decrypted, converting the video into frames, verifying each frame, calling the weight of the steps, and extracting the watermark by adopting an LSB algorithm if the frame is judged by a network to be embedded with the watermark and contains an image of an abnormal scene;
(4) Solving a candy changed graph after extracting the watermark, solving pixel sums of the candy processed graph, and judging which type of A, B, C, D and E the image belongs to according to the size of the pixel sums;
(5) Obtaining an abnormal label according to the mapping relation;
(6) Embedding the abnormal images with watermarks according to the mapping relation to realize the expansion of an original training set, and carrying out target detection on the anomalies to obtain network weights;
(7) If the obtained abnormal number is consistent with the step 25 when the current detection frame is input to the network weight of the step 26, the decoding work is completed; if the watermark is inconsistent or not extracted, the unencrypted video is determined.
Example 4
The invention also discloses a specific method for determining the abnormality of the pointer image form, which comprises the following steps:
(1) The network for target detection collects pointer images, marks the pointer positions to obtain a data set, and performs training by adopting a fasterRCNN;
(2) When the real-time image is input, calling the network in the step 1 to obtain the coordinates of the pointer, and independently storing the pointer image into an image;
(3) Constructing a data set of an independent pointer graph and pointer degrees, constructing an alexnet network for the data set, and realizing the mapping of the independent pointer graph and the pointer degrees;
(4) For the graph obtained in the step 2, calling the network in the step 3 to obtain the pointer degree P1;
(5) For the pointer graph obtained in step 2, performing hough change operation to obtain a maximum line segment, solving the slope corresponding to the maximum line segment, and obtaining a pointer degree p2 according to the mapping relation between the slope and the degree;
(6) Defining the accuracy of the pointer degree obtained by deep learning as acc1, and the accuracy obtained by hough as acc2;
(7) And finally, the pointer degree is p1 (acc 1/acc1+acc 2) +p2 (acc 1/acc1+acc 2).
Finally, it should be noted that: the foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (10)
1. An anomaly monitoring method for an instrument cabinet is characterized by comprising the following steps:
(1) Collecting instrument cabinet data and obtaining an extended image dataset;
(2) Uniformly marking target image data and position signals in the extended image data set to form a marking file, forming a target detection network, storing mAP values and weight files of the network, not adjusting the architecture and model of the network when the mAP values reach the availability network, and storing the network weight w 1 ;
(3) Reading real-time video and calling network weight w 1 And selecting the target frame, and determining the abnormal type, coordinates of the abnormal image and confidence information according to the data result.
2. The abnormality monitoring method of an instrument cabinet according to claim 1, wherein the method of determining an abnormality type is:
the target position i is expressed as f (i) = (X) i ,Y i ,W i ,H i ,c i ,p i ) Wherein X is i For the abscissa of the target position, Y i For the ordinate of the target position, W i Is of width, H i Is of height, c i Class labels for targets, p i Probability of being considered the target by the system; when the aspect ratio W i /H i Greater than threshold beta 1 Calling a digital identification network to obtain a numerical result, and judging whether an abnormality exists according to the numerical value; the step of the digital identification network judging numerical value result is as follows: collecting individual digital graphics to form a data set, training a numberWord recognition network and preserving its weight w 2 Graying and gray projection are carried out on the image to be detected, the processed image is segmented into n pieces, and when i is less than or equal to n, w is called 2 Obtaining the numbers of n pictures, and recombining to obtain a numerical result;
aspect ratio W i /H i Not greater than threshold beta 1 And when the image is in an abnormal state, the image data is input into an abnormal judgment network according to the area, and whether the image is in the abnormal state or not is obtained, wherein the image comprises a pointer, a button, an indicator light, a handle and a defect.
