CN114187282A - Grate cooler bed thickness measuring method based on image segmentation - Google Patents
Grate cooler bed thickness measuring method based on image segmentation Download PDFInfo
- Publication number
- CN114187282A CN114187282A CN202111534823.9A CN202111534823A CN114187282A CN 114187282 A CN114187282 A CN 114187282A CN 202111534823 A CN202111534823 A CN 202111534823A CN 114187282 A CN114187282 A CN 114187282A
- Authority
- CN
- China
- Prior art keywords
- image
- model
- grate cooler
- material layer
- segmentation
- 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
- 238000003709 image segmentation Methods 0.000 title claims abstract description 55
- 238000000034 method Methods 0.000 title claims abstract description 39
- 239000000463 material Substances 0.000 claims abstract description 57
- 238000012545 processing Methods 0.000 claims abstract description 29
- 238000010191 image analysis Methods 0.000 claims abstract description 14
- 238000001914 filtration Methods 0.000 claims abstract description 10
- 238000013135 deep learning Methods 0.000 claims abstract description 9
- 238000013528 artificial neural network Methods 0.000 claims abstract description 7
- 238000005457 optimization Methods 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 30
- 230000011218 segmentation Effects 0.000 claims description 27
- 238000005516 engineering process Methods 0.000 claims description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 12
- 238000012805 post-processing Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 8
- 239000004576 sand Substances 0.000 claims description 8
- 238000000889 atomisation Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 7
- 238000001816 cooling Methods 0.000 claims description 6
- 238000013210 evaluation model Methods 0.000 claims description 6
- 238000002372 labelling Methods 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 6
- 238000011068 loading method Methods 0.000 claims description 5
- HPTJABJPZMULFH-UHFFFAOYSA-N 12-[(Cyclohexylcarbamoyl)amino]dodecanoic acid Chemical compound OC(=O)CCCCCCCCCCCNC(=O)NC1CCCCC1 HPTJABJPZMULFH-UHFFFAOYSA-N 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 4
- 238000003708 edge detection Methods 0.000 claims description 4
- 230000001133 acceleration Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 230000006835 compression Effects 0.000 claims description 3
- 238000007906 compression Methods 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 238000013139 quantization Methods 0.000 claims description 3
- 238000007619 statistical method Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000013138 pruning Methods 0.000 claims description 2
- 238000004458 analytical method Methods 0.000 abstract description 5
- 238000012546 transfer Methods 0.000 abstract description 2
- 230000007613 environmental effect Effects 0.000 abstract 1
- 238000005259 measurement Methods 0.000 description 5
- 239000004568 cement Substances 0.000 description 3
- 238000010304 firing Methods 0.000 description 3
- 238000009499 grossing Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 230000000087 stabilizing effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 239000003245 coal Substances 0.000 description 2
- 238000005520 cutting process Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 238000000691 measurement method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- RWSOTUBLDIXVET-UHFFFAOYSA-N Dihydrogen sulfide Chemical compound S RWSOTUBLDIXVET-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007664 blowing Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000005314 correlation function Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000004821 distillation Methods 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000035699 permeability Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 239000007921 spray Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- 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
- G06T2207/30132—Masonry; Concrete
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a grate cooler bed thickness measuring method based on image segmentation, which comprises the following steps: firstly, marking material layer information inside the grate cooler shot by a high-temperature camera, inputting the marked image into a deep learning image segmentation model, and solving neural network parameters through optimization to obtain an analysis model; then deploying the model to an image analysis platform, predicting the material layer area of the grate cooler in real time, extracting the lower edge information of the image after binarization processing by performing Gaussian filtering and binarization processing on the prediction result, and dividing the edge height by the pixel height of the image to obtain the relative thickness of the material layer; and finally, the relative thickness of the material layer is used as a controlled variable and is transmitted into a control loop, and the running speed of the grate cooler is automatically adjusted, so that the material layer thickness of the grate cooler is in a stable range as thick as possible without eruption blow-through, thereby improving the heat transfer efficiency of the grate cooler and increasing the economic and environmental benefits.
