CN117455828A - Combustion quality detection method based on MLP neural network - Google Patents
Combustion quality detection method based on MLP neural network Download PDFInfo
- Publication number
- CN117455828A CN117455828A CN202310300077.XA CN202310300077A CN117455828A CN 117455828 A CN117455828 A CN 117455828A CN 202310300077 A CN202310300077 A CN 202310300077A CN 117455828 A CN117455828 A CN 117455828A
- Authority
- CN
- China
- Prior art keywords
- flame
- image
- neural network
- model
- training
- 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
- 238000002485 combustion reaction Methods 0.000 title claims abstract description 34
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 29
- 238000001514 detection method Methods 0.000 title claims abstract description 23
- 238000012549 training Methods 0.000 claims abstract description 44
- 238000000034 method Methods 0.000 claims abstract description 29
- 238000001914 filtration Methods 0.000 claims abstract description 24
- 238000007781 pre-processing Methods 0.000 claims abstract description 23
- 238000003062 neural network model Methods 0.000 claims abstract description 19
- 230000000877 morphologic effect Effects 0.000 claims abstract description 18
- 238000012545 processing Methods 0.000 claims abstract description 18
- 238000012360 testing method Methods 0.000 claims description 14
- 238000005516 engineering process Methods 0.000 claims description 10
- 230000011218 segmentation Effects 0.000 claims description 10
- 238000001931 thermography Methods 0.000 claims description 8
- 238000003708 edge detection Methods 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 5
- 239000003086 colorant Substances 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 4
- 238000002360 preparation method Methods 0.000 claims description 4
- 230000000007 visual effect Effects 0.000 claims description 3
- 230000003044 adaptive effect Effects 0.000 claims description 2
- 241001270131 Agaricus moelleri Species 0.000 claims 2
- 239000000446 fuel Substances 0.000 claims 1
- 238000004364 calculation method Methods 0.000 abstract description 3
- 238000004519 manufacturing process Methods 0.000 abstract 1
- 238000005259 measurement Methods 0.000 abstract 1
- 210000002569 neuron Anatomy 0.000 description 12
- 238000011156 evaluation Methods 0.000 description 11
- 238000004422 calculation algorithm Methods 0.000 description 9
- 238000012795 verification Methods 0.000 description 8
- 238000000605 extraction Methods 0.000 description 5
- 238000003709 image segmentation Methods 0.000 description 4
- 230000001788 irregular Effects 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000002123 temporal effect Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 230000003416 augmentation Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000012821 model calculation Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000001681 protective effect Effects 0.000 description 2
- 238000007789 sealing Methods 0.000 description 2
- 241000872198 Serjania polyphylla Species 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 239000003245 coal Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000010349 pulsation Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
- G06T2207/20032—Median filtering
-
- 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
- G06T2207/20036—Morphological image processing
-
- 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
- G06T2207/30168—Image quality inspection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a combustion quality detection method based on an MLP neural network, which comprises the following steps: s1, acquiring corresponding flame combustion videos by an infrared camera; s2, segmenting a flame burning image from a flame burning video through a video segmenter; s3, preprocessing, and adopting median filtering to reduce noise of the flame image; s4, graying the flame image: converting the flame image into a gray scale image for subsequent processing; s5, binarizing the flame image: converting the gray level image into a binary image so as to better extract flame characteristics; s6, morphological processing of the flame image: morphological processing is carried out on the binary image so as to eliminate noise and extract flame characteristics; s7, extracting features of the flame image: extracting characteristics of the flame image so as to facilitate subsequent neural network model training; s8, the neural network model adopts a multi-layer perceptron (MLP) architecture, wherein the MLP architecture comprises an input layer, a hidden layer and an output layer. The input layer is used for receiving the characteristics of the flame image, the hidden layer is used for extracting the characteristics of the flame image, and the output layer is used for outputting the detection result of the flame quality. Compared with the traditional manual measurement, the method has the characteristics of accurate and rapid numerical value calculation and simple operation, thereby avoiding human data errors caused by manual calculation and improving the production efficiency.
