CN115661151B - Method for identifying leakage of hydrogen production system based on machine vision high precision - Google Patents
Method for identifying leakage of hydrogen production system based on machine vision high precision Download PDFInfo
- Publication number
- CN115661151B CN115661151B CN202211680630.9A CN202211680630A CN115661151B CN 115661151 B CN115661151 B CN 115661151B CN 202211680630 A CN202211680630 A CN 202211680630A CN 115661151 B CN115661151 B CN 115661151B
- Authority
- CN
- China
- Prior art keywords
- leakage
- bubble
- image
- hydrogen
- bubbles
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 229910052739 hydrogen Inorganic materials 0.000 title claims abstract description 144
- 239000001257 hydrogen Substances 0.000 title claims abstract description 142
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 title claims abstract description 140
- 238000000034 method Methods 0.000 title claims abstract description 90
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 62
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 33
- 239000007789 gas Substances 0.000 claims abstract description 21
- 238000005984 hydrogenation reaction Methods 0.000 claims abstract description 8
- 230000004927 fusion Effects 0.000 claims abstract description 6
- 230000001629 suppression Effects 0.000 claims abstract description 4
- 238000001514 detection method Methods 0.000 claims description 55
- 238000007654 immersion Methods 0.000 claims description 32
- 238000012549 training Methods 0.000 claims description 25
- 238000012545 processing Methods 0.000 claims description 22
- 238000013135 deep learning Methods 0.000 claims description 18
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 13
- 239000013598 vector Substances 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000009826 distribution Methods 0.000 claims description 8
- 238000012216 screening Methods 0.000 claims description 8
- 238000005286 illumination Methods 0.000 claims description 7
- 238000013139 quantization Methods 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 6
- 230000009191 jumping Effects 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 5
- 238000003384 imaging method Methods 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000010835 comparative analysis Methods 0.000 claims description 2
- 230000007613 environmental effect Effects 0.000 claims description 2
- 238000000605 extraction Methods 0.000 claims description 2
- 230000003313 weakening effect Effects 0.000 claims description 2
- 238000011897 real-time detection Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 13
- 239000003086 colorant Substances 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 238000004880 explosion Methods 0.000 description 3
- 238000007667 floating Methods 0.000 description 3
- 150000002431 hydrogen Chemical class 0.000 description 3
- 230000033001 locomotion Effects 0.000 description 3
- 239000000463 material Substances 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000011521 glass Substances 0.000 description 2
- 239000012535 impurity Substances 0.000 description 2
- WABPQHHGFIMREM-UHFFFAOYSA-N lead(0) Chemical compound [Pb] WABPQHHGFIMREM-UHFFFAOYSA-N 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000004456 color vision Effects 0.000 description 1
- 238000002485 combustion reaction Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 238000005474 detonation Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005868 electrolysis reaction Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 230000004438 eyesight Effects 0.000 description 1
- 238000011049 filling Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000002386 leaching Methods 0.000 description 1
- 238000012067 mathematical method Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000000116 mitigating effect Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000010349 pulsation Effects 0.000 description 1
- 238000002791 soaking Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
- 230000009967 tasteless effect Effects 0.000 description 1
- 238000009423 ventilation Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a method for identifying leakage of a hydrogen production system based on machine vision high precision. It comprises the following steps: the method comprises the following steps: detecting a moving target; step two: non-bubble target suppression based on multi-feature fusion; a multi-feature classifier is designed according to features such as bubble color, texture and shape, false targets are eliminated, and the specific method comprises the following steps: s21: extracting color features of the moving target; s22: extracting the shape characteristics of the moving target; s23: extracting HOG texture features of a moving target; s24: acquiring a strong classifier through an Adaboost algorithm, and judging whether the moving object acquired in the first step is a bubble or a background; step three: and positioning a gas leakage source. The invention has the advantages of online real-time detection and identification of the leakage point and the leakage condition of the hydrogenation device.
Description
Technical Field
The invention relates to the field of hydrogen energy safety and artificial intelligence, in particular to a method for identifying hydrogen gas system leakage based on machine vision high precision.
Background
The hydrogen production by water electrolysis or hydrogen transportation by pipeline is a mature industrial technology, and the following problems exist in the application process: the hydrogen production system belongs to medium and high pressure containers and pipeline systems, and has the characteristics of small density, large diffusion coefficient, low ignition temperature, wide explosion limit (volume fraction of 4-74%), high combustion flame speed and the like, if a large amount of high-pressure hydrogen stored in the hydrogenation station leaks, large-scale combustible mixed gas is easily formed, and explosion or detonation can be caused once the hydrogen is ignited, so that huge casualties and property loss are caused, and the hydrogen production plant has very definite requirements on static electricity, open fire, electrical switches, ventilation, safe clearance with buildings and the like so as to ensure safety; corresponding requirements are provided for factory spaces and the like, and the construction cost of the factory building is increased; the traditional hydrogen leakage alarm technology mainly adopts a hydrogen concentration detection alarm, a detection sensor, a lead and a corresponding power supply which are arranged above or near each main leakage point of a hydrogen production station, when hydrogen leakage occurs in the hydrogen production station, the alarm is connected with a hydrogen emergency cut-off valve and an exhaust fan in the station, so that a gas source is cut off, the accident exhaust fan is started, and combustible mixed gas in a closed space in the station is quickly replaced, so that the risk of fire and explosion is reduced, however, in the method, because the hydrogen concentration detection alarm, the detection sensor and other devices need to carry out transmission, reaction and the like, the delayed feedback of a detection result is caused, and the real-time performance is relatively poor; the hydrogen concentration detection alarm, the detection sensor and the like are mechanical structures, so that accidents are easy to happen in the use process, the detection precision is low or the detection cannot be realized, the devices such as the hydrogen concentration detection alarm, the detection sensor and the like need to be overhauled, replaced and the like, and the reliability is relatively poor; therefore, the prior art can not quickly and accurately position the hydrogen leakage point;
therefore, it is necessary to develop a method for quickly and accurately positioning the hydrogen leakage point, and the method has high real-time performance and high reliability.
Disclosure of Invention
The invention aims to provide a method for identifying leakage of a hydrogen production system based on machine vision high-precision, which can quickly and accurately position hydrogen leakage points, has high real-time performance and high reliability, and can dynamically read leakage bubbles (including a large amount of leakage bubbles and water-insoluble trace leakage bubbles generated by tiny and initial leakage) at key parts of hydrogen-containing equipment in all weather for 24 hours under the condition that the operation of hydrogen-containing equipment, pipelines and valves such as hydrogen production equipment, pipelines and valves is not influenced in a full-immersion environment; the problem of hydrogen easily reveal in open space, it is flammable explosive in aerobic air environment, and the problem of colourless tasteless difficult discovery is solved prior art because devices such as hydrogen concentration detection alarm, detection sensor transmit because it needs, reaction etc. lead to the time delay feedback of testing result, the real-time is relatively poor, and the easy emergence accident in the use, lead to detecting the precision low or can't detect, and need regularly overhaul, operations such as change, the relatively poor defect of reliability.
