CN111027392A - Semi-supervised extraction method for boiler burner flame image quantitative features - Google Patents
Semi-supervised extraction method for boiler burner flame image quantitative features Download PDFInfo
- Publication number
- CN111027392A CN111027392A CN201911100135.4A CN201911100135A CN111027392A CN 111027392 A CN111027392 A CN 111027392A CN 201911100135 A CN201911100135 A CN 201911100135A CN 111027392 A CN111027392 A CN 111027392A
- Authority
- CN
- China
- Prior art keywords
- image
- feature
- model
- flame
- combustion
- 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
- 238000000605 extraction Methods 0.000 title claims abstract description 21
- 238000002485 combustion reaction Methods 0.000 claims abstract description 38
- 238000000034 method Methods 0.000 claims abstract description 24
- 230000008569 process Effects 0.000 claims abstract description 21
- 239000013598 vector Substances 0.000 claims abstract description 15
- 230000008859 change Effects 0.000 claims abstract description 5
- 239000000446 fuel Substances 0.000 claims abstract description 4
- 238000011002 quantification Methods 0.000 claims abstract 3
- 239000003245 coal Substances 0.000 claims description 28
- 238000012549 training Methods 0.000 claims description 11
- 238000013507 mapping Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000013139 quantization Methods 0.000 claims 2
- 230000006835 compression Effects 0.000 claims 1
- 238000007906 compression Methods 0.000 claims 1
- 238000011176 pooling Methods 0.000 description 13
- 238000010586 diagram Methods 0.000 description 7
- 239000002817 coal dust Substances 0.000 description 4
- 238000012545 processing Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000017525 heat dissipation Effects 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000000087 stabilizing effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000010977 unit operation Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Geometry (AREA)
- Control Of Combustion (AREA)
Abstract
本发明公开了一种锅炉燃烧器火焰图像量化特征的半监督提取方法，其所述方法包括以下步骤：步骤1：收集燃料燃烧火焰的视频，然后将视频转化为图像集，并调整图片大小，使其为长宽相等的正方形火焰图像；步骤2：利用步骤1中的图像集训练卷积自编码器模型；步骤3：对卷积自编码模型中编码网络的输出进行可视化，得到一组输入图像的特征图；步骤4：从步骤3中选取最能体现燃烧变化的特征图像，并将该特征图像展开为一个一维向量，作为待选特征向量；步骤5：提取与燃烧状态相关的特征量，计算直接影响燃烧的过程参数与提取的特征量之间的相关系数，选取相关系数大于设定阈值的特征量，然后对步骤4的待选特征向量进行滤波以滤除噪声。
The invention discloses a semi-supervised extraction method for the quantification feature of a boiler burner flame image. The method includes the following steps: Step 1: collect a video of fuel combustion flame, then convert the video into an image set, and adjust the size of the image, Make it a square flame image with equal length and width; Step 2: Use the image set in Step 1 to train the convolutional autoencoder model; Step 3: Visualize the output of the encoding network in the convolutional autoencoder model to get a set of inputs The feature map of the image; Step 4: Select the feature image that can best reflect the combustion change from Step 3, and expand the feature image into a one-dimensional vector as a feature vector to be selected; Step 5: Extract features related to the combustion state Calculate the correlation coefficient between the process parameters that directly affect the combustion and the extracted feature, select the feature whose correlation coefficient is greater than the set threshold, and then filter the feature vector to be selected in step 4 to filter out noise.
Description
Technical Field
The invention relates to the technical field of process monitoring of image processing, in particular to a semi-supervised extraction method of flame image quantitative characteristics of a boiler burner.
Background
Because the fuel characteristics of the coal-fired boiler are changeable, the unit output is frequently changed, the phenomenon of unstable combustion is easy to occur, and the safety of the whole unit operation is directly influenced. Monitoring the stability of combustion in a furnace and optimizing the combustion for adjustments has been a focus of industry attention.
The flame visualization and characterization technology is one of important tools for deeply knowing the combustion of the pulverized coal in the hearth, and aims to provide safety guarantee for combustion adjustment. Because the flame images contain a great deal of information, it is often necessary to integrate a number of different characteristic variables to reflect the combustion process. The traditional flame image feature extraction technology generally adopts a fixed extraction framework, such as denoising, enhancement, segmentation, feature extraction and the like. However, the conventional feature extraction procedure is complicated, and the validity of the extracted feature quantity often depends on the rationality of the image segmentation result.
Therefore, the semi-supervised extraction method for the boiler burner flame image quantitative features is provided for solving the problems in the prior art.