Description
Technical Field
The invention relates to the field of computer image neural networks, in particular to a combustion quality detection method based on a neural network.
Background
In the field of computer image detection, the deep neural network can distinguish images of different categories through good training and shows excellent performance. However, one of the drawbacks of the current image detection technology is poor self-adaptation, and once the target image is polluted by strong noise or has large defects, the target image often cannot obtain ideal results, and in general, the deep neural network only performs better training on a certain identification field planned in advance, that is, most of the neural network models are designed for a closed known system. In real life, known category images do not effectively cover all categories of flame quality produced by different combustibles.
On one hand, the technology has poor self-adaptability, and once a flame image is polluted by stronger noise or has larger defects, an ideal result cannot be obtained; on the other hand, the technology is used for training the neural network, a large amount of data is required to be collected, a plurality of different neural networks are required to be trained according to different combustible materials due to the closed environment of the neural network, and the data processing process is complex.
Purpose(s)
The invention provides a combustion quality detection method based on a neural network. The method can effectively improve the accuracy of flame combustion quality detection, effectively detect the current flame quality by training a better neural network and introducing parameters such as average flame gray level, average gray level of an effective area, flame abundance, area rate of a high-temperature area, flame height, flame width and the like into the neural network, and detect the flame combustion quality of the boiler according to the analysis result of the neural network. The invention has the advantages that: by adopting the neural network technology, the combustion quality of the boiler flame can be accurately detected, so that the combustion efficiency and safety of the boiler flame are improved.
Technical proposal
In order to solve the above problems, the present application proposes a combustion quality detection method based on a neural network, including the steps of:
s1, building a machine vision image acquisition system, and acquiring a flame burning video;
s2, performing identification region segmentation on the collected flame burning video through a video segmenter, and identifying a flame image with stable burning;
s3, preprocessing, and adopting median filtering to reduce noise of the flame image;
s4, preprocessing, namely converting the flame image into a gray image by adopting graying;
s5, preprocessing, namely converting the gray level image into a binary image by binarization;
s6, preprocessing, namely performing morphological processing on the binary image to eliminate noise and extract flame characteristics;
s7, extracting features of the flame image by extracting features of the flame image so as to facilitate subsequent neural network model training;
s8, training the neural network by adopting a multi-layer perceptron (MLP) architecture by the neural network model;
s9, testing and verifying. How the camera collects the flame in the boiler.
Preferentially, the infrared camera is used for controlling the camera to collect corresponding flame burning videos and the video divider is used for dividing the flame video areas, specifically:
the infrared thermal imaging camera adopted by the patent is imaging equipment which integrates a camera, a protective cover, an infrared lamp, a power supply radiating unit and the like into a whole. For reasons of the sealing of the boiler, the camera can collect the flame in the boiler through a thermal imaging technology. The thermal imaging technology can convert heat into visible light, so that a camera can capture an image of flame, therefore, flame data is obtained by controlling the camera to collect corresponding flame burning videos through an infrared thermal camera, the flame video areas are segmented through a video divider, then the data collected by the camera are processed, and in the flame image collection process, shooting parameters of the camera, including exposure time, exposure compensation, white balance, image gain and the like, need to be set, so that the quality of the captured flame image is ensured.