3. The abnormality monitoring method of an instrument cabinet according to claim 1, wherein the algorithm of the abnormality judgment network adopts a single training network method, a fusion training network method or an image processing method;
the individual training network method comprises the steps of: respectively collecting and training a pointer, a button, an indicator lamp, a handle and a normal map and an abnormal map of a defect to form a pointer network, a signal lamp network and a defect network, carrying out data enhancement, calculating the weight of each network, calling different network weights according to an image form, and obtaining whether the network is abnormal or not, wherein the specific identification method comprises the following steps: obtaining normal and abnormal samples, building a network model, and storing the weight of the network model when the accuracy of the network model reaches more than 99%; detecting a video image in real time, calling the weight of a stored network model after obtaining image data, and judging whether the image data is abnormal or not;
the fusion training network method comprises the following steps: forming normal and abnormal figures of the pointer, the button, the indicator light, the handle and the defect into a comprehensive data set, carrying out data enhancement, corresponding to the normal label and the abnormal label, training a network to store weights, and calling the weights to the target position image to obtain whether the target position image is abnormal or not;
the image processing method includes the steps of:
preprocessing and thresholding images of the pointer, the handle and the button, highlighting the height of the button, fitting the slopes of the pointer and the handle, and judging whether the slopes of the real-time pointer and the handle images or the height of the button are abnormal or not;
performing color analysis on the signal lamp image, and judging whether the signal lamp image is abnormal or not according to the real-time signal lamp color;
preprocessing a defect image, calculating the defect area, and judging whether the defect is a defect or not according to the defect area of the real-time defect image;
the image preprocessing comprises graying and binarization.
4. The anomaly monitoring method for an instrument cabinet according to claim 1, wherein the specific steps of the anomaly determination network training are:
the rest of the original background removing target is marked as Q, the obtained abnormal data image is respectively processed by 11 methods of rotation, brightness increase, brightness reduction, filtering, noise addition, clipping, blurring, masking processing, binarization, LBP change and gabor change, the processed 11 result images are respectively input into n abnormal judging models, each model outputs 11 results which are considered to be abnormal, when more than n/2 models are considered to be abnormal, the data image is considered to be abnormal data, the results of the 11 abnormal data are combined with a plurality of original background images, and the data image is expanded to an image data set as a new sample.
5. The abnormality monitoring method of an instrument cabinet according to claim 1, wherein the acquisition method of the extended image dataset abnormality is:
detecting original data in real time, if abnormality exists, taking the abnormal image as one of accumulation numbers, when certain data is accumulated, putting the data into a discrimination network, determining that the accumulated data are all abnormal data, and recording the abnormal data as B; assuming that the original sample library data set is A, the initial index is the accuracy a obtained through training, B is uploaded and expanded to the original data set to form a new data set A+B, training is carried out on the data set A+ B, A, B, the obtained accuracy values are respectively marked as a1, a2 and a3, if a1 and a3 are larger than a2, the data set of the original sample library is updated to be A+B, the index is updated to be the maximum value of a1 and a3, and otherwise, the index is not updated.
6. The anomaly monitoring method for an instrument cabinet of claim 1, wherein the extended image dataset is data-extended by a data enhancement method and model-trained by a semi-supervised method, the data enhancement method comprising the steps of:
(1) Acquiring the width ww and the height hh of an original image pic1, defining an original target in the original image as (xx 0, yy0, ww0, hh 0), wherein four variables respectively represent the abscissa, the ordinate, the width and the height of the original target;
(2) Randomly executing scaling operation on the original image and the original target and repeating for a plurality of times to realize data enhancement processing operation;
the zoom-out operation includes the steps of:
taking a random number t0 smaller than 1 as a reduction ratio, reducing the original target into a new image pic2 with width w0 x t0 and height hh0 x t0, wherein the abscissa range of pic2 is [ xx0 x t0: xx0 x0+ ww0 x t0] and the ordinate range is [ yy0 x t0: yy0 x t0+ hh0 x t0]; generating an image pic3 with width and height of ww0 and hh0 respectively, placing pic2 at the positions (ee 1, ee 2), wherein ee1 is smaller than ww-ww0, ee2 is smaller than hh-hh0, and in pic3, updating the position of the target reduced image pic2 to be (ee 1+ xx0 x t0, ee2+ yy0 x t 0); repeating the steps for a plurality of times to respectively obtain different enhanced images, and putting the enhanced images into a database;
the amplifying operation is as follows:
taking one of which is more than 1 and less thanAs an amplification ratio, amplifying an original target into a new image pic4 with the width and the height of ww0 t1 and hh0 t1 respectively, and randomly generating initial positions (ee 3 and ee 4) in the original image pic1 to place pic4; repeating the steps for a plurality of times to respectively obtain different enhanced images, and putting the enhanced images into a database;
the training method of the semi-supervision method comprises the following steps:
(1) Constructing two channels for training, wherein the first is an original target, and the second is an image generated by a shrinking operation;
(1) After training an original target, obtaining a feature map fe1, wherein the abscissa range of the fe1 position is [ xx0:xx0+ww0], and the ordinate range is [ yy0:y0+hh0 ];
(3) Training the image of which the original target is reduced to obtain a feature image fe2;
(4) Obtaining the coordinate position of fe1 by adopting the Roi-Align method, and obtaining aim1 for the target by assuming the target position to be positioned
(5) According to the above description, the position of the object in the original graph is updated to be the position (ee 1+xx0 t0, ee2+yy0 t 0) of pic2, and the object position of the feature graph fe2 is obtained by a roi-Align method, so as to obtain aim2 for the object;
(6) Aim1, aim are subjected to an immesize operation to fix the size, the similarity loss of an aim1.Aim2 is added to the loss function of the whole network, cross-correlation values are defined or the image is flattened into row vectors, and the absolute error sum between the row vectors is calculated;
(7) And (5) training is completed according to the steps.