Description
Technical Field
The invention relates to a grate cooler bed thickness measuring method based on image segmentation.
Background
The grate cooler is a key device in the cement production process, the thickness of a material layer of the grate cooler is important for the grate cooler, and if the material layer is too thin, the time for cooling air to pass through the material layer is too short, and the heat transfer efficiency is not high; the material bed is too thick, so that the air permeability of the material bed is influenced, and further local spray-shaped thorough blowing occurs, so that most of cooling air is blown away from a spray position, the purpose of sufficient heat exchange with clinker is not achieved, and the heat exchange efficiency is also reduced; therefore, accurate identification of the thickness of the grate cooler material layer becomes a key problem for limiting the heat exchange efficiency of the grate cooler control and firing system.
At present, the thickness of a grate cooler material layer has three identification modes: the method is characterized by indirect quantities (pressure under a grate, secondary air temperature, hydraulic signals of a grate cooler and the like) related to the thickness of a material layer; secondly, extracting the material layer edge identification through traditional image processing; thirdly, direct measurement is carried out through a mechanical device; however, the above three methods all have certain problems, and the indirect characterization method comprises the following steps: the model of the indirect quantity and the material layer thickness is complex and presents a nonlinear characteristic, the relation between the indirect quantity and the material layer thickness is greatly influenced by working conditions, the material layer thickness measurement is often low in precision and difficult to fall to the ground for application; for the traditional image edge extraction method: the output result has large fluctuation and poor stability, and the influence of high-temperature atomization and the like cannot be overcome; direct measurement method for mechanical devices: the measuring device has high installation cost and needs to be maintained continuously, certain interference exists in the cooling and transportation of the burnt clinker by contact measurement, and the accuracy of the measuring result is low, so that a waiting solution is needed urgently.
Disclosure of Invention
Aiming at the current situation of the prior art, the invention aims to provide the grate cooler material layer thickness measuring method based on image segmentation, which greatly improves the identification precision and the model stability, effectively improves the accuracy of the model prediction result, and further can achieve the purposes of stabilizing the grate cooler material layer thickness, saving the coal consumption and enhancing the heat exchange efficiency of a firing system.
The technical scheme adopted by the invention for solving the technical problems is as follows: a grate cooler material layer thickness measuring method based on image segmentation is characterized by comprising the following steps:
s1, performing offline training of the image segmentation network model based on deep learning;
s2, carrying out image segmentation on the internal image of the grate cooler in real time based on the trained image segmentation model;
deploying the trained segmentation model to an image analysis platform on line, simultaneously acquiring image data inside the grate cooler by the image platform in real time, and performing segmentation processing on the image inside the grate cooler acquired in real time by the image segmentation model to obtain a segmentation result;
s3, carrying out post-processing conversion on the segmentation result to obtain the thickness of the grate cooler material layer;
and carrying out post-processing on the result after the real-time prediction by adopting the traditional image processing technology such as maximum connected region identification, edge extraction and the like to obtain the thickness of the grate cooler material layer.
Preferably, in the step S1, the image information inside the grate cooler shot by the high-temperature camera is collected, and after the annotation processing, a suitable image segmentation model is selected for training and evaluation, which specifically includes, but is not limited to, the following contents and step sequences:
s11: acquiring the internal image information of the grate cooler shot by a high-temperature camera offline;
s12: marking a material layer area of the image information in the grate cooler or a side wall area which is not covered by the material layer by using a marking tool;
s13: dividing the image data after the labeling into data sets according to a training set, a verification set and a test set;
s14: selecting a proper image segmentation neural network model, and setting initial parameters such as a solver, a learning rate, the number of rounds and the like;
s15: inputting the image information of the training set and the verification set into the image segmentation model, and obtaining a converged model parameter through iterative optimization;
s16: evaluating the optimized model by using an AP or mAP index, if the evaluation model meets the use requirement, considering the deployment to be online, and if the evaluation model does not meet the use requirement, entering the step S11 to renew the image data or entering the step S14 to adjust the parameters of the training model to retrain the model;
s17: and performing model compression and acceleration on the model after offline training by using technologies such as model pruning, model quantization and the like, and storing a model file.