Preferentially, the preprocessing method of the flame Image comprises the steps of preprocessing an Image by utilizing a singlechip, wherein the preprocessing comprises median filtering, graying, binarizing and morphological processing of the Image, and then extracting features of the flame Image by using two tools of matlab and Image J; the related processing of the flame image is specifically as follows:
intensity variations were studied by histogram technique corresponding to all three planes (R-plane, G-plane and B-plane). Analysis has shown that median filtering eliminates noise to a greater extent than other types of filters. The image of flames is then subjected to graying and binarization and morphological processing, in particular, a gray image refers to the use of black hues in the image to display objects, i.e. the use of black reference colors, and the use of black colors of different saturation levels to display the image, whereas the color values stored in each pixel of the gray image are also referred to as gray levels, the gray levels generally ranging from 0 to 255, 255 representing white and 0 representing black. Graying of the flame image in the patent; the gray level value of the pixel point on the image is set to be 0 or 255, namely the whole image is obviously provided with only black and white visual effects; carrying out morphological processing on the binary image to thoroughly eliminate noise and extract flame characteristics; edge detection to extract features from the useful portion of the flame image. Various features, such as average intensity, direction, area, centroid, standard deviation, median, mode, etc. of the high temperature flame have been extracted. Tools for feature extraction are MATLAB and Image J. Wherein the flame quality detection algorithm of the present invention comprises the steps of:
(1) And (3) acquiring flame images: capturing a flame image using a flame image acquisition system;
(2) Pretreatment of flame images: preprocessing the captured flame image to extract characteristics of the flame image;
(3) Flame quality detection: and analyzing the characteristics of the flame image by using the trained neural network model to detect flame quality.
Preferably, the flame algorithm for the flame pretreatment stage is:
weighted average method:
G(x,y)＝0.299R(x,y)+0.587G(x,y)+0.114B(x,y)
binarization of flame image:
T(x,y)＝
\begin{cases}
1,&\text{if}G(x,y)>T\\
0,&\text{otherwise}
\end{cases}
morphological processing of images of flames:
F(x,y)＝E(x,y)*T(x,y)
wherein E (x, y) is the result of morphological operations, and T (x, y) is a binary image.
Preferably, the neural network model training of the present invention is through data preparation: preparing a training dataset comprising features of flame images and tags of flame quality; model construction: constructing a multi-layer perceptron (MLP) model, wherein the MLP model comprises an input layer, a hidden layer and an output layer; model training: training the model using the training dataset to fit a flame quality tag; model evaluation: the model is evaluated using the test dataset to determine the accuracy of the model.
Preferably, in step S8, the following steps are specifically included:
s80, data preparation: preparing 100 flame burned images
S81, constructing a model: model construction: a multi-layer perceptron (MLP) model is constructed, including an input layer, a hidden layer, and an output layer.
Input layer: there are 5 neurons set to represent the color, brightness, contour, texture, shape of the flame, respectively, which receive information of each feature in the input data to provide enough information for the hidden layer to perform feature extraction.
The reason for choosing these 5 parameters as metrics is
Color: the color of the flame will typically change with temperature, from blue to yellow to orange to red. Thus, color features may be extracted from the flame image to help determine the extent of combustion of the flame.
Brightness: when a flame burns, its brightness is typically much higher than the surrounding environment, so that the brightness features can be extracted from the flame image.
Profile: the profile of the flame is generally irregular and may vary continuously. By identifying and tracking these profile variations, it is possible to help determine the combustion of the flame.
Texture: flame texture is also an important feature, which is often manifested as distortion and deformation of the flame. By identifying these textural features, the flame stability can be aided.
Shape: the shape of the flame is also generally irregular, and the flame burning condition can be judged by analyzing the shape of the flame image.
Hidden layer: the number of neurons in the hidden layer is a reasonable choice at 450 as determined by constant adjustment and experimentation, with less likelihood that the model will perform poorly and more likelihood that the model will be overfitted.
Output layer: the output layer has a neuron for outputting a binary classification result of whether the flame is sufficiently burned, and 1 indicates sufficient combustion and 0 indicates insufficient combustion. And applying a sigmoid function to convert the output of the neuron into a probability value whose mathematical expression is:
f(x)＝1/(1+exp(-x))
where x is the input of the neuron and f (x) is the output.