In order to achieve the purpose, the technical scheme of the invention is as follows: the method for identifying the leakage of the hydrogen production system based on machine vision high precision is characterized in that: the method comprises the steps of adopting a device for identifying the leakage of a hydrogen production system with high precision based on machine vision, collecting images of underwater hydrogen-containing equipment by an industrial camera in a full-submerged mode, and identifying bubbles generated by tiny leakage of the easily-leaked parts such as pipelines, joints, valves and the like of the underwater hydrogen-containing equipment by utilizing a machine vision algorithm, thereby realizing online real-time detection and identifying the leakage point and the leakage condition of the hydrogen production system;
the method comprises the following steps:
the method comprises the following steps: detecting a moving target;
step two: non-bubble target suppression based on multi-feature fusion;
a multi-feature classifier is designed according to features such as bubble color, texture and shape, false targets are eliminated, and the specific method comprises the following steps:
s21: extracting color features of the moving target;
s22: extracting the shape characteristics of the moving target;
s23: extracting HOG texture features of a moving target;
s24: acquiring a strong classifier through an Adaboost algorithm, and judging whether the moving object acquired in the first step is a bubble or a background;
step three: and positioning a gas leakage source.
In the above technical solution, in the step one, the moving object is detected by the specific method:
s11: reading a video stream image;
s12: confirming whether the current image is a first frame image;
if yes, initializing an RGB-VIBE model, and jumping to the step S11;
if not, jumping to the step S13;
s13: updating the RGB-VIBE model;
s14: detecting a moving target by adopting a frame difference method;
s15: and outputting the position of the moving target.
In the above technical solution, in step S21, the color feature of the moving object is extracted, and the bubbles are classified based on the color feature, and the specific method includes:
s211: acquiring a target image;
s212: color space conversion;
using the HSI color space, the image is converted from the RGB color space to the HSI color space by the following formula:
s213: color quantization;
the HSI color space is quantized, the chrominance is quantized to 8 spaces, the saturation is divided into 3 spaces, and the luminance is divided into 3 spaces, that is, the color space is divided into 72 intervals, and the specific quantization values are as follows:
s214: extracting a statistical histogram;
the statistical histogram of the image features is a one-dimensional discrete function, as shown in the following formula:
wherein: k represents the characteristic value of the image; l is the number of characteristic obtainable values; n is k The number of pixels with a characteristic value of k in the image is obtained; n is the total number of image pixels;
the color histogram is characterized in that: after the image is rotated, the color histogram of the image is not changed; the color histogram is easy to extract and the similarity between the two histograms is easier to calculate.
In the above technical solution, in step S23, the method for extracting the texture feature of the moving object HOG includes:
s231: converting the input color image into a gray image;
s232: standardizing the color space of the input image by using a Gamma correction method;
s233: calculating a gradient; capturing contour information, and further weakening the interference of illumination;
s234: projecting the gradient to a gradient direction of the cell; providing a code for the local image area;
s235: normalizing all cells on the block;
s236: collecting and obtaining HOG characteristics of all blocks in a detection space;
all overlapping blocks in the detection window are collected for HOG features and combined into a final feature vector for classification.
In the above technical solution, in step S24, a strong classifier is obtained through an Adaboost algorithm, and it is determined that the moving object obtained in the first step is a bubble or a background, where the specific method is as follows:
provided with an inputnEach training sample was:
wherein->
Is an input training sample, is>
Respectively representing a positive sample and a negative sample, wherein the number of positive samples is->
The number of negative samples is->
，/>
Wherein the training sample is a sample of air bubbles or background;
s241: initializing weights for each sample
；/>
S242: for each one
，/>
Number of weak classifiers:
(1) normalizing the weights to a probability distribution;
wherein: wherein w is a weight, and subscript i, j indicates that the indices all belong to the interval of 1, … …, n; d (i) indicates that i belongs to a sample interval; n is the number of training samples;
(2) for each feature
Training a weak classifier>
Calculating the weighted error rate of the weak classifiers corresponding to all the characteristics; weighted error rate->
The calculation formula of (c) is:
wherein: w (x) is a function of the calculated weights;
(3) selecting the best weak classifier
(possess the bestSmall error rate->
）；
(4) Adjusting the weight according to the optimal weak classifier;
wherein,
indicates that it is correctly classified>
Indicating that it was misclassified;
adjustment factor
Can be based on>
Calculating;
s243: obtaining a final strong classifier;
the final strong classifier is:
wherein: h (x) is a strong classifier;
representing a weak classifier;
superimposed weight coefficients
Can be based on>
Calculating;
and judging the bubbles or the background through the numerical value calculated by the strong classifier, wherein the bubbles are taken when h (x) is 1, and the background is taken when h (x) is 0.
In the above technical solution, in the third step, the leakage air source is positioned, and the specific method is as follows:
carrying out vertical projection on the detected bubble image region, then selecting a horizontal line and a vertical line in a projection histogram, setting the horizontal line and the vertical line as coordinate axes, and acquiring the center coordinates of all bubbles
；
Firstly, classifying the circle center coordinates of the bubbles by adopting a kmeans clustering method aiming at the circle center coordinates of the bubbles;
after the number of the bubble source regions is divided, each bubble source is respectively positioned, and the specific method comprises the following steps:
assuming that the number of the bubble sources is K, the circle center coordinate of the bubble in the kth bubble source region is defined as
Determining the position of the bubble source according to the following formula, and obtaining the coordinates of the position of the bubble source:
wherein the coordinates of the center of the circle are defined as
；（/>
，/>
) Representing the coordinates of a to-be-solved leakage air source; i represents the ith leakage gas source; />
Indicating the number of the identified bubbles in the k-th bubble source region;
the positioning of the gas leakage source also comprises the calibration of the position of the gas leakage source, namely the position of the bubble source is defined according to the position coordinate of the bubble source.
In the technical scheme, the device for identifying the leakage of the hydrogen production system based on the machine vision high precision comprises a full immersion environment system, hydrogen-containing equipment in the hydrogen production equipment, a transparent observation window, an industrial camera, a reference background and an image processing system;
the hydrogen-containing equipment in the hydrogen production equipment is positioned in the full-leaching environment system;
the transparent observation window is arranged on the full-immersion environment system;
the reference background is positioned on the full immersion environment system;
the industrial camera is arranged at the transparent observation window and is positioned in front of the reference background; the hydrogen-containing device in the hydrogen production device is positioned between the reference background and the industrial camera;
the image processing system comprises an image acquisition card, a bubble outline identification and capture AI algorithm processing unit and a non-leakage bubble screening deep learning unit;
the image acquisition card is used for acquiring and preprocessing images acquired by the industrial camera lens;
the bubble outline recognition and capture AI algorithm processing unit recognizes the leakage condition of hydrogen-containing equipment in the hydrogenation equipment in a full-immersion environment system at high precision by high-speed and high-dynamic machine vision imaging and by means of an edge recognition algorithm, and captures and recognizes leaked bubbles and leakage points by combining the comparison of a reference background;
the non-leakage bubble screening deep learning unit is used for distinguishing and screening non-leakage random bubbles and leakage hydrogen bubbles through the comparative analysis and the deep learning of the leakage hydrogen bubbles on various other types of bubbles generated by non-hydrogen leakage reasons in an observation window obtained in real time, so as to avoid false alarm and misjudgment;
the device for identifying the leakage of the hydrogen production system based on machine vision high-precision acquires image information of a hydrogen-containing device, a pipeline and a valve in a full-immersion environment system in real time through a camera, sends a real-time image to an image acquisition card, receives digital video data digitized by the camera through the image acquisition card, and stores the image in a computer memory; the bubble profile recognition and capture AI algorithm processing unit extracts and analyzes image features, the non-leakage bubble discrimination deep learning unit discriminates and eliminates non-bubble targets, and possible leakage areas are positioned and output through a leakage gas source.