Preferably, in the step S2, the trained image segmentation model is deployed to an image platform, and a real-time segmentation process is performed on the image acquired in real time, which includes but is not limited to the following steps and sequence:
s21: deploying the image analysis platform on a physical server or a cloud server;
s22: the physical server or the cloud server has a deep learning basic environment, such as a CUDA operation supporting function;
s23: the image analysis platform has the functions of reading video streams and extracting frames and the capability of periodically calling the trained model, wherein the video streams can be network video streams or off-line video streams;
s24: loading the model after offline training to the image platform, and starting an operation algorithm model;
s25: atomizing and flying sand of the video stream frame-extracted image data through a dark channel prior technology or other image processing technologies to enhance the image definition;
s26: inputting the preprocessed image into the started image segmentation model according to a certain period, executing a segmentation task, and outputting a segmentation result.
Preferably, in the step S3, the segmentation result after the real-time prediction is post-processed to obtain the thickness of the grate cooler material layer, which includes but is not limited to the following contents and sequence of steps:
s31: performing Gaussian filtering and binarization processing on the segmented image data;
s32: acquiring a maximum connected region in the binary image;
s33: performing edge detection on the maximum connected region, acquiring a maximum pixel height y value in an x axis with the same pixel width in an edge curve to form a one-dimensional vector, and performing statistical analysis on effective boundary data of a grate cooler material layer according to historical data to obtain the range of the x axis;
s34: performing abnormal value and filtering processing on the one-dimensional vector to obtain a preprocessed one-dimensional vector;
s35: calculating the average value or the median value of the preprocessed one-dimensional vector, and limiting the calculation result within a certain range to be used as a final material layer thickness value;
s36: and dividing the final bed thickness value by the pixel height of the original image to obtain the relative bed thickness.
Preferably, in the step S36, the relative thickness of the grate cooler material layer predicted in real time is introduced into the control loop to be used as a controlled variable, and by manipulating the speed of the grate cooler, the stable control of the thickness of the grate cooler material layer can be realized, the material thickness is stabilized within a set range, and the heat exchange efficiency is improved while the cooling quality of clinker is ensured.
Preferably, the image segmentation neural network of step S14 has an image enhancement function, and can effectively solve the problem that the conventional image processing such as high-temperature image atomization and sand flying cannot overcome.
Preferably, the image segmentation neural network of step S14 may adopt a training model with a small sample size, so as to avoid the complicated processes of downloading, preprocessing, and labeling of mass data.
Compared with the prior art, the invention has the following beneficial effects:
firstly, compared with the traditional image algorithm identification method, the grate cooler bed thickness identification method based on the deep learning image segmentation algorithm has the advantages that the identification precision is greatly improved, and particularly the anti-atomization and anti-flying sand interference capabilities of the model are remarkably enhanced;
secondly, the stability of the model is greatly improved, the long-time online running rate reaches more than 99%, and the stability of the model running is ensured from a data source and an output result respectively through the video stream offline automatic reconnection function, the identification result abnormal value processing, the smoothing and other post-processing methods;
thirdly, the accuracy of the model prediction result is improved, and after the prediction model deployed to the site stably runs for a period of time, the analysis and prediction of the change trend between the material layer thickness and the comprehensive pressure under the grate has a certain positive correlation, so that the process basic knowledge is met;
and finally, an accurate and stable grate cooler bed thickness identification result is taken as a controlled variable and is introduced into a grate cooler control loop, and the purposes of stabilizing the grate cooler bed thickness, saving coal consumption and enhancing the heat exchange efficiency of a firing system can be achieved by controlling the speed of the grate cooler and the frequency of a fan.
Drawings
FIG. 1 is a block diagram of the framework of the algorithm of the present invention, including its major constituent steps.
FIG. 2 is a flow chart of an implementation of the present invention, including flow steps in an implementation.