S82, model training: firstly, data preprocessing: including image normalization, resizing, augmentation, etc. The data set is then partitioned: the data set is divided into a training set, a verification set and a test set, 70% is divided into the training set, 20% is divided into the verification set, and 10% is divided into the test set. Then building a model: parameters and methods are specifically built according to the MPL architecture design model, and refer to the foregoing. The model is then compiled: a loss function, an optimizer and an evaluation index are selected and the model is compiled. Training a model: the data of the training set is input into the model, training is carried out through a back propagation algorithm, and the weight and bias of the model are updated until a certain training frequency or a certain training precision is achieved. And (3) verifying a model: and inputting data of the verification set into the model, calculating loss and evaluation indexes of the model, and judging generalization capability of the model. And (3) adjusting a model: and (3) adjusting and optimizing the model according to the result of the verification set, such as adjusting model parameters, increasing the neuron number of the hidden layer and the like. Test model: inputting data of the test set into the model, calculating loss and evaluation indexes of the model, and evaluating performance of the model;
s83, model evaluation, namely after multiple experiments, the model calculation model prediction result is high in proportion to the actual label, and the accuracy is high.
Accuracy (Accuracy): the proportion of the model prediction result to the actual label is calculated, and the model prediction result is one of the most commonly used evaluation indexes. The confusion matrix may be used to calculate the accuracy, i.e. the number of correctly classified samples divided by the total number of samples.
Drawings
FIG. 1 is a general flow chart of the present invention;
FIG. 2 is an image acquisition and neural network training of the present invention;
FIG. 3 is a flow chart of the flame height routine of the present invention;
FIG. 4 is a flame threshold determination of the present invention;
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in further detail with reference to the accompanying drawings.
S1, building a machine vision image acquisition system, and acquiring a flame burning video;
s2, performing identification region segmentation on the collected flame burning video through a video segmenter, and identifying a flame image with stable burning;
s3, preprocessing, and adopting median filtering to reduce noise of the flame image;
s4, preprocessing, namely converting the flame image into a gray image by adopting graying;
s5, preprocessing, namely converting the gray level image into a binary image by binarization;
s6, preprocessing, namely performing morphological processing on the binary image to eliminate noise and extract flame characteristics;
s7, extracting features of the flame image by extracting features of the flame image so as to facilitate subsequent neural network model training;
s8, training the neural network by adopting a multi-layer perceptron (MLP) architecture by the neural network model;
s9, testing and verifying. How the camera collects the flame in the boiler
S1, building a machine vision image acquisition system, and acquiring a flame burning video;
the infrared thermal imaging camera adopted by the patent is imaging equipment which integrates a camera, a protective cover, an infrared lamp, a power supply radiating unit and the like into a whole. For reasons of the sealing of the boiler, the camera can collect the flame in the boiler through a thermal imaging technology. The thermal imaging technology can convert heat into visible light, so that a camera can capture an image of flame, therefore, flame data is obtained by controlling the camera to collect corresponding flame burning videos through an infrared thermal camera, the flame video areas are segmented through a video divider, then the data collected by the camera are processed, and in the flame image collection process, shooting parameters of the camera, including exposure time, exposure compensation, white balance, image gain and the like, need to be set, so that the quality of the captured flame image is ensured.
S2, performing identification region segmentation on the collected flame burning video through a video segmenter, and identifying a flame image with stable burning; the flame combustion video is then divided into frames for further analysis. The video divider is used for gradually extracting flame images from the video.
The algorithm of the video divider integrates the image segmentation algorithm, the digital processing and the time correlation of the video sequence to process the flame image of the flame video; image segmentation algorithms typically use spatial information such as color, gray scale, edges, texture, etc. of the image for segmentation. Image segmentation is generally application-specific and can be classified into single-level and multi-level methods. Video segmentation software employs a single-level approach, typically an edge-based approach, or the like. The multi-level method is better applied, such as splitting and combining, morphological method, wavelet method and the like. Since the image segmentation method does not utilize the limitations of the temporal features and other information of the video sequence, the effectiveness of the segmentation algorithm can be improved by considering the temporal correlation of the video sequence. Video segmentation typically uses information of the video image in both spatial and temporal axes.
The flame burning image with high resolution and real-time update rate can be obtained by utilizing video segmentation.