Disclosure of Invention
The invention discloses a semi-supervised extraction method of quantified characteristics of a flame image of a boiler burner, which comprises the following steps of:
step 1: collecting a video of fuel combustion flame, converting the video into an image set, and adjusting the size of an image to make the image be a square flame image with the same length and width;
step 2: training a convolutional self-encoder model by using the image set in the step 1;
and step 3: visualizing the output of a coding network in a convolutional self-coder model to obtain a group of characteristic graphs of input images;
and 4, step 4: selecting the characteristic image which can reflect combustion change most from the step 3, and unfolding the characteristic image into a one-dimensional vector as a candidate characteristic vector;
and 5: and (4) extracting the characteristic quantity related to the combustion state, calculating a correlation coefficient between the process parameter directly influencing combustion and the extracted characteristic quantity, selecting the characteristic quantity of which the correlation coefficient is greater than a set threshold value, and filtering the to-be-selected characteristic vector in the step (4) to filter noise.
Preferably, when the video is converted into the image set in step 1, the image set is established by performing image stretching and compressing preprocessing on the image according to needs and combining the image recording time.
Preferably, the training step of step 2 comprises:
step 2.1: randomly selecting a part of flame images from the data set as the input of the convolution self-encoder model, and outputting the model as a reconstructed image with the size completely consistent with that of the input image;
step 2.2: given an input image of the encoder model as X and an output image as Y, the model is composed of a decoding process g (-) and an encoding process f (-) and is mathematically described as:
f:X→F
g:F→X
Y＝g(f(X))
step 2.3: randomly selecting a part of flame images from the image set as the input of the convolution self-encoder model, and setting a loss function by a model training target that a model output image is consistent with an input image:
wherein, W and b are weight and bias parameters related to a convolution process in the f mapping and g mapping processes; n is the number of picture samples in each training process; xiIs the ith input sample of the coding model; y isiThe ith output of the corresponding decoding model;
step 2.4: and training the convolutional self-encoder model by using cross loss entropy as a loss function until the loss function value is reduced to an expected range or the cycle number reaches an expected value.
Preferably, the input image feature map of step 3 is a filtered image after a convolution operation.
Preferably, the process parameters of step 5 directly influencing combustion include: coal feeding amount, primary air quantity and coal dust concentration.
The invention provides a semi-supervised extraction method of quantified characteristics of a flame image of a boiler burner, which simplifies the steps of extracting the characteristics of the flame image, reduces the workload of manually determining the characteristic quantity, reduces the information loss caused by subjective components, and can effectively extract deeper information contained in the flame image.
Drawings
FIG. 1 is a flow chart of a semi-supervised extraction method of quantified features of a boiler burner flame image.
Fig. 2 is a schematic diagram of a three-layer self-encoder structure.
FIG. 3 is a schematic structural diagram of a convolution self-encoder in a flame image feature extraction process.
Fig. 4 is a diagram comparing an encoder output picture with an input picture.
FIG. 5 is a graph showing the comparison between the characteristic parameters of the output of the encoder after being filtered by an exponential sliding window and the coal feeding amount.
FIG. 6 is a graph showing the correlation coefficient between the output characteristic parameters of the encoder and the coal feeding rate.
FIG. 7 is a graph comparing the output characteristic parameter of the encoder with the primary wind temperature after being filtered by an exponential sliding window.
Fig. 8 is a graph comparing correlation coefficients between each characteristic parameter output by the encoder and the primary air temperature.
Detailed Description
In order to make the implementation objects, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be described in more detail below with reference to the accompanying drawings in the embodiments of the present invention. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are only some, but not all embodiments of the invention. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In this example, images of flames burning for 4 hours were taken for analysis. Simultaneously selecting variables related to the combustion process of the boiler: coal feeding amount and unit load. Wherein y in FIG. 5 shows the variation of the coal feeding amount in the [0s,14400s ] interval. As can be seen from the figure, the corresponding coal mill is cut off at 5700s, the coal feeding rate begins to gradually decrease, and the coal feeding rate of the coal mill is reduced to 0t/h at 6300 s; the F-mill was restarted at 7320s until the stabilizing force was restored. According to the combustion principle, the concentration of the pulverized coal is the most main influence factor of the ignition and combustion characteristics of the pulverized coal airflow, and the combustion stability of the boiler is directly influenced. At low coal dust concentrations, the coal dust air mixture requires less heat to ignite and less heat to supply, thus quickly heating the coal dust air stream to the critical state of ignition. Because the concentration of the pulverized coal is low, the discharged heat is less, the combustion is not strong, continuous flame cannot be formed, the outward heat dissipation is larger, and the temperature level is lower, so that the combustion is unstable. Therefore, when the coal feed rate is close to 0t/h, the combustion state is unstable. And selecting the coal feeding amount as a related variable of the boiler combustion state change, taking the coal feeding amount as an index for evaluating the combustion state, comparing the index with the characteristic amount extracted from the encoding model later, and verifying the effectiveness of the characteristic extraction method.