FIG. 3 is the image segmentation neural network structure principle of the algorithm of the present invention, and the present invention adopts the network structure development based on U-Net, but is not limited to the image segmentation network structure models such as ICNet and DeepLab v3 +.
FIG. 4 is an illustration of an embodiment of the invention for marking a grate cooler image shot by a high-temperature camera by using a marking tool.
FIG. 5 is a pseudo-color image obtained by image segmentation of a grate cooler image shot by a high-temperature camera according to the algorithm of the invention.
FIG. 6 shows a first application effect of the invention in the identification of the thickness of the grate cooler bed of a cement plant.
Fig. 7 shows a second application effect of the invention in the identification of the thickness of the grate cooler bed of a cement plant.
Detailed Description
Unless defined otherwise, technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention belongs. The use of "first," "second," and similar terms in the present application do not denote any order, quantity, or importance, but rather the terms are used to distinguish one element from another. The word "comprising" or "comprises", and the like, means that the element or item listed before the word covers the element or item listed after the word and its equivalents, but does not exclude other elements or items. The terms "connected" or "coupled" and the like are not restricted to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "upper", "lower", "left", "right", and the like are used merely to indicate relative positional relationships, and when the absolute position of the object being described is changed, the relative positional relationships may also be changed accordingly.
To maintain the following description of the embodiments of the present invention clear and concise, a detailed description of known functions and known components of the invention have been omitted.
As shown in fig. 1 to 7, a method for measuring the thickness of a grate cooler bed based on image segmentation comprises the following steps:
s1, performing offline training of the image segmentation network model based on deep learning;
s2, carrying out image segmentation on the internal image of the grate cooler in real time based on the trained image segmentation model;
deploying the trained segmentation model on an image analysis platform on line, simultaneously acquiring image data inside the grate cooler by the image platform in real time, and performing segmentation processing on the image inside the grate cooler acquired in real time by the image segmentation model to obtain a segmentation result;
s3, converting the segmentation result into the thickness of the grate cooler material layer through post-processing;
and (3) carrying out post-processing on the result after real-time prediction by adopting the traditional image processing technology such as maximum connected region identification, edge extraction and the like to obtain the thickness of the grate cooler material layer.
The following describes in detail a specific implementation of the method for measuring the thickness of the grate cooler bed based on image segmentation according to a specific embodiment.
The method comprises the following steps of firstly, training an image segmentation offline model based on deep learning, and specifically, realizing the following method:
(1) collecting the image information inside the grate cooler shot by a high-temperature camera, wherein the high-temperature camera is connected with a high-temperature protection lens through a CS standard threaded interface, and the high-temperature protection lens is inserted into the grate cooler through an opening on the side edge of the grate cooler; an RJ45 network port is reserved on the side edge of the high-temperature camera, and the high-temperature camera is connected with a production monitoring server through a super-six type network cable and is connected with a production video; by writing a python script, the import can read a video stream and extract an opencv-python packet of an image frame, then read a field video stream according to protocols such as rtmp or rtsp and the like, and finally extract frames at certain intervals to store the image.
(2) And (3) marking a material layer area of the image information in the grate cooler or a side wall area which is not covered by the material layer by using a labelme or hectogram intelligent marking tool and the like, wherein as shown in fig. 4, a schematic diagram of marking the side wall which is not covered by the material layer is shown, and the relative thickness of the material layer is the ratio of the pixel height of the whole image minus the pixel value of the height of the side wall to the pixel height of the whole image.
Specifically, a labelme labeling tool is used for sequentially labeling about 300 images in the image storage folder according to fig. 4, wherein the 300 images include the internal material layer image of the grate cooler with high definition and the internal material layer image of the grate cooler with serious atomization or flying sand; and storing the marked images in two folders according to the original images and json formats respectively.
(3) And dividing the marked about 300 image data into data sets according to the training set, the verification set and the test set, wherein the division ratio is 7:2:1 in sequence.