S3, preprocessing, namely adopting median filtering to reduce noise of the flame image
Because the image acquisition process is interfered by the environment, and noise originally carried by the system is added, the noise in the image greatly interferes with the extraction of the flame image, so that the gray image is filtered, and the interference caused by external unstable factors is reduced.
The filtering is done using median filtering. Since the flame image is a color image, intensity variations are studied by comparing histogram techniques corresponding to all three planes (R-plane, G-plane, and B-plane) of various other filters, and various common filtering methods include mean filtering, median filtering, maximum-minimum filtering, and the like, and in order to sufficiently compare the effects of various filtering, the images are filtered by MATLAB software in these several common filtering methods, respectively. Analysis has shown that median filtering eliminates noise to a greater extent than average and adaptive filters. Once the noise is removed, edge detection is performed to extract features from the useful portion of the flame image.
The median filtering of 3*3 neighborhood window is finally selected as a filtering scheme through screening. The main idea of median filtering in the case of digital images stored in the form of a two-dimensional matrix is to choose a filtering window and then process a sequence of two-dimensional digital images, which can be expressed as { x } ij (i,j)∈I 2 The two-dimensional median filtering method can be defined as:
s4, preprocessing, namely converting the flame image into a gray image by adopting graying;
the gray image is to display an object by using black tone, that is, using black reference color, and displaying an image by using black with different saturation, wherein the color value stored in each pixel of the gray image is also called gray, the gray range is generally 0-255, 255 represents white, 0 represents black, and the patent adopts a weighted average method to perform graying treatment;
the weighted average method is to average the weighted average of three RGB color components according to different weights as a final gray value according to the importance of the colors and other indexes. According to the color recognition condition of human eyes, three coefficients of the formula are selected as weights.
Gray(i,j)＝0.299*R(i,j)+0.578*G(i,j)+0.114*B(i,j)；
S5, preprocessing, namely converting the gray level image into a binary image by binarization;
the gray-scale image is set to be 0 or 255, namely the gray value of the pixel point on the image is set to be 0 or 255, namely the whole image is provided with obvious visual effects of only black and white.
Binarization of flame image:
T(x,y)＝
\begin{cases}
1,&\text{if}G(x,y)>T\\
0,&\text{otherwise}
\end{cases}
s6, preprocessing, namely performing morphological processing on the binary image to eliminate noise and extract flame characteristics;
carrying out morphological processing on the binary image to eliminate noise and extract flame characteristics;
morphological processing of images of flames:
F(x,y)＝E(x,y)*T(x,y)
wherein E (x, y) is the result of morphological operations, and T (x, y) is a binary image.
S7, extracting features of the flame image by extracting features of the flame image so as to facilitate subsequent neural network model training;
extracting features of the flame Image through edge detection and using two tools, namely matlab and Image J;
the patent selects flame average gray level, effective area average gray level, flame abundance, high temperature area rate, flame height and flame width as combustion parameters for research, and the following describes the formulas of partial flame characteristic parameter calculation.
Calculating the average gray level of the flame:
the average gray level of the flame is obtained by setting a threshold value.
g i For the gray value of the ith pixel point in the flame digital image, NL is the number of pixel rows and NS is the number of pixel columns.
Calculating the average gray level of the flame effective area:
l (x) is a step function defined as:
E th for the gray threshold value of the set effective average gray, the average gray of the effective area represents the average gray in the effective area where the flame burns, and the average light intensity of the flame in the area can be accurately displayed.
The formula of the flame quality detection algorithm is:
Q＝f(X)
wherein Q is a detection result of flame quality, X is a feature of a flame image, and f (X) is an output of the neural network model.
Calculating flame abundance:
g th for a predetermined threshold of flame abundance, values are taken herein:
g th ＝E th ＝40
high temperature area ratio
The high-temperature area ratio reflects that the area of the flame complete combustion area accounting for the flame image is in the flame complete combustion area, the pulverized coal is fully combusted, the combustion pulsation and the flame area fluctuation are severe, the temperature is highest, and the change of the area of the high-temperature area can reflect the flame combustion stability in the furnace chamber most.