In the technical scheme, the device for identifying the leakage of the hydrogen production system based on the machine vision high-precision further comprises a special light source; the special light source is positioned in the full-immersion environment system;
the device for identifying the leakage of the hydrogen production system based on the machine vision high precision also comprises an early warning device; the early warning device is arranged on the image processing system and used for sending out an early warning signal when the bubble leakage condition exceeding a safety coefficient range or a set threshold value occurs;
the early warning signal includes voice signal and image signal, and the early warning signal reports to the police according to bubble leakage condition 3 grades, is respectively: a mild leakage alarm, a moderate leakage alarm and a severe leakage alarm;
when a light leakage alarm is given, the early warning device sends out a low-frequency prompt tone and displays a blue area at the identified leakage point;
when the moderate leakage alarms, the early warning device sends out an intermediate frequency prompt tone and displays a yellow area at the identified leakage point;
when the severe leakage alarms, the early warning device sends out a high-frequency prompt tone and displays a red area at the identified leakage point.
The invention has the following advantages:
(1) The method adopts the mode of completely soaking the hydrogenation equipment in water, and identifies bubbles generated by leakage of easily leaked parts such as pipelines, joints, valves and the like of the hydrogen-containing equipment by using the images of the underwater hydrogen-containing equipment acquired by an industrial camera and by using a machine vision algorithm, thereby realizing online real-time high-precision detection and discovery of the leakage point of the hydrogenation equipment;
(2) The system realizes real-time high-precision capture and identification of tiny hydrogen leakage of a target object without contact, a built-in sensor, a lead wire and a measurement source based on a machine vision image identification algorithm by an optical system (a special light source), an image acquisition module (an industrial camera), an image processing system (an image acquisition card, a bubble outline identification and capture AI algorithm processing unit, a deep learning unit and the like), and early warning according to a set threshold;
(3) According to the invention, through the non-leakage bubble discrimination deep learning unit, various other types of bubbles generated by non-hydrogen leakage reasons in the observation window obtained in real time, such as bubbles caused by vibration, small bubbles caused by micro impurities in water and the like, are discriminated and discriminated by the contrast analysis and the deep learning of the leakage hydrogen bubbles, so that the analysis of the variation trend of output pressure pulsation by misinformation and misjudgment is avoided, and the prediction and early warning of relevant unstable or dangerous states are sent out, thereby improving the detection precision;
(4) The system provided by the invention is convenient to obtain materials, can be used as large and medium water containers such as containers and water pools, and is used for providing the full immersion underwater environment of hydrogen-containing equipment such as hydrogen production devices, hydrogen delivery pipelines, pipeline valves and the like; the transparent observation window is arranged for machine vision observation and can be made of high-strength organic glass material; the method is based on the full immersion environment, and the reference background is arranged behind the hydrogen-containing equipment and is used as the contrast background of the hydrogen-containing equipment, so that the size and the floating track of leaked bubbles can be more clearly displayed in an image or a video; the invention is provided with an early warning device which is used for sending out an early warning signal when the bubble leakage condition exceeds a safety coefficient range or a set threshold value occurs; the optical system camera is set and calibrated to be a high-resolution and low-frame-frequency monocular or monocular camera, and a color imaging camera is selected according to a monitored target, so that the detection precision of bubble leakage (including large-amount bubble leakage and water-insoluble trace bubble leakage generated by micro and initial leakage) of a hydrogen production device is improved;
the machine vision micro-bubble identification method can also be used in the occasions of online real-time identification and leakage detection of other water-insoluble gas systems except hydrogen.
Drawings
FIG. 1 is a diagram of the operation state of hydrogen production in the full immersion environment system with leakage bubbles and non-leakage bubbles;
FIG. 2 is a flow chart of moving object detection in the present invention;
FIG. 3 is a flow chart of the present invention for bubble classification based on color features;
FIG. 4 is a schematic view of the present invention for detecting and calibrating a leak point of an underwater hydrogen-containing apparatus;
FIG. 5 is a flow chart of the present invention for identifying the use of a hydrogen production system based on machine vision with high accuracy;
FIG. 6 is a flow chart of a method for identifying leakage of a hydrogen production system based on machine vision with high accuracy according to the present invention;
in fig. 1, M denotes a leakage bubble; n represents a non-leaking bubble;
the original image taken by the camera in the left image in fig. 4, and the small boxes in the left image in fig. 4 represent the bubbles identified by the present invention; the right image in fig. 4 is a binarized image recognized by a computer, black in the right image in fig. 4 is a background, white points are bubbles recognized by the present invention, which correspond to small boxes in the left image in fig. 4, and a frame line C indicates a circumscribed range where a large number of bubbles appear;
in the left diagram in fig. 4, box line a circles out a possible source of leakage air; circled by box line B is the identified source of the leak.
Detailed Description
The embodiments of the present invention will be described in detail with reference to the accompanying drawings, which are not intended to limit the present invention, but are merely exemplary. While the advantages of the invention will be clear and readily understood by the description.
The invention is a high-precision method for identifying the leakage of a hydrogen production device by combining machine vision with AI algorithm without a hydrogen detection sensor, a lead wire and a corresponding power supply, and the machine vision identification method has ultrahigh reliability and real-time property because no measuring element such as a sensor is needed; the invention provides a method for detecting and discovering leakage of hydrogen-containing equipment for hydrogen production by fully placing the relevant equipment for hydrogen production under water, and utilizing the characteristic that hydrogen is insoluble in water, and detecting and discovering leakage in time through high-precision identification of micro bubbles generated by leakage; in the full immersion mode, once any hydrogen-containing equipment leaks, even if a trace amount of hydrogen leaks, hydrogen leaking into water also appears in a bubble mode; by arranging the industrial camera, bubbles leaked into water can be quickly and accurately captured by utilizing the machine vision algorithm, and leakage points and leakage amount of hydrogen-containing equipment can be identified at the first time by screening small random bubbles which are generated due to vibration and other reasons and are not leaked, so that leakage can be found in time and alarm at the first time; the method can thoroughly avoid the risk of flammability and explosiveness of hydrogen leakage in a conventional air mode, and realize safe hydrogen production, hydrogenation, hydrogen storage and hydrogen utilization;
according to the method, by building a completely new machine vision and AI algorithm, micro leakage of hydrogen production and storage equipment in a full immersion environment can be identified on line without a hydrogen detection sensor, a lead and a power supply, and the method has the functions of real-time acquisition, long-term monitoring, first-time discovery and alarming; through deep learning of a machine, the method can effectively distinguish various non-leakage small bubbles which are sporadic due to equipment vibration or other reasons, so that high-precision identification, zero-leakage report and ultralow false report are realized; the method based on the invention can greatly improve the safety of hydrogen energy development and use, and has very important significance for promoting the industrial development of hydrogen energy and the application of green renewable energy.