Specifically, a hundred-degree PaddleX tool is used for converting a format and dividing a data set, wherein the conversion format is used for converting the annotated json format file into a data format required by PaddleSeg, and if a Torch or TensorFlow frame is used for training a model, the annotated data format is converted into a specific format required by model training, and the data set is divided.
(4) Selecting a proper image segmentation neural network model, and setting initial parameters such as parameters of a solver, a learning rate, the number of rounds and the like;
specifically, the image segmentation neural network usually has structures such as U-Net, ICNet and DeepLab v3+, a U-Net structure training model is selected, an initial parameter batch _ size is set to be 4, iters are set to be 10000, a solver optimizer adopts a random gradient descent sgd, and a learning rate is set in a dynamic change mode from large to small according to a polynomial attenuation mode.
(5) Inputting image information of the training set and the verification set into an image segmentation model, and obtaining a converged model parameter through iterative optimization;
specifically, the model is trained according to set parameters, and the model in the training process is stored according to a certain number of iters periods in the training process; and simultaneously evaluating the stored training model, and if the training model evaluated this time is superior to the last training model, storing the model as the optimal model.
(6) And (3) evaluating the optimized model by using an AP (access point) or mAP (target area) index, considering the deployment on-line if the evaluation model meets the use requirement, and if the evaluation model does not meet the use requirement, entering the step (1) to renew the image data or entering the step (4) to adjust the parameters of the training model to retrain the model.
(7) Model compression and acceleration are carried out on the model after offline training by using the technologies of model cutting, model quantization and the like, and an offline model file is stored.
Specifically, the trained and evaluated model is further compressed according to the complexity of the model, so that the prediction speed of the model is increased, and the occupation of the model on the memory is reduced; the model is compressed and accelerated mainly from the aspects of model cutting, quantification, distillation and the like.
Secondly, deploying the trained segmentation model on an image analysis platform on line, acquiring image data inside the grate cooler in real time by the image platform, and segmenting the image inside the grate cooler acquired in real time by the image segmentation model to obtain a segmentation result, wherein the segmentation result is realized by the following specific method:
(1) deploying the image analysis platform on a physical server or a cloud server;
specifically, the image analysis platform is installed on a designated server, and relevant files including an IP address, a video frame rate, an image analysis high-width pixel value, a log file storage address and the like are configured correctly according to the requirements of configuration files, so that the image platform can read and predict video information.
(2) The physical server or the cloud server has a deep learning basic environment, such as a CUDA operation supporting function;
specifically, since the image segmentation algorithm is classified according to the pixel level, and the requirements on the video memory and the computing power are high, the server needs to have a CUDA operation function in order to better ensure the analysis performance.
(3) The image analysis platform has the functions of reading video stream and extracting frames and the capability of periodically calling the trained model, wherein the video stream can be a network video stream or an off-line video stream;
specifically, the image analysis platform comprises two parts, wherein one part is an image algorithm operation server and is responsible for real-time video stream frame extraction and image algorithm prediction and outputting a prediction result; the other part is a front section configuration and alarm display end which is responsible for camera information configuration, image loading algorithm, configuration analysis task and display alarm information.
(4) And loading the offline trained model to an image platform, and starting and running the algorithm model.
(5) The video stream frame extraction image data is subjected to atomization and sand flying through a dark channel prior technology or other image processing technologies, so that the image definition is enhanced;
(6) inputting the preprocessed image into a started image segmentation model, executing a segmentation task according to a certain period, and outputting a segmentation result;
specifically, loading a grate cooler bed thickness model which is trained offline to an image platform, selecting a grate cooler shooting camera, configuring a grate cooler bed thickness prediction model to the grate cooler camera to form an analysis task, setting parameters such as an ROI (region of interest), an operation condition, an operation period and an operation time for the task, and finally starting the task to execute a grate cooler bed thickness image segmentation task according to a certain period to form an image segmentation result.