In addition, the average intensity, direction, area, centroid, standard deviation, median, mode, etc. of the high temperature flame have been extracted. Tools for feature extraction are MATLAB and Image J.
S8, training the neural network by adopting a multi-layer perceptron (MLP) architecture by the neural network model;
a multi-layer perceptron (MLP) architecture, including an input layer, a hidden layer, and an output layer. The input layer is used for receiving the characteristics of the flame image, the hidden layer is used for extracting the characteristics of the flame image, and the output layer is used for outputting the detection result of the flame quality.
S80, data preparation: preparing 100 flame burned images
S81, constructing a model: model construction: a multi-layer perceptron (MLP) model is constructed, including an input layer, a hidden layer, and an output layer.
Input layer: there are 5 neurons set to represent the color, brightness, contour, texture, shape of the flame, respectively, which receive information of each feature in the input data to provide enough information for the hidden layer to perform feature extraction.
The reason for choosing these 5 parameters as metrics is
Color: the color of the flame will typically change with temperature, from blue to yellow to orange to red. Thus, color features may be extracted from the flame image to help determine the extent of combustion of the flame.
Brightness: when a flame burns, its brightness is typically much higher than the surrounding environment, so that the brightness features can be extracted from the flame image.
Profile: the profile of the flame is generally irregular and may vary continuously. By identifying and tracking these profile variations, it is possible to help determine the combustion of the flame.
Texture: flame texture is also an important feature, which is often manifested as distortion and deformation of the flame. By identifying these textural features, the flame stability can be aided.
Shape: the shape of the flame is also generally irregular, and the flame burning condition can be judged by analyzing the shape of the flame image.
Hidden layer: the number of neurons in the hidden layer is a reasonable choice at 450 as determined by constant adjustment and experimentation, with less likelihood that the model will perform poorly and more likelihood that the model will be overfitted.
Output layer: the output layer has a neuron for outputting a binary classification result of whether the flame is sufficiently burned, and 1 indicates sufficient combustion and 0 indicates insufficient combustion. And applying a sigmoid function to convert the output of the neuron into a probability value whose mathematical expression is:
f(x)＝1/(1+exp(-x))
where x is the input of the neuron and f (x) is the output.
S82, model training: firstly, data preprocessing: including image normalization, resizing, augmentation, etc. The data set is then partitioned: the data set is divided into a training set, a verification set and a test set, 70% is divided into the training set, 20% is divided into the verification set, and 10% is divided into the test set. Then building a model: parameters and methods are specifically built according to the MPL architecture design model, and refer to the foregoing. The model is then compiled: a loss function, an optimizer and an evaluation index are selected and the model is compiled. Training a model: the data of the training set is input into the model, training is carried out through a back propagation algorithm, and the weight and bias of the model are updated until a certain training frequency or a certain training precision is achieved. And (3) verifying a model: and inputting data of the verification set into the model, calculating loss and evaluation indexes of the model, and judging generalization capability of the model. And (3) adjusting a model: and (3) adjusting and optimizing the model according to the result of the verification set, such as adjusting model parameters, increasing the neuron number of the hidden layer and the like. Test model: inputting data of the test set into the model, calculating loss and evaluation indexes of the model, and evaluating performance of the model;
s83, model evaluation, namely after multiple experiments, the model calculation model prediction result is high in proportion to the actual label, and the accuracy is high.
Accuracy (Accuracy): the proportion of the model prediction result to the actual label is calculated, and the model prediction result is one of the most commonly used evaluation indexes. The confusion matrix may be used to calculate the accuracy, i.e. the number of correctly classified samples divided by the total number of samples.