With reference to the accompanying drawings: a method for identifying leakage of a hydrogen production system based on machine vision high precision comprises the steps of adopting a device for identifying leakage of the hydrogen production system based on machine vision high precision, enabling a hydrogen production device to be completely immersed in water, collecting images of underwater hydrogen-containing devices through an industrial camera, and identifying bubbles generated by tiny leakage of easily-leaked parts such as pipelines, joints, valves and the like of the underwater hydrogen-containing devices by utilizing a machine vision algorithm, so that leakage points and leakage conditions of the hydrogen production device are detected and identified in real time on line;
the method comprises the following steps:
the method comprises the following steps: detecting a moving target;
after detecting the moving bubbles by using the VIBE algorithm, further checking whether the moving target is a real bubble;
step two: non-bubble target suppression based on multi-feature fusion;
a multi-feature classifier is designed according to features such as bubble color, texture and shape, false targets are eliminated, and the specific method comprises the following steps:
s21: extracting color features of the moving target;
s22: extracting the shape characteristics of the moving object (the method for extracting the shape characteristics of the moving object is the prior art);
s23: extracting HOG texture features of a moving target;
s24: acquiring a strong classifier through an Adaboost algorithm, and judging whether the moving object acquired in the first step is a bubble or a background;
then, detecting a bubble characteristic diagram based on a single-stage multi-scale YOLOv3-SPP network;
the network structure diagram of YOLOv3-SPP is mainly divided into three parts: (1) Backbone network, in YOLOv3-SPP, darknet53 is used as backbone network; (2) Spatial Pyramid (SPP) (2) multi-scale target detection network, YOLOv3-SPP uses three detection scales, namely detection scale 1 (the size is 1/32 of the input image), detection scale 2 (the size is 1/16 of the input image) and detection scale 3 (the size is 1/8 of the input image) in the graph;
the specific method for detecting the bubble characteristic diagram based on the single-stage multi-scale YOLOv3-SPP network comprises the following steps:
performing up-sampling on the feature map output by the SPP and performing feature fusion on the feature map with the corresponding scale generated by the backbone network to obtain the final feature maps with three detection scales;
step three: positioning a gas leakage source;
in the invention, due to the action of buoyancy, the generated bubbles usually present a state of uniform linear upward motion, so that firstly, a moving target in a scene is preliminarily positioned in a moving target background modeling mode in the step one, and then, in the step two, the size, the color test and the texture information of the bubbles are further determined; training a deep learning classifier so as to eliminate non-bubble targets in a motion scene; then, detecting the bubble characteristic diagram based on a single-stage multi-scale YOLOv3-SPP network to obtain characteristic diagrams of three detection scales, and detecting, distinguishing and positioning bubbles with different scales based on the characteristic diagrams; finally, positioning a gas leakage source by adopting a mathematical method based on the detected bubble characteristics in the third step; the invention realizes the real-time high-precision identification under various leakage conditions (such as trace leakage and large amount of leakage) in the hydrogen preparation and addition system through the steps executed in sequence.
Further, in step one, the moving object detection is color image VIBE moving object detection (the moving object detection method is the prior art);
the method for detecting the moving target specifically comprises the following steps:
s11: reading a video stream image;
s12: confirming whether the current image is a first frame image (namely a first picture in the video);
if yes, initializing an RGB-VIBE model, and jumping to the step S11;
if not, jumping to the step S13;
s13: updating the RGB-VIBE model;
s14: detecting a moving target by adopting a frame difference method;
s15: and outputting the position of the moving target.
Further, in step S21, the color features of the moving object are extracted, and the bubbles are classified based on the color features, and the specific method includes:
s211: acquiring a target image;
s212: color space conversion;
when calculating the similarity of two images, it is common to extract their color features and compare them in a specific color space; the purpose of the color space is to represent colors by using primary colors according to a certain standard, such as RGB, HSI, HSV, etc.; the similarity of RGB color spaces cannot represent the similarity of colors; for example, the RGB color on the query image is (200, 150,0) and the RGB color of the image library image is (200, 200,0), which are very similar in RGB color space, but very different in color; the HSI and HSV color space have no problem in the aspect, and are very suitable for distinguishing by naked eyes of people, and better reflect the perception and identification capability of the people to the color; therefore, the present invention herein employs the HSI color space, and the image can be converted from the RGB color space to the HSI color space by the following formula:
s213: color quantization;
the color types of an image are usually very many, and if the similarity is directly calculated, a large feature storage space is consumed; experiments show that the increase of the dimension of the color histogram can effectively improve the retrieval accuracy, but when the dimension is increased to a certain degree, the improvement of the retrieval accuracy is small and possibly reduced; if the colors are properly quantized and then calculated, the calculated amount is much less, and the calculation efficiency is improved; the color quantization is to quantize the H, S and I3 components at unequal intervals according to human color perception and then analyze and calculate a large amount of color models; therefore, the present invention quantizes the HSI color space, quantizes the chroma into 8 spaces, divides the saturation into 3 spaces, and divides the luminance into 3 spaces, i.e., the color space is divided into 72 intervals, and the specific quantization values are as follows:
s214: extracting a statistical histogram;
the image has a plurality of description modes, and a color histogram is one of the description modes; the color histogram describes the distribution of image colors over the color space; two common histograms are found: counting the histogram and accumulating the histogram;
an image can be described by means of a statistical histogram of image features; the statistical histogram of the image features is a one-dimensional discrete function, as shown in the following formula:
wherein: k represents the feature value of the image, L is the number of the feature values, n k The number of pixels with a characteristic value of k in the image, and N is the total number of the image pixels;
color histograms have many advantages: after the image is rotated, the color histogram of the image is not changed; the color histograms are easy to extract, and the similarity between the two histograms is easy to calculate;
in the statistical histogram, after the HSI space is quantized, the value range of H is [0, 1, …, 7], the value range of S is [0, 1, 2], and the value range of I is [0, 1, 2]; synthesizing HSI into a one-dimensional characteristic vector by the following formula, and distributing three components of H, S and I on the one-dimensional vector; in the formula, the hue H takes a weight of 9, the saturation weight of 3, and the brightness weight of 1; because the hue contains most information, the weight of the hue is larger, and the weight of the saturation and the brightness is smaller; obtaining the value range of G as [0, 1, …, 71] according to a formula;
calculating similarity based on the image features in the feature library and the statistical histogram in step S214 (the calculation method for calculating similarity by using statistical histogram is the prior art);
directly judging whether the target image is a bubble or a background according to the result of the similarity calculation;
the similarity calculation result and the bubble in the feature library have high similarity and are bubbles;
the result of similarity calculation is the background with high similarity to the background in the feature library.