Thirdly, carrying out post-processing on the segmentation result after real-time prediction by adopting the traditional image processing technology such as maximum connected region identification, edge extraction and other technical means to obtain the thickness of the grate cooler material layer, and specifically realizing the method by the following steps:
(1) performing Gaussian filtering and binarization processing on the segmented image data;
specifically, Gaussian filtering and binarization processing are carried out on the segmented pseudo-color image by adopting a correlation function in opencv, the Gaussian filtering is used for smoothing the segmentation result, and the binarization processing is convenient for subsequent edge detection; generally, an image after binarization processing contains a plurality of irregular connected regions with different sizes, and a wall region to be detected is the largest connected region, and other regions are interference regions.
(2) Acquiring a maximum connected region in a binary image;
specifically, all connected regions are first acquired using findContours in opencv, then the area of each connected region is calculated using a contourarray function, and finally the maximum connected region is acquired from the connected region area.
(3) Performing edge detection on the maximum connected region, acquiring a maximum pixel height y value in an x axis with the same pixel width in an edge curve to form a one-dimensional vector, and statistically analyzing effective boundary data of a grate cooler material layer according to historical data within the range of the x axis to obtain the effective boundary data;
specifically, X-axis and y-axis coordinate data (X-axis and Y-axis coordinate data) of the outer contour of the maximum connected region are extracted1,Y1). In addition, an X-axis data range X in the visible range of the side wall when the grate cooler operates stably is obtained according to statistical analysis0(ii) a At X0Scanning the Y axis in the range, and taking the Y value corresponding to the same horizontal axis to form a one-dimensional vector Y2。
(4) For one-dimensional vector Y2Carrying out abnormal value and filtering processing to obtain a preprocessed one-dimensional vector Y3；
Specifically, the pair Y is processed using the 3sigma method or other outlier processing method2Processing, and performing first-order filtering or other smoothing means to obtain a preprocessed one-dimensional vector Y3。
(5) For the preprocessed one-dimensional vector Y3Calculating a mean value or a median value, and limiting a calculation result within a certain range to be used as a final material layer thickness value;
(6) and dividing the final bed thickness value by the pixel height of the original image to obtain the relative bed thickness.
Compared with the traditional image processing and measuring method, the method provided by the invention can better overcome the influence of the red river, flying sand, image atomization and the like, and the measuring precision is greatly improved; compared with a variable indirect representation measurement method, the method provided by the invention can adapt to different working condition changes, can overcome the situation that an indirect representation variable can only adapt to a linear relation, and enhances the measurement stability; compared with a direct measuring device, the method provided by the invention is a non-contact measuring method, can avoid interference on a clinker cooling process, has small equipment maintenance workload, and improves the measuring accuracy and stability. In addition, the measurement result is introduced into the control loop, so that the aim of stabilizing the thickness of the grate cooler bed can be fulfilled, the clinker yield is improved, and the heat exchange efficiency of the grate cooler is improved.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those skilled in the art that various changes in the embodiments and modifications thereof may be made, and equivalents may be substituted for elements thereof; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (7)
1. A grate cooler material layer thickness measuring method based on image segmentation is characterized by comprising the following steps:
s1, performing offline training of the image segmentation network model based on deep learning;
s2, carrying out image segmentation on the internal image of the grate cooler in real time based on the trained image segmentation model;
deploying the trained segmentation model to an image analysis platform on line, simultaneously acquiring image data inside the grate cooler by the image platform in real time, and performing segmentation processing on the image inside the grate cooler acquired in real time by the image segmentation model to obtain a segmentation result;
s3, carrying out post-processing conversion on the segmentation result to obtain the thickness of the grate cooler material layer;
and carrying out post-processing on the result after the real-time prediction by adopting the traditional image processing technology such as maximum connected region identification, edge extraction and the like to obtain the thickness of the grate cooler material layer.