Precision (Precision) and Recall (Recall): when the sample is unbalanced, the accuracy does not fully reflect the performance of the model. Precision and recall may be used to address this issue. The precision rate indicates the proportion of samples classified as positive samples that are truly positive samples, and the recall rate indicates the proportion of all positive samples that are correctly classified as positive samples. The confusion matrix may be used to calculate the precision and recall. F1 value (F1 Score): taking into account both accuracy and recall, the F1 value can be used to evaluate model performance. The F1 value is the harmonic mean of the precision and recall, i.e., 2 precision recall/(precision + recall).
ROC curve and AUC values: the ROC curve (Receiver Operating Characteristic Curve) is used to evaluate the performance of the classification model at different thresholds, and can be obtained by plotting true and false positive rates at different thresholds. AUC values (Area Under the Curve) represent the area under the ROC curve, typically used to compare the performance of different models, with larger AUC values representing better model performance. Loss Function (Loss Function): the loss function may be used to evaluate the training effect of the model, common loss functions include cross entropy loss, mean square error loss, and the like. Smaller loss function value means better model training effect
S9, testing and verifying.
Claims (9)
1. A combustion quality detection method based on an MLP neural network is characterized by comprising the following steps of S1, building a machine vision image acquisition system, and acquiring flame combustion videos; s2, performing identification region segmentation on the collected flame burning video through a video segmenter, and identifying a flame image with stable burning; s3, preprocessing, and adopting median filtering to reduce noise of the flame image; s4, preprocessing, namely converting the flame image into a gray image by adopting graying; s5, preprocessing, namely converting the gray level image into a binary image by binarization; s6, preprocessing, namely performing morphological processing on the binary image to eliminate noise and extract flame characteristics; s7, extracting features of the flame image by extracting features of the flame image so as to facilitate subsequent neural network model training; s8, training the neural network by adopting a multi-layer perceptron (MLP) architecture by the neural network model; s9, testing and verifying.
2. A neural network-based system according to claim 1The method for detecting the combustion quality of the fuel cell is characterized by comprising the following steps of: the image acquisition system in the step S1 acquires a flame burning video; the problem that solves at first is how the camera gathers the flame in the boiler: the camera can collect flame in the boiler through a thermal imaging technology; the thermal imaging technology can convert heat into visible light, so that a camera can capture an image of flame, and in the flame image acquisition process, the shooting parameters of the camera, including exposure time, exposure compensation, white balance, image gain and the like, need to be set to ensure the quality of the captured flame image. The combustion quality detection method based on the neural network according to claim 1, wherein: and (3) carrying out identification region segmentation on the collected flame burning video by the video in the step (S1) through a video segmenter, and identifying a flame image with stable burning. The combustion quality detection method based on the neural network according to claim 1, wherein: the filtering is done using median filtering. Since the flame image is a color image, intensity variations are studied by comparing histogram techniques corresponding to all three planes (R plane, G plane, and B plane) of various other filters; analysis has shown that median filtering eliminates noise to a greater extent than average and adaptive filters; once the noise is removed, edge detection is performed to extract features from the useful portion of the flame image. The combustion quality detection method based on the neural network according to claim 1, wherein: by gray scale image is meant that objects are displayed in the image with black hues, i.e. with the colors of the black bit standard, and the image is displayed with black of different saturation, whereas the color values stored in each pixel of the gray scale image are also called gray scales, the gray scales typically being in the range of 0-255, 255 representing white and 0 representing black. The graying of the flame image in this patent adopts a weighted average method:。
3. the combustion quality detection method based on the neural network according to claim 1, wherein: the gray image is formed by binarizing the gray value of the pixel point on the image to be 0 or 255, namely, the gray value of the pixel point on the image is set to be 0 or 255, namely, the whole image is displayed with obvious visual effects of only black and white, and the flame image is binarized.
4. The combustion quality detection method based on the neural network according to claim 1, wherein: carrying out morphological processing on the binary image to eliminate noise and extract flame characteristics; morphological processing of images of flames:
5. wherein E (x, y) is the result of morphological operations, and T (x, y) is a binary image.