Further, in step S22, the method for extracting the shape feature of the moving object includes:
the methods for describing shape feature parameters mainly have two types: region-based feature parameters and boundary-based feature parameters; the regional characteristic parameters are mainly the description of shape characteristic parameters obtained through the collection of all pixel points in the region; the parameters can be geometric parameters, density parameters, two-dimensional transformation (such as Fourier transformation and wavelet transformation) coefficients or energy spectrums of regions and the like; for the description of the shape characteristics, a moment invariant method, a Fourier descriptor, an edge histogram method, a wavelet important coefficient method, a wavelet profile representation method, a geometric parameter method and the like are typical;
1) Area characteristics;
the area of a region is a basic feature of a region, which describes the size of the region; for digital images, the area of a region is defined as the number of pixel points in the region;
2) A dimensional characteristic;
the length and width of the target are defined as the length and width of the minimum circumscribed rectangle of the target;
3) Squareness degree;
the squareness is defined as the ratio of the area of the object to the area of the smallest circumscribing rectangle (MER) of the object; the degree of rectangularity reflects the filling degree of an object in a minimum external rectangle, the degree of rectangularity of the rectangle is 1, and the degree of rectangularity of a circle is
The squareness of the triangle is 0.5; for other shapes, the range of the rectangular degree is (0,1); rectangular, circular and irregular shapes can be distinguished by using the rectangular degree;
4) Circularity;
the circularity reflects the degree of an object approaching a circular shape, and is also called Compactness (Compactness), and is defined as the ratio of the area a of a region 4 pi times to the square of the perimeter P (there is a document defined as the ratio of the square of the perimeter to the area of a region 4 pi times).
Further, in step S23, a HOG (Histogram of Oriented Gradient) feature is a local region descriptor, which is a feature that is constructed by calculating a Gradient direction Histogram on a local region and can well describe the edge of the human body; it is not sensitive to illumination variations and small amounts of drift; procedure for Hog feature extraction proposed by Dalal: dividing a sample image into a plurality of units (cells) of pixels, averagely dividing the gradient direction into 9 intervals (bins), carrying out histogram statistics on the gradient directions of all the pixels in each direction interval in each unit to obtain a 9-dimensional feature vector, forming a block (block) by every adjacent 4 units, connecting the feature vectors in one block to obtain a 36-dimensional feature vector, and scanning the sample image by using the block, wherein the scanning step length is one unit; finally, the characteristics of all the blocks are connected in series to obtain the characteristics of the human body; for example, for a 64 × 128 image, every 2 × 2 unit (16 × 16 pixels) constitutes a block, with 4 × 9=36 features in each block, with 8 pixels as a step size, then there will be 7 scan windows in the horizontal direction and 15 scan windows in the vertical direction; that is, 64 × 128 pictures, have 36 × 7 × 15=3780 features in total;
the method for extracting the HOG texture features of the moving object comprises the following specific steps:
s231: converting the input color image into a gray image;
s232: standardizing (normalizing) the color space of the input image by using a Gamma correction method; the method aims to adjust the contrast of the image, reduce the influence caused by local shadow and illumination change of the image and inhibit the interference of noise;
s233: calculating a gradient; the purpose is to capture profile information while further mitigating the interference of illumination;
s234: projecting the gradient to a gradient direction of the cell; the object is to provide a coding for local image areas;
s235: normalizing all cells on the block; normalization allows further compression of lighting, shadows, and edges, and typically each cell is shared by multiple different blocks, but its normalization is based on different blocks, so the computation results are also different; thus, the features of one cell may appear in the final vector multiple times with different results; the block descriptor after normalization is called HOG descriptor;
s236: HOG characteristics of all blocks in a detection space are collected;
this step is to collect the HOG features of all the overlapped blocks in the detection window and combine them into the final feature vector for classification.
Step one, in step S24, adaboost is an iterative algorithm, and the core idea is to train different classifiers (weak classifiers) for the same training set, and then to assemble these weak classifiers to form a stronger final classifier (strong classifier); the algorithm is realized by changing data distribution, and the weight of each sample is determined according to whether the classification of each sample in each training set is correct and the accuracy of the last overall classification; sending the new data set with the modified weight value to a lower-layer classifier for training, and finally fusing classifiers obtained by each training as a final decision classifier; the Adaboost classifier is used to eliminate some unnecessary training data features and put the key on the key training data;
different training sets in the Adaboost algorithm are realized by adjusting the weight corresponding to each sample; at the beginning, the corresponding weight of each sample is the same, namely n is the number of samples, and a weak classifier is trained under the sample distribution; for the samples with wrong classification, the corresponding weight is increased; for the correctly classified samples, the weight is reduced, so that the wrongly classified samples are highlighted, and a new sample distribution is obtained; training the weak classifiers again under the new sample distribution to obtain the weak classifiers; by analogy, pass through
A sub-cycle is obtained>
A weak classifier for classifying the received signal>
Superposing (boost) the weak classifiers according to a certain weight to obtain a strong classifier which is finally needed;
obtaining a strong classifier through an Adaboost algorithm, and judging whether the moving object obtained in the first step is a bubble or a background, wherein the specific method comprises the following steps:
provided with an inputnEach training sample was:
in which>
Is an input training sample, is>
Respectively representing a positive sample and a negative sample, wherein the number of positive samples is->
The number of negative samples is->
，/>
(ii) a Wherein the training sample is a sample of air bubbles or background;
s241: initializing weights for each sample
；
S242: for each one
，/>
Number of weak classifiers:
(1) normalizing the weights to a probability distribution;
wherein: wherein w is a weight, and subscript i, j indicates that the indices all belong to the interval of 1, … …, n; d (i) indicates that i belongs to one sample interval; n is the number of training samples;
(2) for each feature
Training a weak classifier>
Calculating the weighted error rate of the weak classifiers corresponding to all the features; weighted error rate->
The calculation formula of (2) is as follows:
wherein: w (x) is a function of the calculated weights;
(3) selecting the best weak classifier
(having a minimum error Rate>
) ；
(4) Adjusting the weight according to the optimal weak classifier;
wherein:
indicates that it is correctly classified>
Indicating that it was misclassified;
adjustment factor
Can be based on>
Calculating;
s243: obtaining a final strong classifier;
the final strong classifier is:
wherein: h (x) is a strong classifier;
represents a weak classifier;
superimposed weight coefficients
Can be based on>
Calculating;
and judging the bubbles or the background through the numerical value calculated by the strong classifier, wherein the bubbles are taken when h (x) is 1, and the background is taken when h (x) is 0.
Further, the feature maps of the three detection scales are generated in the following specific manner:
detection scale 1: the feature map output by the SPP is a feature map of a detection scale 1; the feature map size of the detection scale 1 is 1/32 of that of the input image, namely, the input image is subjected to down-sampling by 32 times to obtain the feature map of the detection scale 1; the characteristic diagram of the detection scale 1 has a large receptive field and is suitable for detecting objects with large scale;
detection scale 2: the feature map output by the 92 th layer of the network is up-sampled (the up-sampling adopts a bilinear interpolation method), and then the feature map is spliced with the feature map output by the 62 nd layer of Darknet53 to obtain a feature map of a detection scale 2; detecting the size of the receptive field of the feature map of the scale 2, and being suitable for detecting the targets with the medium size;
detection scale 3: the output characteristic diagram of the 104 th layer of the network is up-sampled and then is subjected to characteristic splicing with the characteristic diagram output by the 37 th layer of Darknet53, so that a characteristic diagram of a detection scale 3 is obtained; the detection scale 3 is small in the receptive field of the characteristic diagram, and is suitable for detecting the target with small size.