The following describes a combustion process flame image feature extraction method based on a convolution self-encoder model, with reference to the accompanying drawings and embodiments.
The flame image feature extraction method based on the combination of the convolutional neural network model and the self-encoder model shown in FIG. 2 as shown in FIG. 1 comprises the following implementation steps:
converting a flame video into a square image sequence with equal length and width, wherein the flame image in the embodiment is a color image, the number of channels is 3, and the length and the width are respectively set as 128;
training a convolutional auto-encoder model, as shown in fig. 3, where the encoding part: inputting a flame image with the size of 128 × 128 × 3, convolving the flame image by using 64 convolution kernels with the size of 3 × 3, then performing pooling (in the embodiment, the pooling is performed in a non-overlapping maximum pooling with the step size of 2), and outputting a feature map with the size of 64 × 64 × 64; performing convolution by using 32 convolution kernels with the size of 3 multiplied by 3, performing pooling processing, and outputting a feature map with the size of 32 multiplied by 32; and finally, performing convolution by using 16 convolution kernels with the size of 3 multiplied by 3, performing pooling processing, and outputting a feature map with the size of 16 multiplied by 16. The decoding process is symmetrical to the decoding network, firstly, the 16 multiplied by 16 by 3 deconvolution kernels are adopted to deconvolute the feature map of the size of 16 multiplied by 16 which is output by encoding, and the feature map of the size of 16 multiplied by 16 is output; then, performing inverse pooling (inverse pooling is inverse pooling, which can only approximate reduction of pooling process, in this embodiment, the employed inverse pooling mode is non-overlapping maximum inverse pooling with step length of 2), and performing deconvolution by using 32 deconvolution kernels with size of 3 × 3, and outputting feature maps with size of 32 × 32 × 32; then, performing inverse pooling operation, performing deconvolution by adopting 64 deconvolution kernels with the size of 3 × 3, and outputting a feature map with the size of 64 × 64 × 64; finally, performing inverse pooling operation, performing inverse convolution by using 3 inverse convolution kernels with the size of 3 multiplied by 3, and outputting a feature map with the size of 128 multiplied by 3, namely outputting Y from the encoding network.
Further, encoding and decoding the flame image data set, requiring that the output Y of the model can restore the input X as much as possible, and selecting the cross entropy as a measure of the error between the input X and the output Y in this embodiment specifically includes:
and setting the cycle number of the training process to be 5000 times, converging the loss function to an acceptable range or an expected value, and finishing the model training at the moment.
The encoded network output is visualized. The flame images are input into a coding network and output into 16 characteristic maps with the size of 16 multiplied by 16. FIG. 4 is an original flame image and corresponding coded network output characteristic diagram under three different coal feeding conditions, corresponding to three states of stable, unstable and fire extinguishing respectively. By observing and analyzing the output characteristic diagrams, the shaded part in the 13 th characteristic diagram represents the area of the pulverized coal, and the characteristic quantity can be independently extracted as the characteristic quantity reflecting the combustion process.
Further, the brightness mean value of the 13 th feature map is extracted for further analysis.
Expanding the output of the coding network into a one-dimensional vector; the coding network outputs a feature map of size 16 × 16 × 16, and the number of feature quantities after expansion into one-dimensional vectors is 4096.
The characteristic quantity that reflects the state of the combustion process is extracted. In this embodiment, the coal supply amount and the primary air temperature are selected as process parameters that directly affect the combustion state, correlation coefficients between each eigenvector output by the coding network and the coal supply amount (primary air temperature) are respectively calculated, and the eigenvector with a higher absolute value of the correlation coefficient with the process parameters is selected as the eigenvalue reflecting the combustion process. FIG. 5 shows the extracted flame image feature variables after being filtered by an exponential sliding window, where x1 is the feature variable with the highest positive correlation number with the coal feeding rate in the encoder output feature vector, and x2 is the feature variable with the highest negative correlation coefficient with the coal feeding rate in the encoder output feature vector; x3 is the average value of the brightness of the 13 th characteristic diagram output by the encoder, and y is the variation of the coal feeding rate in the combustion process. It can be seen from the figure that the extracted characteristic quantity can effectively reflect the change of the combustion process to a certain extent, wherein x2 is in inverse proportion to the coal feeding amount along with the advancing of the combustion process, and x1 and x3 are in direct proportion to the coal feeding amount, so that the combustion state can be accurately reflected. FIG. 6 shows the correlation coefficient between each characteristic parameter and the coal feeding rate output by the encoder, wherein the maximum value of the positive correlation coefficient is 0.909998, and the maximum value of the negative correlation coefficient is-0.922144. Similarly, fig. 7 shows the extracted flame image characteristic parameters after being filtered by an exponential sliding window, where x1 is a characteristic variable with the highest positive correlation coefficient with the primary air temperature in the encoder output characteristic vector, x2 is a characteristic variable with the highest negative correlation coefficient with the primary air temperature in the encoder output characteristic vector, and y is the primary air temperature variation condition in the combustion process. FIG. 8 shows the correlation coefficient between each characteristic parameter and the primary air temperature output by the encoder, wherein the maximum value of the positive correlation coefficient is 0.903492, and the maximum value of the negative correlation coefficient is-0.938299.