6. And extracting the average intensity, direction, area, mass center, standard error, median and mode of flames in the effective flame Image by using 2 software and using an edge detection method through edge detection and using two tools to extract characteristics of the flame Image, so that the subsequent neural network model training is realized.
7. The combustion quality detection method based on the neural network according to claim 1, wherein: the step S8 specifically includes the following steps: s80, a neural network model adopts a multi-layer perceptron (MLP) architecture, wherein the neural network model comprises an input layer, a hidden layer and an output layer; the input layer is used for receiving the characteristics of the flame image, the hidden layer is used for extracting the characteristics of the flame image, and the output layer is used for outputting the detection result of the flame quality.
S8, training a neural network model.
9. The neural network model training of the invention comprises the following steps: s80, data preparation: preparing a training data set comprising features of flame images and labels of flame quality S81, model construction: constructing a multi-layer perceptron (MLP) model, wherein the MLP model comprises an input layer, a hidden layer and an output layer, S82, model training: training the model using the training dataset to fit the flame quality label, S83, model assessment: the model is evaluated using the test dataset to determine the accuracy of the model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310300077.XA CN117455828A (en) | 2023-03-26 | 2023-03-26 | Combustion quality detection method based on MLP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310300077.XA CN117455828A (en) | 2023-03-26 | 2023-03-26 | Combustion quality detection method based on MLP neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117455828A true CN117455828A (en) | 2024-01-26 |
Family
ID=89582369
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310300077.XA Pending CN117455828A (en) | 2023-03-26 | 2023-03-26 | Combustion quality detection method based on MLP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117455828A (en) |
-
2023
- 2023-03-26 CN CN202310300077.XA patent/CN117455828A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021000524A1 (en) | Hole protection cap detection method and apparatus, computer device and storage medium | |
| CN108416968B (en) | Fire early warning method and device | |
| Al-Hiary et al. | Fast and accurate detection and classification of plant diseases | |
| US7936377B2 (en) | Method and system for optimizing an image for improved analysis of material and illumination image features | |
| CN110276284A (en) | Flame identification method, device, equipment and storage medium based on video quality assessment | |
| CN111126136A (en) | Smoke concentration quantification method based on image recognition | |
| CN106295124A (en) | Utilize the method that multiple image detecting technique comprehensively analyzes gene polyadenylation signal figure likelihood probability amount | |
| CN103034838A (en) | Special vehicle instrument type identification and calibration method based on image characteristics | |
| CN103914708A (en) | Food variety detection method and system based on machine vision | |
| Laga et al. | Image-based plant stornata phenotyping | |
| CN115526889B (en) | Nondestructive testing method of boiler pressure pipeline based on image processing | |
| CN111582359A (en) | Image identification method and device, electronic equipment and medium | |
| CN112258490A (en) | Low-emissivity coating intelligent damage detection method based on optical and infrared image fusion | |
| CN111931700A (en) | Corn variety authenticity identification method and identification system based on multiple classifiers | |
| CN106951863A (en) | A kind of substation equipment infrared image change detecting method based on random forest | |
| CN116167964A (en) | Tumor classification method and system based on tumor hyperspectral image | |
| CN113192038A (en) | Method for identifying and monitoring abnormal smoke and fire in existing flame environment based on deep learning | |
| CN108960413A (en) | A kind of depth convolutional neural networks method applied to screw surface defects detection | |
| CN117576564B (en) | Disease and pest identification early warning method and system for tea planting | |
| Utaminingrum et al. | Alphabet Sign Language Recognition Using K-Nearest Neighbor Optimization. | |
| Zohra et al. | A linear regression model for estimating facial image quality | |
| Zhang et al. | Fabric defect detection based on visual saliency map and SVM | |
| CN114065798A (en) | Visual identification method and device based on machine identification | |
| Dhanuja et al. | Areca nut disease detection using image processing technology | |
| CN114359539B (en) | Intelligent identification method for high-spectrum image of parasite in sashimi |
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 |