7. The abnormality monitoring method of an instrument cabinet according to claim 1, characterized by drawing an abnormality trend curve from the abnormality, comprising the steps of:
for abnormal numbers, 0 represents high values, 1 represents low values, 2 represents normal values, and a continuous curve is formulated according to the above, so that a graph in a digital form is obtained;
for the image form being defective, the historical data is plotted as a curve in the form of a defect of 0 and a defect of 1;
for other anomalies, 6 anomaly forms are defined, and the data are plotted as a curve by using 0 to represent only one display lamp fault, 1 to represent multiple display lamp faults, 2 to represent handle position anomalies, 3 to represent 0 and 2 to coexist, 4 to represent 1 and 2 to coexist, and 5 to represent logic anomalies.
8. The anomaly monitoring method of the instrument and meter cabinet according to claim 1, wherein a safe encryption scheme is adopted for the anomaly image, a watermark is embedded into the anomaly image, the watermark image is selected according to pixel values of anomaly and CANDY change, then a deep learning network is constructed according to the embedded watermark image, discrimination of an embedded watermark frame is realized, and watermark extraction algorithm is adopted for watermark extraction.
9. The anomaly monitoring method for an instrument cabinet according to claim 8, wherein the secure encryption scheme comprises the steps of:
encryption process:
(11) Collecting an abnormal image as a training sample;
(12) Five types of anomalies, namely a pointer, a button, an indicator light, a handle and a defect, are numbered 1,2,3,4 and 5;
(13) Collecting a large number of watermark images, respectively solving a graph after candy change, solving pixel sums for the graph after candy processing, dividing the watermark images into 5 groups according to the pixel sums, and marking the watermark images as A, B, C, D and E according to the sequence from large to small;
(14) For A, B, C, D and E, constructing a mapping relation between the sequence of the A, B, C, D and E and the abnormal marks 1,2,3,4 and 5, wherein the mapping relation is non-sequence mapping;
(15) When a video detected in real time is detected, if an abnormality exists in a certain frame, selecting an image group corresponding to the embedded watermark according to the abnormal sequence number, selecting two images from the corresponding group watermark, and embedding the video watermark by adopting an LSB algorithm, wherein the central positions of the two images are random positions near the central coordinates of the abnormal image;
the decryption process adopts a deep learning algorithm and LSB to realize the extraction of video watermark and the acquisition of information:
(21) Selecting a plurality of images which are embedded with watermarks and contain abnormal scenes, selecting a plurality of images which are not embedded with watermarks and do not contain abnormal scenes, and forming a training set by the three parts;
(22) Building a convolutional neural network for the training set of the first part, training the three parts by adopting a vgg deep learning algorithm, obtaining network performance through a LOSS curve and an accuracy curve, and storing a weight file;
(23) When the video is decrypted, converting the video into frames, verifying each frame, calling the weight of the steps, and extracting the watermark by adopting an LSB algorithm if the frame is judged by a network to be embedded with the watermark and contains an image of an abnormal scene;
(24) Solving a candy changed graph after extracting the watermark, solving pixel sums of the candy processed graph, and judging which type of A, B, C, D and E the image belongs to according to the size of the pixel sums;
(25) Obtaining an abnormal label according to the mapping relation;
(26) Embedding the abnormal images with watermarks according to the mapping relation to realize the expansion of an original training set, and carrying out target detection on the anomalies to obtain network weights;
(27) If the obtained abnormal number is consistent with the step 25 when the current detection frame is input to the network weight of the step 26, the decoding work is completed; if the watermark is inconsistent or not extracted, the unencrypted video is determined.