Finally, it should be pointed out that: the above examples are only for illustrating the technical solutions of the present invention, and are not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (5)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911100135.4A CN111027392A (en) | 2019-11-12 | 2019-11-12 | Semi-supervised extraction method for boiler burner flame image quantitative features |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911100135.4A CN111027392A (en) | 2019-11-12 | 2019-11-12 | Semi-supervised extraction method for boiler burner flame image quantitative features |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111027392A true CN111027392A (en) | 2020-04-17 |
Family
ID=70201233
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911100135.4A Pending CN111027392A (en) | 2019-11-12 | 2019-11-12 | Semi-supervised extraction method for boiler burner flame image quantitative features |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111027392A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115977496A (en) * | 2023-02-24 | 2023-04-18 | 重庆长安汽车股份有限公司 | Vehicle door control method, system, equipment and medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108664953A (en) * | 2018-05-23 | 2018-10-16 | 清华大学 | A kind of image characteristic extracting method based on convolution self-encoding encoder model |
-
2019
- 2019-11-12 CN CN201911100135.4A patent/CN111027392A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108664953A (en) * | 2018-05-23 | 2018-10-16 | 清华大学 | A kind of image characteristic extracting method based on convolution self-encoding encoder model |
Non-Patent Citations (1)
| Title |
|---|
| TIAN QIU ET AL.: "An Unsupervised Classification Method for Flame Image of Pulverized Coal Combustion Based on Convolutional Auto-Encoder and Hidden Markov Model", 《ENERGIES》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115977496A (en) * | 2023-02-24 | 2023-04-18 | 重庆长安汽车股份有限公司 | Vehicle door control method, system, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109308696B (en) | No-reference image quality evaluation method based on hierarchical feature fusion network | |
| WO2020155929A1 (en) | Method for determining rock mass integrity | |
| CN106846305B (en) | A kind of boiler combustion stability monitoring method based on the more characteristics of image of flame | |
| CN108896499A (en) | In conjunction with principal component analysis and the polynomial spectral reflectance recovery method of regularization | |
| CN110378848B (en) | An Image Dehazing Method Based on Derivative Graph Fusion Strategy | |
| CN110728728B (en) | A Method of Image Reconstruction Based on Non-local Regularization in Compressed Sensing Network | |
| CN111027392A (en) | Semi-supervised extraction method for boiler burner flame image quantitative features | |
| CN110717495A (en) | Identification method of solid waste incineration conditions based on multi-scale color moment feature and random forest | |
| CN117484031A (en) | Photovoltaic module welding processing equipment | |
| CN112634171B (en) | Image defogging method and storage medium based on Bayesian convolutional neural network | |
| CN106303524B (en) | Video Double Compression Detection Method Based on Predictive Residual Variation Mode | |
| CN107147909B (en) | Variance-based recompression JPEG image original quantization step length estimation method | |
| CN112712483A (en) | High-reflection removing method based on light field double-color reflection model and total variation | |
| CN104951800A (en) | Resource exploitation-type area-oriented remote sensing image fusion method | |
| CN113989162B (en) | Method for defogging flame of factory based on neural network | |
| CN112560672A (en) | Fire image recognition method based on SVM parameter optimization | |
| CN111695507B (en) | Static gesture recognition method based on improved VGGNet network and PCA | |
| CN116778301A (en) | A method and system for quantitative detection of furnace flame combustion state | |
| CN111091580A (en) | A Standing Tree Image Segmentation Method Based on Improved ResNet-UNet Network | |
| CN109829377A (en) | A kind of pedestrian's recognition methods again based on depth cosine metric learning | |
| CN115575404A (en) | Fruit appearance quality detection method based on ratio multispectral image | |
| CN115578619A (en) | Combustion stability diagnosis method based on multi-source information fusion | |
| CN108154501B (en) | Adaptive evaluation method for image segmentation quality of spiral blade based on gray distribution | |
| CN113222244A (en) | Online boiler combustion optimization method based on flame combustion image judgment | |
| CN112364904A (en) | Model pruning method based on random sampling |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200417 |
|
| RJ01 | Rejection of invention patent application after publication |