Step one, in step three, the leakage air source is positioned, and the specific method comprises the following steps:
carrying out vertical projection on the detected bubble image region, then selecting a horizontal line and a vertical line in a projection histogram, setting the horizontal line and the vertical line as coordinate axes, and acquiring the center coordinates of all bubbles
；
Firstly, classifying the circle center coordinates of the bubbles by adopting a kmeans clustering method aiming at the circle center coordinates of the bubbles;
the moving direction of the bubbles is vertical to the upper part from the air leakage source, and the number of the bubbles is gradually reduced; therefore, the similar sizes of the abscissas are selected as clustering characteristics, the centers of the bubbles with the characteristics are divided into a class, and the class is identified as the bubbles generated in the same bubble source region;
after the number of the bubble source regions is divided, each bubble source is respectively positioned, and as the air leakage sources are distributed below all bubbles, the position of the bubble source can be confirmed according to the characteristics, and the specific method comprises the following steps:
assuming that the number of the bubble sources is K, the circle center coordinate of the bubble in the kth bubble source region is defined as
The position of the bubble source can be determined according to the following formula, and the coordinates of the position of the bubble source can be obtained:
wherein the coordinates of the center of the circle are defined as
；（/>
，/>
) To representThe coordinates of the leakage air source are solved; i represents the ith leakage gas source; x represents the abscissa; y represents the ordinate; />
Indicating the number of the identified bubbles in the k-th bubble source region;
the positioning of the gas leakage source also comprises the calibration of the position of the gas leakage source, namely the position of the bubble source is defined according to the position coordinate of the bubble source.
Further, the device for identifying the leakage of the hydrogen production system based on machine vision high precision comprises a full immersion environment system, hydrogen-containing equipment in the hydrogen production equipment, a transparent observation window, an industrial camera, a reference background and image processing system, a power supply and a control cable; the implementation system of the method for identifying the leakage of the hydrogen production system based on machine vision high precision mainly comprises a full-immersion operation environment, an image acquisition module, an image processing system, an early warning and other state output interaction interfaces;
the hydrogen-containing equipment in the hydrogen production equipment is positioned in the full-immersion environmental system; the full-immersion environment system can be large and medium water containers such as containers, pools and the like and is used for providing a full-immersion underwater environment for hydrogen-containing equipment such as hydrogen production devices, hydrogen conveying pipelines, pipeline valves and the like; the hydrogen-containing equipment in the hydrogen production equipment is equipment for producing hydrogen, hydrogenating, storing hydrogen, conveying hydrogen and using a hydrogen system to contain hydrogen with a certain pressure, and comprises a hydrogen production device and corresponding pipelines and valves thereof;
the transparent observation window is arranged on the full-immersion environment system; the transparent observation window is a machine vision observation window, is a special observation window for vision monitoring on the full-immersion environment system, and can be made of high-strength organic glass material;
the reference background is positioned on the full immersion environment system; the reference background is used for highlighting, and the contrast background of the equipment is provided, so that the size and the floating track of the leaked bubbles can be more clearly shown in an image or a video;
the industrial camera is arranged at the transparent observation window and is positioned in front of the reference background; the hydrogen-containing device in the hydrogen production device is positioned between the reference background and the industrial camera; the industrial camera is a monocular or multi-view camera, an area-array camera or a line-scan camera can be selected, and a high-dynamic full-color camera is selected for capturing and identifying bubbles leaked from underwater equipment in the observation window; the resolution of the camera is calculated according to the actual image width and precision requirements, so that 1920 pixels × 1080 pixels with higher resolution are selected to meet the requirements of high precision and dynamic identification; generally, the long-term status monitoring is adopted, and when the tiny leakage begins to occur, the bubbles are small, the floating speed is high, so the frame rate can be selected to be a high frame rate, such as 20fps; because the camera in the invention is generally fixed in position, the lens can adopt fixed focus and fixed aperture;
the power supply and the control cable are power supplies of non-contact equipment in a full-immersion environment, and the power supply and the cable for the non-sensor are used for connecting the power cable access and the control cable connection of the industrial camera and the supplementary lighting equipment;
the image processing system comprises an image acquisition card, a bubble outline identification and capture AI algorithm processing unit and a non-leakage bubble screening deep learning unit;
the image acquisition card is used for acquiring and preprocessing images acquired by the industrial Camera lens, the data interface determines the transmission bandwidth according to the resolution and the frame rate, and the USB3.0, camera Link or GigE interface can be selected by combining the transmission distance; preferably a USB3.0 interface;
the bubble outline recognition and capture AI algorithm processing unit recognizes the leakage condition of hydrogen-containing equipment in the hydrogenation equipment in a full-immersion environment system at high precision by high-speed and high-dynamic machine vision imaging and by means of an edge recognition algorithm, and captures and recognizes leaked bubbles and leakage points by combining the comparison of a reference background;
the non-leakage bubble discrimination deep learning unit is used for carrying out contrastive analysis and deep learning on various other types of bubbles generated by non-hydrogen leakage reasons in an observation window obtained in real time, such as bubbles caused by vibration, small bubbles caused by tiny impurities in water and the like, so as to distinguish and discriminate non-leakage random bubbles from leakage hydrogen bubbles and avoid misinformation and misjudgment;
the device for identifying the leakage of the hydrogen production system based on machine vision high-precision acquires image information of a hydrogen-containing device, a pipeline and a valve in a full-immersion environment system in real time through a camera, sends a real-time image to an image acquisition card, receives digital video data digitized by the camera through the image acquisition card, and stores the image in a computer memory; the bubble outline recognition and capture AI algorithm processing unit extracts and analyzes image features, the non-leakage bubble discrimination deep learning unit discriminates and eliminates non-bubble targets, and possible leakage areas are positioned and output through a leakage gas source;
the invention realizes moving target detection and non-bubble target inhibition based on multi-feature fusion through the bubble contour recognition and capture AI algorithm processing unit, is used for distinguishing background and bubbles and improving the detection precision of the bubbles, realizes bubble feature map detection based on a single-stage multi-scale YOLOv3-SPP network through the non-leakage bubble discrimination deep learning unit, is used for distinguishing leakage bubbles from non-leakage bubbles, and is also used for accurately detecting trace bubble leakage under the conditions of lower movement rate, smaller quantity, smaller shape and the like, and the detection range of the invention is improved, so that the invention is not only suitable for leakage detection of a large quantity of bubbles, but also suitable for leakage detection of trace bubbles, and further improves the detection range and the detection precision of the invention.