10. The abnormality monitoring method of an instrument cabinet according to claim 2, wherein the step of determining that the pointer image form is abnormal includes:
(1) The network for target detection collects pointer images, marks the pointer positions to obtain a data set, and performs training by adopting a fasterRCNN;
(2) When the real-time image is input, calling the network in the step 1 to obtain the coordinates of the pointer, and independently storing the pointer image into an image;
(3) Constructing a data set of an independent pointer graph and pointer degrees, constructing an alexnet network for the data set, and realizing the mapping of the independent pointer graph and the pointer degrees;
(4) For the graph obtained in the step 2, calling the network in the step 3 to obtain the pointer degree P1;
(5) For the pointer graph obtained in step 2, performing hough change operation to obtain a maximum line segment, solving the slope corresponding to the maximum line segment, and obtaining a pointer degree p2 according to the mapping relation between the slope and the degree;
(6) Defining the accuracy of the pointer degree obtained by deep learning as acc1, and the accuracy obtained by hough as acc2;
(7) And finally, the pointer degree is p1 (acc 1/acc1+acc 2) +p2 (acc 1/acc1+acc 2).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310460460.1A CN116645329A (en) | 2023-04-26 | 2023-04-26 | Abnormality monitoring method for instrument and meter cabinet |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310460460.1A CN116645329A (en) | 2023-04-26 | 2023-04-26 | Abnormality monitoring method for instrument and meter cabinet |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116645329A true CN116645329A (en) | 2023-08-25 |
Family
ID=87617829
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310460460.1A Pending CN116645329A (en) | 2023-04-26 | 2023-04-26 | Abnormality monitoring method for instrument and meter cabinet |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116645329A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117095411A (en) * | 2023-10-16 | 2023-11-21 | 青岛文达通科技股份有限公司 | Detection method and system based on image fault recognition |
-
2023
- 2023-04-26 CN CN202310460460.1A patent/CN116645329A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117095411A (en) * | 2023-10-16 | 2023-11-21 | 青岛文达通科技股份有限公司 | Detection method and system based on image fault recognition |
CN117095411B (en) * | 2023-10-16 | 2024-01-23 | 青岛文达通科技股份有限公司 | Detection method and system based on image fault recognition |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Tao et al. | Unsupervised anomaly detection for surface defects with dual-siamese network | |
CN109740609B (en) | Track gauge detection method and device | |
CN112307919B (en) | Improved YOLOv 3-based digital information area identification method in document image | |
CN111539330B (en) | Transformer substation digital display instrument identification method based on double-SVM multi-classifier | |
CN116645329A (en) | Abnormality monitoring method for instrument and meter cabinet | |
CN111212291A (en) | DFL-CNN network-based video intra-frame object removal tamper detection method | |
CN114758329A (en) | System and method for predicting temperature of target area in thermal imaging graph based on deep learning | |
Li et al. | Robust median filtering detection based on the difference of frequency residuals | |
CN115995056A (en) | Automatic bridge disease identification method based on deep learning | |
Devereux et al. | A new approach for crack detection and sizing in nuclear reactor cores | |
CN113660484B (en) | Audio and video attribute comparison method, system, terminal and medium based on audio and video content | |
Zhang et al. | Prnu-based image forgery localization with deep multi-scale fusion | |
CN116912184B (en) | Weak supervision depth restoration image tampering positioning method and system based on tampering area separation and area constraint loss | |
Guan et al. | An effective image steganalysis method based on neighborhood information of pixels | |
CN115112669B (en) | Pavement nondestructive testing identification method based on small sample | |
Zhao et al. | Image tampering detection via semantic segmentation network | |
CN112614094B (en) | Insulator string abnormity positioning and identifying method based on sequence state coding | |
CN114663731A (en) | Training method and system of license plate detection model, and license plate detection method and system | |
CN113936300A (en) | Construction site personnel identification method, readable storage medium and electronic device | |
CN114120097A (en) | Distribution network engineering on-site transformer detection method and system based on machine vision | |
CN112990350A (en) | Target detection network training method and target detection network-based coal and gangue identification method | |
CN115082865B (en) | Bridge crane intrusion dangerous behavior early warning method and system based on visual image recognition | |
CN117634006B (en) | BIM technology-based sleeve embedded engineering management system and method | |
CN117354495B (en) | Video monitoring quality diagnosis method and system based on deep learning | |
CN112788331B (en) | Video recompression detection method, terminal equipment and storage medium |
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 |