2. The method for measuring the thickness of the grate cooler material layer based on the image segmentation according to claim 1, wherein in the step S1, the image information inside the grate cooler shot by the high-temperature camera is collected, and after the annotation processing, a proper image segmentation model is selected for training and evaluation, specifically including but not limited to the following contents and steps:
s11: acquiring the internal image information of the grate cooler shot by a high-temperature camera offline;
s12: marking a material layer area of the image information in the grate cooler or a side wall area which is not covered by the material layer by using a marking tool;
s13: dividing the image data after the labeling into data sets according to a training set, a verification set and a test set;
s14: selecting a proper image segmentation neural network model, and setting initial parameters such as a solver, a learning rate, the number of rounds and the like;
s15: inputting the image information of the training set and the verification set into the image segmentation model, and obtaining a converged model parameter through iterative optimization;
s16: evaluating the optimized model by using an AP or mAP index, if the evaluation model meets the use requirement, considering the deployment to be online, and if the evaluation model does not meet the use requirement, entering the step S11 to renew the image data or entering the step S14 to adjust the parameters of the training model to retrain the model;
s17: and performing model compression and acceleration on the model after offline training by using technologies such as model pruning, model quantization and the like, and storing a model file.
3. The method for measuring the thickness of the grate cooler bed based on the image segmentation as claimed in claim 2, wherein in the step S2, the trained image segmentation model is deployed to an image platform, and the real-time segmentation process is performed on the image acquired in real time, which includes but is not limited to the following steps and sequence:
s21: deploying the image analysis platform on a physical server or a cloud server;
s22: the physical server or the cloud server has a deep learning basic environment, such as a CUDA operation supporting function;
s23: the image analysis platform has the functions of reading video streams and extracting frames and the capability of periodically calling the trained model, wherein the video streams can be network video streams or off-line video streams;
s24: loading the model after offline training to the image platform, and starting an operation algorithm model;
s25: atomizing and flying sand of the video stream frame-extracted image data through a dark channel prior technology or other image processing technologies to enhance the image definition;
s26: inputting the preprocessed image into the started image segmentation model according to a certain period, executing a segmentation task, and outputting a segmentation result.
4. The method for measuring the thickness of the grate cooler material layer based on the image segmentation as claimed in claim 3, wherein the step S3 is to perform post-processing on the segmentation result after the real-time prediction to obtain the thickness of the grate cooler material layer, and the method includes but is not limited to the following contents and steps:
s31: performing Gaussian filtering and binarization processing on the segmented image data;
s32: acquiring a maximum connected region in the binary image;
s33: performing edge detection on the maximum connected region, acquiring a maximum pixel height y value in an x axis with the same pixel width in an edge curve to form a one-dimensional vector, and performing statistical analysis on effective boundary data of a grate cooler material layer according to historical data to obtain the range of the x axis;
s34: performing abnormal value and filtering processing on the one-dimensional vector to obtain a preprocessed one-dimensional vector;
s35: calculating the average value or the median value of the preprocessed one-dimensional vector, and limiting the calculation result within a certain range to be used as a final material layer thickness value;
s36: and dividing the final bed thickness value by the pixel height of the original image to obtain the relative bed thickness.
5. The method for measuring the thickness of the grate cooler material layer based on the image segmentation as claimed in claim 4, wherein in the step S36, the relative thickness of the grate cooler material layer predicted in real time is introduced into a control loop and used as a controlled variable, the stable control of the thickness of the grate cooler material layer can be realized by controlling the speed of the grate cooler, the thickness of the material layer is stabilized within a set range, and the heat exchange efficiency is improved while the cooling quality of clinker is ensured.
6. The method for measuring the thickness of the grate cooler material layer based on the image segmentation as claimed in claim 1, wherein the image segmentation neural network of the step S14 has an image enhancement function, and can effectively solve the problem that the traditional image processing such as high-temperature image atomization and sand flying cannot overcome.