Further, the device for identifying the leakage of the hydrogen production system based on the machine vision high-precision also comprises a special light source; the special light source is positioned in the full-immersion environment system, is used for internal illumination in the full-immersion environment and is used for supplementary illumination only during night identification;
the leaked bubbles can be identified under the indoor normal lighting condition, the special light source considers necessary supplementary light sources to improve the identification accuracy and precision, and according to the object condition of an identification target (such as a pipeline, a valve, an electrolytic tank and the like), a front surface light source or a front side light source can be selected, and the influence of reflection light on machine vision identification is avoided; the light source is provided with a lighting controller, and the brightness of the light source is adjusted according to the field environment;
the device for identifying the leakage of the hydrogen production system based on the machine vision high precision also comprises an early warning device; the early warning device is arranged on the image processing system and used for sending out an early warning signal when the bubble leakage condition exceeding a safety coefficient range or a set threshold value occurs;
the early warning signal includes voice signal and image signal, and the early warning signal reports to the police according to bubble leakage condition 3 grades, is respectively: a mild leakage alarm, a moderate leakage alarm and a severe leakage alarm;
when a light leakage alarm is given, the early warning device sends out a low-frequency prompt tone and displays a blue area at the identified leakage point;
when the moderate leakage alarms, the early warning device sends out an intermediate frequency prompt tone and displays a yellow area at the identified leakage point;
when the severe leakage alarms, the early warning device sends out a high-frequency prompt tone and displays a red area at the identified leakage point.
Other parts not described belong to the prior art.
Claims (7)
1. A method for identifying leakage of a hydrogen production system based on machine vision high precision is characterized by comprising the following steps: the method comprises the steps of adopting a device for identifying the leakage of a hydrogen production system with high precision based on machine vision, fully immersing hydrogen production equipment in water, acquiring images of underwater hydrogen-containing equipment by an industrial camera, and identifying bubbles generated by the leakage of the easily-leaked part of the underwater hydrogen-containing equipment by utilizing a machine vision algorithm, so as to detect and identify the leakage point and the leakage condition of the hydrogen production device in real time on line;
the method comprises the following steps:
the method comprises the following steps: detecting a moving target;
step two: non-bubble target suppression based on multi-feature fusion;
a multi-feature classifier is designed according to the features of the color, the texture and the shape of the bubble, and the false target is eliminated, wherein the specific method comprises the following steps:
s21: extracting color features of the moving target;
s22: extracting the shape characteristics of the moving object;
s23: extracting HOG texture features of a moving target;
s24: acquiring a strong classifier through an Adaboost algorithm, and judging whether the moving object acquired in the first step is a bubble or a background;
step three: positioning a gas leakage source;
in the third step, the air leakage source is positioned, and the specific method comprises the following steps:
carrying out vertical projection on the detected bubble image region, then selecting a horizontal line and a vertical line in a projection histogram, setting the horizontal line and the vertical line as coordinate axes, and acquiring the center coordinates of all bubbles
；
Firstly, classifying the circle center coordinates of the bubbles by adopting a kmeans clustering method aiming at the circle center coordinates of the bubbles;
after the number of the bubble source regions is divided, each bubble source is respectively positioned, and the specific method comprises the following steps:
assuming that the number of the bubble sources is K, the center coordinates of the bubbles in the kth bubble source area are defined as
Determining the position of the bubble source according to the following formula, and obtaining the coordinates of the position of the bubble source:
wherein the coordinates of the center of the circle are defined as
；（/>
，/>
) Representing the coordinates of a to-be-solved leakage air source; i represents the ith leakage gas source; />
Indicating the number of bubbles identified in the k-th bubble source region；
The positioning of the gas leakage source also comprises the calibration of the position of the gas leakage source, namely the position of the bubble source is defined according to the position coordinate of the bubble source.
2. The machine-vision-based high-precision method for identifying leaks in a hydrogen production system according to claim 1, wherein the method comprises the following steps: in the first step, the moving object is detected, and the specific method comprises the following steps:
s11: reading a video stream image;
s12: confirming whether the current image is a first frame image;
if yes, initializing an RGB-VIBE model, and jumping to the step S11;
if not, jumping to the step S13;
s13: updating the RGB-VIBE model;
s14: detecting a moving target by adopting a frame difference method;
s15: and outputting the position of the moving target.
3. The machine vision-based high-precision method for identifying leaks in hydrogen production systems according to claim 1 or 2, wherein the method comprises the following steps: in step S21, a color feature of the moving object is extracted, and the bubbles are classified based on the color feature, and the specific method includes:
s211: acquiring a target image;
s212: color space conversion;
using the HSI color space, the image is converted from the RGB color space to the HSI color space by the following formula:
s213: color quantization;
the HSI color space is quantized, the chroma is quantized into 8 spaces, the saturation is divided into 3 spaces, the brightness is divided into 3 spaces, namely, the color space is divided into 72 intervals, and specific quantization values are as follows:
s214: extracting a statistical histogram;
the extraction formula of the statistical histogram of the image features is as follows:
wherein: k is a radical of 0 Representing the feature value of the image; l is the number of characteristic obtainable values;
for images having a characteristic value of k 0 The number of pixels of (a); n is the total number of image pixels.
4. The machine-vision-based high-precision method for identifying leaks in hydrogen production systems according to claim 3, wherein the method comprises the following steps: in step S23, the method for extracting the texture feature of the moving object HOG includes:
s231: converting the input color image into a gray image;
s232: carrying out color space standardization on an input image by adopting a Gamma correction method;
s233: calculating a gradient; capturing contour information, and further weakening the interference of illumination;
s234: projecting the gradient to a gradient direction of the cell; providing a code for the local image area;
s235: normalizing all cells on the block;
s236: collecting and obtaining HOG characteristics of all blocks in a detection space;
all overlapping blocks in the detection window are collected for HOG features and combined into a final feature vector for classification.
5. The machine-vision-based high-precision method for identifying leaks in hydrogen production systems according to claim 4, wherein the method comprises the following steps: in step S24, a strong classifier is obtained through an Adaboost algorithm, and the moving object obtained in the first step is determined to be a bubble or a background, and the specific method is as follows:
provided with an inputnEach training sample was:
wherein->
Is an input of a training sample of the training,
respectively represent positive and negative samples, wherein the number of positive samples is->
The number of negative samples is->
，/>
(ii) a Wherein the training sample is a sample of air bubbles or background;
s241: initializing weights for each sample
；
S242: for each one
，/>
Number of weak classifiers:
(1) normalizing the weights to a probability distribution;
wherein w is a weight; d (i) indicates that i belongs to a sample interval; n is the number of training samples;
(2) for each feature
Training a weak classifier>
Calculating the weighted error rate of the weak classifiers corresponding to all the characteristics; weighted error rate->
The calculation formula of (c) is:
wherein: w (x) is a function of the calculated weights;
(3) selecting the best weak classifier
：
(4) Adjusting the weight according to the optimal weak classifier;
wherein
Indicates that it is correctly classified>
Indicating that it was misclassified; />
Adjustment factor
According to>
Calculating;
s243: obtaining a final strong classifier;
the final strong classifier is:
wherein: h (x) is a strong classifier;
represents a weak classifier;
superimposed weight coefficients
Can be based on>
Calculating;
and judging the bubble or the background through the numerical value calculated by the strong classifier, wherein the bubble is taken as h (x) when 1 is taken, and the background is taken as h (x) when 0 is taken.