7. The method for measuring the thickness of the grate cooler bed based on the image segmentation as claimed in claim 6, wherein the image segmentation neural network of the step S14 can adopt a training model with a small sample size, thereby avoiding the tedious processes of downloading, preprocessing and labeling mass data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111534823.9A CN114187282A (en) | 2021-12-15 | 2021-12-15 | Grate cooler bed thickness measuring method based on image segmentation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111534823.9A CN114187282A (en) | 2021-12-15 | 2021-12-15 | Grate cooler bed thickness measuring method based on image segmentation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114187282A true CN114187282A (en) | 2022-03-15 |
Family
ID=80543954
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111534823.9A Pending CN114187282A (en) | 2021-12-15 | 2021-12-15 | Grate cooler bed thickness measuring method based on image segmentation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114187282A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115446420A (en) * | 2022-10-25 | 2022-12-09 | 天地上海采掘装备科技有限公司 | Automatic flame cutting method for tooth holder welding notch |
| CN117934468A (en) * | 2024-03-22 | 2024-04-26 | 泰安中联水泥有限公司 | Image processing-based grate cooler fault prediction method and system |
-
2021
- 2021-12-15 CN CN202111534823.9A patent/CN114187282A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115446420A (en) * | 2022-10-25 | 2022-12-09 | 天地上海采掘装备科技有限公司 | Automatic flame cutting method for tooth holder welding notch |
| CN117934468A (en) * | 2024-03-22 | 2024-04-26 | 泰安中联水泥有限公司 | Image processing-based grate cooler fault prediction method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105956618B (en) | Converter steelmaking blowing state identification system and method based on image dynamic and static characteristics | |
| CN114187282A (en) | Grate cooler bed thickness measuring method based on image segmentation | |
| CN113139521B (en) | Pedestrian boundary crossing monitoring method for electric power monitoring | |
| CN112819802B (en) | Method for supervising and predicting blast furnace condition abnormality based on tuyere information deep learning | |
| CN110264457B (en) | Welding seam autonomous identification method based on rotating area candidate network | |
| US11900676B2 (en) | Method and apparatus for detecting target in video, computing device, and storage medium | |
| CN102706274B (en) | System for accurately positioning mechanical part by machine vision in industrially-structured scene | |
| CN112613454A (en) | Electric power infrastructure construction site violation identification method and system | |
| CN110057820B (en) | Method, system and storage medium for on-line detection of chlorine-hydrogen ratio of hydrogen chloride synthesis furnace | |
| CN113344967B (en) | Dynamic target identification tracking method under complex background | |
| CN110555377A (en) | pedestrian detection and tracking method based on fisheye camera overlook shooting | |
| CN112989910A (en) | Power target detection method and device, computer equipment and storage medium | |
| CN114511519A (en) | Train bottom bolt loss detection method based on image processing | |
| CN113643206A (en) | Cow breathing condition detection method | |
| CN111626104B (en) | Cable hidden trouble point detection method and device based on unmanned aerial vehicle infrared thermal image | |
| CN106778675B (en) | A kind of recognition methods of target in video image object and device | |
| US11068734B2 (en) | Client terminal for performing hybrid machine vision and method thereof | |
| CN116934674A (en) | Large-granularity silicon material detection and removal system based on YOLOv8 network | |
| CN111339824A (en) | Road surface sprinkled object detection method based on machine vision | |
| CN108734158B (en) | Real-time train number identification method and device | |
| CN115761580A (en) | Detection method of water surface floater, edge calculation terminal and storage medium | |
| CN105095897A (en) | Gradient image and similarity weight based digit identification method | |
| CN108800916A (en) | A kind of grate-cooler thickness of feed layer Model Predictive Control Algorithm based on machine vision | |
| CN111652055B (en) | Intelligent switch instrument identification method based on two-stage positioning | |
| CN114638805A (en) | Track slab crack detection method and system 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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20230801 Address after: No. 320 and 436 Pingtang East Road, Puxing Street, Xinjin District, Chengdu City, Sichuan Province, 611400 Applicant after: Sinoma Intelligent Technology (Chengdu) Co.,Ltd. Address before: 303 Hanzhongmen street, Gulou District, Nanjing City, Jiangsu Province Applicant before: NANJING KISEN INTERNATIONAL ENGINEERING Co.,Ltd. |
|
| TA01 | Transfer of patent application right |