6. The machine-vision-based high-precision method for identifying leaks in hydrogen production systems according to claim 5, wherein the method comprises the following steps: the device for identifying the leakage of the hydrogen production system based on machine vision high precision comprises a full immersion environment system, hydrogen-containing equipment in the hydrogen production equipment, a transparent observation window, an industrial camera, a reference background and an image processing system;
the hydrogen-containing equipment in the hydrogen production equipment is positioned in the full-immersion environmental system;
the transparent observation window is arranged on the full-immersion environment system;
the reference background is positioned on the full immersion environment system;
the industrial camera is arranged at the transparent observation window and is positioned in front of the reference background; the hydrogen-containing device in the hydrogen production device is positioned between the reference background and the industrial camera;
the image processing system comprises an image acquisition card, a bubble outline identification and capturing AI algorithm processing unit and a non-leakage bubble screening deep learning unit;
the image acquisition card is used for acquiring and preprocessing images acquired by the industrial camera lens;
the bubble outline recognition and capture AI algorithm processing unit recognizes the leakage condition of hydrogen-containing equipment in the hydrogenation equipment in a full-immersion environment system at high precision by high-speed and high-dynamic machine vision imaging and by means of an edge recognition algorithm, and captures and recognizes leaked bubbles and leakage points by combining the comparison of a reference background;
the non-leakage bubble screening deep learning unit is used for distinguishing and screening non-leakage random bubbles and leakage hydrogen bubbles through the comparative analysis and the deep learning of the leakage hydrogen bubbles on various other types of bubbles generated by non-hydrogen leakage reasons in an observation window obtained in real time, so as to avoid false alarm and misjudgment;
the device for identifying the leakage of the hydrogen production system based on machine vision high-precision acquires image information of a hydrogen-containing device, a pipeline and a valve in a full-immersion environment system in real time through a camera, sends a real-time image to an image acquisition card, receives digital video data digitized by the camera through the image acquisition card, and stores the image in a computer memory; the bubble profile recognition and capturing AI algorithm processing unit extracts and analyzes image features, the non-leakage bubble discrimination deep learning unit discriminates and eliminates non-bubble targets, and a leakage area is positioned and output through a leakage gas source.
7. The machine-vision-based high-precision method for identifying leaks in hydrogen production systems according to claim 6, wherein the method comprises the following steps: the device for identifying the leakage of the hydrogen production system based on the machine vision high precision also comprises a special light source; the special light source is positioned in the full-immersion environment system;
the device for identifying the leakage of the hydrogen production system based on the machine vision high precision also comprises an early warning device; the early warning device is arranged on the image processing system and used for sending out an early warning signal when the bubble leakage condition exceeding a safety coefficient range or a set threshold value occurs; the early warning signal includes voice signal and image signal, and the early warning signal reports to the police according to bubble leakage condition 3 grades, is respectively: a mild leakage alarm, a moderate leakage alarm and a severe leakage alarm; when a light leakage alarm is given, the early warning device sends out a low-frequency prompt tone and displays a blue area at the identified leakage point; when the moderate leakage alarms, the early warning device sends out an intermediate frequency prompt tone and displays a yellow area at the identified leakage point; when the severe leakage alarms, the early warning device sends out a high-frequency prompt tone and displays a red area at the identified leakage point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211680630.9A CN115661151B (en) | 2022-12-27 | 2022-12-27 | Method for identifying leakage of hydrogen production system based on machine vision high precision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211680630.9A CN115661151B (en) | 2022-12-27 | 2022-12-27 | Method for identifying leakage of hydrogen production system based on machine vision high precision |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115661151A CN115661151A (en) | 2023-01-31 |
| CN115661151B true CN115661151B (en) | 2023-03-28 |
Family
ID=85023087
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211680630.9A Active CN115661151B (en) | 2022-12-27 | 2022-12-27 | Method for identifying leakage of hydrogen production system based on machine vision high precision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115661151B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116958503B (en) * | 2023-09-19 | 2024-03-12 | 广东新泰隆环保集团有限公司 | Image processing-based sludge drying grade identification method and system |
| CN117893643B (en) * | 2024-03-18 | 2024-06-11 | 四川弘和数智集团有限公司 | Method, device, equipment and medium for generating gas leakage image of oil and gas station compressor |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103839279A (en) * | 2014-03-18 | 2014-06-04 | 湖州师范学院 | Adhesion object segmentation method based on VIBE in object detection |
| CN112085713A (en) * | 2020-08-28 | 2020-12-15 | 格力电器（芜湖）有限公司 | Method, device, equipment and system for automatically detecting leakage point |
| AU2021401329A1 (en) * | 2020-12-16 | 2023-07-13 | Draeger Medical Systems, Inc. | Using machine learning algorithms to automatically determine exposure in gas detection tubes |
-
2022
- 2022-12-27 CN CN202211680630.9A patent/CN115661151B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN115661151A (en) | 2023-01-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115661151B (en) | Method for identifying leakage of hydrogen production system based on machine vision high precision | |
| CN111723748B (en) | Infrared remote sensing image ship detection method | |
| CN109871902B (en) | SAR small sample identification method based on super-resolution countermeasure generation cascade network | |
| CN109308447A (en) | The method of equipment operating parameter and operating status is automatically extracted in remote monitoriong of electric power | |
| CN112861635B (en) | Fire disaster and smoke real-time detection method based on deep learning | |
| CN110598693A (en) | Ship plate identification method based on fast-RCNN | |
| CN113361326B (en) | Wisdom power plant management and control system based on computer vision target detection | |
| CN113869162A (en) | Violation identification method and system based on artificial intelligence | |
| CN113947570B (en) | Crack identification method based on machine learning algorithm and computer vision | |
| CN109308448A (en) | A method of it prevents from becoming distribution maloperation using image processing techniques | |
| CN113033315A (en) | Rare earth mining high-resolution image identification and positioning method | |
| CN117557775B (en) | Substation power equipment detection method and system based on infrared and visible light fusion | |
| CN115761537A (en) | Power transmission line foreign matter intrusion identification method oriented to dynamic characteristic supplement mechanism | |
| CN114998783A (en) | Front-end equipment for video analysis of smoke, fire and personnel behaviors | |
| CN115171006B (en) | Detection method for automatically identifying person entering electric power dangerous area based on deep learning | |
| CN116403085A (en) | Multi-pointer instrument identification method and device for transformer substation based on multitasking | |
| CN114241189B (en) | Ship black smoke recognition method based on deep learning | |
| CN114820676A (en) | Equipment running state identification method and device | |
| CN114998248A (en) | Automatic identification method for defect images of power transmission and transformation equipment | |
| Vijaylaxmi et al. | Fire detection using YCbCr color model | |
| CN109886133A (en) | A kind of ship detection method and system based on remote sensing image | |
| CN113420724B (en) | Unmanned aerial vehicle remote sensing river inlet and outlet recognition method and device | |
| CN118072205B (en) | Unmanned aerial vehicle aerial photography small target detection method based on feature transformation and sample optimization | |
| CN113971755B (en) | All-weather sea surface target detection method based on improved YOLOV model | |
| CN118552766A (en) | Dangerous chemical plant fire detection method based on mixed Gaussian background model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |