A dual-light fusion-based boiler combustion state detection method for thermoelectric systems

Abstract

The present application relates to a kind of thermoelectric system boiler combustion state detection methods based on dual light fusion, visible light camera and infrared camera are used to collect boiler flame combustion image respectively, after processing, image is fused by dual light fusion algorithm, with the continuous image after fusion as a group of feature sequence input three-dimensional convolution neural network and classify, and the classification obtains boiler flame combustion state;Boiler airflow state is obtained by the infrared image collected;Combining boiler flame combustion state and boiler airflow state, the boiler combustion state is obtained.The present application maintains higher robustness under complex environment and different combustion states, can more comprehensively and accurately reflect the boiler combustion state, improve detection accuracy;It is suitable for various types of thermoelectric system boiler, has strong universality;The combustion state of boiler can be evaluated from multiple levels, the combustion situation inside boiler is accurately reflected in real time;High recognition accuracy, wide application range, strong anti-interference ability.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent control of cogeneration systems, and in particular to a method for detecting the combustion status of boilers in cogeneration systems based on dual-light fusion. Background Technology

[0002] With the continuous growth of industrial production and energy demand, monitoring the flame combustion status of boilers, as key equipment in thermal power systems, is particularly important. Boiler combustion status is closely related to the operating efficiency, safety, and environmental performance of thermal power systems; obtaining boiler combustion status data allows for better control of system operation.

[0003] Traditional indirect methods for detecting boiler flame combustion status primarily rely on visible light or infrared images. However, due to technological limitations, these methods often fail to achieve ideal results in complex industrial environments. Visible light flame detection can monitor the color and shape of the flame, providing intuitive combustion status information. Its advantage lies in its high resolution, clearly displaying flame shape and changes. However, visible light images struggle to provide accurate temperature information and are easily interfered with in high-dust and high-temperature environments. Infrared image flame detection, on the other hand, captures the infrared radiation of the flame, providing information on its temperature distribution. It is less affected by high-temperature environments, can penetrate smoke, and provides reliable boiler flame temperature data. However, infrared images typically have lower resolution, making it difficult to clearly display the detailed shape of the flame, and are susceptible to interference from the infrared radiation of slag, coke, and fly ash. Furthermore, boiler combustion status is also affected by gas flow factors. The stability and uniformity of the gas flow are crucial to the boiler's combustion efficiency and stability, and changes in gas composition directly affect the combustion reaction. Traditional gas flow detection techniques often rely on contact sensors, which are easily affected by environmental factors such as high temperatures and dust, leading to inaccurate measurement results.

[0004] Therefore, a single technical means is insufficient to comprehensively and accurately reflect the combustion state of the boiler flame.

[0005] With advancements in image processing and computer vision technologies, fusion algorithms have seen significant development. This paper aims to design a fusion algorithm that efficiently registers and aligns visible light and infrared images, combining features from both types of images to achieve both high resolution and high temperature accuracy. Compared to traditional single-image techniques, this algorithm will provide more comprehensive and accurate combustion status information in complex industrial environments, thereby significantly improving the operating efficiency and safety of thermoelectric systems. Summary of the Invention

[0006] This invention solves the problems existing in the prior art and provides a method for detecting the combustion status of a boiler in a thermoelectric system based on dual-light fusion.

[0007] The technical solution adopted in this invention is a method for detecting the combustion status of a boiler in a thermoelectric system based on dual-light fusion. The method uses a visible light camera and an infrared camera to acquire boiler flame combustion images respectively. After processing, the images are fused using a dual-light fusion algorithm. The fused continuous images are used as a set of feature sequences and input into a three-dimensional convolutional neural network for classification to obtain the boiler flame combustion status.

[0008] By collecting infrared images, the temperature distribution and radiation characteristics of the gas inside the boiler are detected, and the boiler airflow status is obtained.

[0009] The boiler combustion state is determined by combining the boiler flame combustion state and the boiler airflow state.

[0010] Preferably, the method includes the following steps:

[0011] S1 is equipped with a visible light camera and an infrared camera to acquire visible light images of the boiler flame at the same sampling frequency. v and infrared image I i ;

[0012] S2 in visible light image I v and infrared image I i Feature points are acquired and matched, and the infrared image is then analyzed based on these feature points. i Align to visible light image I v superior;

[0013] S3 will align the infrared image I i 'and visible light image I v Multi-resolution weighted fusion was performed using the Laplacian pyramid to obtain a two-light fused image I. fused ;

[0014] S4 segments the collection of several consecutive frames of dual-light fusion images into feature sub-sequences, and inputs the historically acquired feature sub-sequences into a three-dimensional convolutional neural network for training. Unlike conventional convolutional neural networks, which can only process two-dimensional images, the three-dimensional convolutional neural network here refers to replacing the layers in a conventional convolutional neural network with three-dimensional layers, so that it can be used to process three-dimensional sequence data.

[0015] S5 inputs real-time dual-light fusion image sequences into a trained 3D convolutional neural network to classify the flame combustion state and output the boiler flame combustion state.

[0016] S6 via infrared image I i The sequence characterizes the rate of change of boiler temperature, obtains the airflow velocity v, and uses optical flow to track consecutive frames of infrared images I. i The velocity vector field of the airflow is obtained by measuring the change in the gas temperature pixel. Obtain the boiler airflow status;

[0017] S7 uses the characteristics of the boiler airflow state f airflow Characteristics of boiler flame combustion state f flame Weighted fusion is performed to obtain the boiler combustion state characteristics f. combined This allows us to determine the boiler combustion status.

[0018] Preferably, S2 includes the following steps:

[0019] S2.1 The SIFT algorithm is used to detect feature point sets {P} in visible light and infrared images respectively. v}、{P i};

[0020] S2.2 uses the corresponding feature point set {P v}、{P i The feature descriptor set {D} v} and {D i Calculate the similarity of feature points to obtain the best matching feature point pair;

[0021] S2.3 Based on the matched feature point pairs, an affine transformation is performed on the infrared image to transform the infrared image I... i Align to visible light image I v superior.

[0022] Preferably, in S2.1, P v ={p v1 ,p v2 ,…,p vn}, P i ={p i1 ,p i2 ,…,p im}, where n and m are the feature point sets {P} v}、{P i The number of elements in};

[0023] In S2.2, D v ={d v1 ,d v2 ,…,d vn}, D i ={d i1 ,d i2 ,…,d im}, thus obtaining a set of matching pairs M, satisfying,

[0024]

[0025] Where ε is the similarity threshold, and p vk and p ilThis represents a pair of matching feature points in a visible light image and an infrared image.

[0026] Preferably, S2.3 includes the following steps:

[0027] S2.3.1 Extract the coordinates of the matched feature point pairs to obtain the coordinates of the feature points on the visible light image and the feature point coordinates on the infrared image; in actual operation, a large number of feature point pairs will be obtained, and after affine transformation of each, the incorrect feature point pairs will be filtered out by comparison;

[0028] S2.3.2 Define an affine transformation model T that satisfies,

[0029]

[0030] Where a and b are scaling factors in the x and y directions, respectively, and c and d are rotation factors, t x and t y These are the translation factors in the x and y directions, respectively;

[0031] Let the original coordinates of the matching point in the infrared image be (x, y), and the point after the affine transformation of model T be (x′, y′), satisfying x′=ax+by+t. x y'=cx+dy+t y ;

[0032] S2.3.3 Combine at least three sets of matching point pairs to calculate a, b, c, d, and t in model T. x and t y ;

[0033] S2.3.4 Transfer infrared image I i Model T, after applying the calculated parameters, is aligned to the visible light image I. v Above, the aligned infrared image I i The pixel value at each pixel location (x', y') is equal to the original infrared image I. i The pixel value at the corresponding position (x, y) satisfies the following:

[0034]

[0035] Preferably, S3 includes the following steps:

[0036] S3.1 Visible light image I v and aligned infrared image I i 'Construct the Gaussian pyramid to satisfy...'

[0037]

[0038] Wherein, the superscript of G is the layer number of the Gaussian pyramid image, with layer 0 as the original image, and pyrDown is the downsampling operation;

[0039] S3.2 Based on the obtained Gaussian pyramid, the visible light image I v and aligned infrared image I i 'Construct the Laplace Pyramid to satisfy,'

[0040]

[0041]

[0042] Where pyrUp is the upsampling operation;

[0043] S3.3 performs weighted fusion on each Laplacian image layer to satisfy the following conditions:

[0044]

[0045] Where w1 and w2 are weighting coefficients, satisfying w1 + w2 = 1; obviously, w1, w2 ∈ (0, 1);

[0046] S3.4 Reconstruction of Two-Light Fusion Image Based on Fusion of Laplacian Pyramid I fused ,satisfy,

[0047]

[0048] Preferably, in S4, the set of 50 consecutive dual-light fusion images within 2 seconds is segmented into feature subsequences;

[0049] Here, 25 frames are sampled per second. In order to identify the state of flame combustion, every 50 frames is set as a subsequence (segment) to ensure data continuity. By using long continuous image sequences, the accuracy of classifying combustion states is improved.

[0050] Preferably, in S4, training the model includes the following steps:

[0051] S4.1 Acquire historically acquired dual-light fusion images and obtain feature sub-sequences. Preprocess the feature sub-sequences to meet the input requirements.

[0052] S4.2 Label the feature subsequences according to the historical operating status of the boiler;

[0053] S4.3 Input the feature subsequence into the three-dimensional convolutional neural network to obtain the forward propagation prediction results, and then backpropagate to update the network weights until the training iteration ends.

[0054] Preferably, the three-dimensional convolutional neural network includes an input layer, a three-dimensional convolutional layer, a three-dimensional pooling layer, a residual block group, a global average pooling layer, a fully connected layer, and a Softmax classifier arranged in sequence.

[0055] The residual block group includes several residual blocks, and each residual block includes two three-dimensional convolutional layers, two three-dimensional batch normalization layers and a ReLU activation function arranged in sequence.

[0056] Preferably, S6 includes the following steps:

[0057] S6.1 Through continuous infrared images I i The sequence is used to obtain the airflow velocity v, which satisfies the following conditions:

[0058]

[0059] Where v(x,y) is the airflow velocity at position (x,y), ΔT(x,y) is the temperature change at position (x,y), Δt is the time interval, and α is the thermal diffusivity of the gas.

[0060] S6.2 employs the optical flow method, based on continuous infrared light images I i Sequence to obtain the velocity vector field of airflow satisfy,

[0061]

[0062] in, Let (x,y) be the airflow velocity vector field at position (x,y). and Let be the gradient of the pixel at position (x,y) in the spatial direction. This represents the temporal gradient of the image.

[0063] This invention relates to a method for detecting the combustion state of a boiler in a thermoelectric system based on dual-light fusion. The method involves acquiring boiler flame combustion images using a visible light camera and an infrared camera, respectively. After processing, the images are fused using a dual-light fusion algorithm. The fused continuous images are then used as a set of feature sequences input into a three-dimensional convolutional neural network for classification, thereby obtaining the boiler flame combustion state. The boiler airflow state is obtained through the acquired infrared images. Finally, the boiler combustion state is determined by combining the boiler flame combustion state and the boiler airflow state.

[0064] The beneficial effects of this invention are as follows:

[0065] (1) Combustion status detection is performed using dual-light fusion images. The fused image not only contains detailed information from the visible light image but also contains temperature information from the infrared image. By making full use of the complementary characteristics of the two images, it can maintain high robustness in complex environments and under different combustion states, and can more comprehensively and accurately reflect the combustion status of the boiler, thus improving the detection accuracy.

[0066] (2) Using a non-contact machine vision detection method, it can process various fuel combustion modes, including stoker combustion, chamber combustion and fluidized bed combustion, through different boiler flame datasets. It is applicable to various types of thermal power system boilers and has strong versatility.

[0067] (3) By using continuous multi-frame flame images as input to the three-dimensional classification model, the changes in flame combustion state can be dynamically captured and analyzed. By combining airflow state detection, the combustion state of the boiler can be evaluated from multiple levels. The dynamic changes in airflow and the flame combustion state work together to enable the present invention to reflect the combustion situation inside the boiler in real time and accurately.

[0068] This invention has the advantages of high recognition accuracy, wide applicability, and strong anti-interference ability. Attached Figure Description

[0069] Figure 1 This is a flowchart of the boiler combustion state detection method of the present invention;

[0070] Figure 2 This is a flowchart of the dual-light fusion algorithm of the present invention;

[0071] Figure 3 This is a structural diagram of the three-dimensional convolutional neural network model of the present invention. Detailed Implementation

[0072] The present invention will be further described in detail below with reference to embodiments, but the scope of protection of the present invention is not limited thereto.

[0073] This invention relates to a method for detecting the combustion status of a boiler in a thermoelectric system based on dual-light fusion. The method uses a visible light camera and an infrared camera to acquire boiler flame combustion images respectively. After processing, the images are fused using a dual-light fusion algorithm. The fused continuous images are then used as a set of feature sequences and input into a three-dimensional convolutional neural network for classification to obtain the boiler flame combustion status.

[0074] The boiler airflow status is obtained by collecting infrared images;

[0075] The boiler combustion state is determined by combining the boiler flame combustion state and the boiler airflow state.

[0076] Specifically, the method includes the following steps:

[0077] S1 is equipped with a visible light camera and an infrared camera to acquire visible light images of the boiler flame at the same sampling frequency. v and infrared image I i ;

[0078] S2 in visible light image I v and infrared image I i Feature points are acquired and matched, and the infrared image is then analyzed based on these feature points. i Align to visible light image I v superior;

[0079] S3 will align the infrared image I i 'and visible light image I v Multi-resolution weighted fusion was performed using the Laplacian pyramid to obtain a two-light fused image I. fused ;

[0080] S4 segments the collection of several consecutive frames of dual-light fusion images into feature subsequences, and inputs the historically acquired feature subsequences into a three-dimensional convolutional neural network for training.

[0081] S5 inputs real-time dual-light fusion image sequences into a trained 3D convolutional neural network to classify the flame combustion state and output the boiler flame combustion state.

[0082] S6 via infrared image I i The sequence characterizes the rate of change of boiler temperature, obtains the airflow velocity v, and uses optical flow to track consecutive frames of infrared images I. i The velocity vector field of the airflow is obtained by measuring the change in the gas temperature pixel. Obtain the boiler airflow status;

[0083] S7 uses the characteristics of the boiler airflow state f airflow Characteristics of boiler flame combustion state f flame Weighted fusion is performed to obtain the boiler combustion state characteristics f. combined This allows us to determine the boiler combustion status.

[0084] The method will be further explained below with reference to specific implementation methods.

[0085] S1 is equipped with a visible light camera and an infrared camera to acquire visible light images of the boiler flame at the same sampling frequency. v and infrared image I i ;

[0086] Visible light images of boiler flame combustion are acquired using a binocular camera. v ;

[0087] Infrared images of boiler flames are acquired using a binocular infrared camera. i ;

[0088] In practical applications, visible light cameras and infrared cameras are integrated into a single binocular camera to facilitate the acquisition of images at different positions.

[0089] S2 in visible light image I v and infrared image I i Feature points are acquired and matched, and the infrared image is then analyzed based on these feature points. i Align to visible light image I v superior.

[0090] (2-1) Using the SIFT algorithm to detect the set of feature points {P} in a visible light image. v}, the set of feature points of an infrared image {P i},include:

[0091] Using the SIFT algorithm in visible light image I v The set of feature points to be detected in the middle {P v};

[0092] P v ={p v1 ,p v2 ,…,p vn}

[0093] Using the SIFT algorithm in infrared image I i The set of feature points to be detected in the middle {P i};

[0094] P i ={p i1 ,p i2 ,…,p im}

[0095] (2-2) Use the corresponding feature descriptor set {D} v} and {D i Calculate the similarity of feature points and find the best match, including:

[0096] For visible light image I v Feature point set {P v}, its feature descriptor is {D v};

[0097] D v ={d v1 ,d v2 ,…,d vn}

[0098] For infrared image I i Feature point set {P i}, its feature descriptor is {D i};

[0099] D i ={d i1 ,d i2 ,…,d im}

[0100] By comparing the similarity of feature descriptors, the best match for each pair of feature points is found. The similarity is calculated using Euclidean distance, and the matching function is expressed as:

[0101] M={(p vk ,p il )|||d vk -d il ||≤ε}

[0102] Where M is the set of matching pairs, and ε is the similarity threshold; here p vk and p il This represents a pair of matching feature points in a visible light image and an infrared image;

[0103] (2-3) Using matched feature points, perform an affine transformation on the infrared image to transform the infrared image I... i Align to visible light image I v Above, including:

[0104] (2-3-1) Extract the coordinates of the matching point pairs, where the coordinates S of the matching point on the visible light image are... v The coordinates of the matching point S on the infrared image i ,

[0105] S v ={(x vk ,y vk )|(p vk ,p il )∈M}

[0106] S i ={(x il ,y il )|(p vk ,p il )∈M}

[0107] (2-3-2) Calculate the affine transformation model T using matching point pairs;

[0108] In affine transformations, the transformation matrix T describes linear transformations and translations from one plane to another. The affine transformation matrix T is typically represented as a 3×3 matrix, but one row is kept to be [0,0,1] because an affine transformation is a linear transformation plus a translation. Specifically, it is represented as:

[0109]

[0110] Where a and b are scaling factors, representing scaling in the x and y directions respectively; c and d are rotation factors, representing the degree of image rotation; t x and t y It is the translation factor, which represents the translation of the image in the x and y directions;

[0111] The matching point in the original infrared image is (x, y), and the point after affine transformation is (x′, y′), which can be calculated using the following formula:

[0112]

[0113] The calculation yields:

[0114] x' = ax + by + t x

[0115] y'=cx+dy+t y

[0116] (2-3-3) To calculate the matrix parameters in the affine transformation model T, at least three matching point pairs are required. Let the matching point pairs be... Locate points n = 1, 2, 3, such that these matching points satisfy the following equation:

[0117]

[0118] These equations can be represented as a system of linear equations:

[0119]

[0120] By solving this system of linear equations, the parameters a, b, c, d, and t of the affine transformation matrix can be obtained. x and t y ;

[0121] (2-3-4) Infrared image I i The transformation model T calculated by the application is aligned to the visible light image I. v This requires the aligned infrared image I. i The pixel value at each pixel location (x', y') is equal to the original infrared image I. i The formula for the pixel value at the corresponding position (x, y) is as follows:

[0122] I i '(x',y') T =T×I i (x,y) T

[0123] The infrared image I can be obtained using the above formula. i Each pixel position is aligned to the visible light image I through an affine transformation.v Above, where I i 'This is the aligned infrared image.'

[0124] S3 will align the infrared image I i 'and visible light image I v Multi-resolution weighted fusion was performed using the Laplacian pyramid to obtain a two-light fused image I. fused .

[0125] (3-1) Visible light image I v and aligned infrared image I i The formula for constructing a Gaussian pyramid is as follows:

[0126]

[0127]

[0128] The superscript of G indicates the number of layers in the Gaussian pyramid image, with layer 0 being the original image. pyrDown represents the downsampling operation, which reduces the image size by half. Each layer of the Gaussian pyramid image is the downsampled result of the previous layer.

[0129] (3-2) Based on the obtained Gaussian pyramid, the visible light image I v and aligned infrared image I i The formula for constructing the Laplace pyramid is as follows:

[0130]

[0131]

[0132] (3-3) where pyrUp represents the upsampling operation, and then the Laplacian images of each layer are weighted and fused, as shown in the following formula:

[0133]

[0134] Among them, w1 and w2 are weighting coefficients, satisfying w1 + w2 = 1;

[0135] (3-4) Then, the two-light fused image I is reconstructed from the fused Laplace pyramid. fused :

[0136]

[0137] S4 segments the collection of several consecutive frames of dual-light fusion images into feature sub-sequences, and inputs the historically acquired feature sub-sequences into a three-dimensional convolutional neural network for training.

[0138] Specifically, to combine 50 consecutive dual-light fusion images within two seconds {Ifused The segments are divided into feature subsequences, and the historically collected feature subsequences are input into a three-dimensional convolutional integral class network for training.

[0139] In this invention, boiler flame combustion is a dynamic process, and the boiler combustion state cannot be measured solely from the features of a single dual-light fusion image. Therefore, in order to make the detection model focus on the dynamic information of boiler combustion, the historically sampled dual-light fusion image data is segmented into subsequences of 50 frames within two seconds. At this point, the input data has changed from two-dimensional image data to three-dimensional sequence data with time-series information. Traditional two-dimensional image classification networks have difficulty processing this type of data well, while three-dimensional convolutional classification networks extend two-dimensional convolution to three-dimensional convolution in order to capture features in the time series, thereby better capturing the dynamic characteristics of the boiler flame combustion state.

[0140] The two-light fused image subsequences are input into a 3D convolutional integral class network for training, including:

[0141] (4-1) Data Preprocessing: The historically acquired dual-light fusion images are divided into subsequences of 50 frames within two seconds, and each subsequence is used as the input to the three-dimensional convolutional classification network. Simultaneously, to improve training performance, the input data needs to be preprocessed, uniformly adjusting the size of each image to H×W. The shape of the network input layer is then (N, C, H, W), where N is the number of frames (50 in this embodiment), C is the number of channels (the dual-light fusion image has four channels), and H and W are the height and width of the image, respectively.

[0142] (4-2) Data labeling: According to the historical operating status of the boiler, the collected data subsequences are labeled. In this embodiment, the boiler flame combustion status is divided into seven states: low load stable combustion, low load unstable combustion, high load stable combustion, high load unstable combustion, load increase combustion process, load decrease combustion process, and flame extinguishing.

[0143] (4-3) Model Construction: The 3D convolutional classifier network used in this embodiment is a 3D ResNet neural network. This network structure includes a (N,C,H,W) shaped network input, 3D convolutional layers and 3D pooling layers, stacked residual blocks, a global average pooling layer, a fully connected layer, and a Softmax classifier, outputting combustion state categories. The 3D convolutional layers apply 3D convolution operations to the input data, outputting feature maps.

[0144] Conv3D(N,C,H,W)→(N,C',H',W')

[0145] Where Conv3D represents the 3D convolution operation, C' is the number of output channels, and H' and W' are the height and width of the convolutional feature map. The 3D pooling layer downsamples the feature map to reduce data dimensionality while retaining important features.

[0146] MaxPool3D(N,C',H',W')→(N,C',H”,W”)

[0147] Where MaxPool3D represents the three-dimensional max pooling operation, and H” and W” are the height and width of the feature map after pooling;

[0148] Following the first 3D convolutional layer and 3D pooling layer are several residual blocks, each containing two 3D convolutional layers, two 3D batch normalization layers, and a ReLU activation function:

[0149] Residual Block=Conv3D(C',C')→BatchNorm3D→ReLU

[0150] →Conv3D(C',C')→BatchNorm3D

[0151] Where Residual Block represents a residual block and BatchNorm3D represents a three-dimensional batch normalization layer; by stacking residual blocks, a deeper network structure can be built, resulting in better performance in classification tasks;

[0152] The global average pooling layer performs global average pooling on the feature map and outputs a feature vector of fixed size, represented as:

[0153] GlobalAvgPool3D(N,C',H',W')→(N,C')

[0154] Where GlobalAvgPool3D represents the global average pooling operation. The fully connected layer maps the feature vector to the classifier output, uses the Softmax function to calculate the probability of each class, and the class with the highest probability is the class predicted by the model. The final output is the class of the boiler flame combustion state.

[0155] FC(N,C')→(N,num_classes)

[0156] Softmax(N,num_classes)→(num_classes)

[0157] Where FC represents a fully connected layer, and num_classes represents the categories.

[0158] The training process includes the following steps:

[0159] (4-3-1) Forward propagation is the process of passing the input data through the network layer by layer to finally obtain the predicted output, that is, the process of obtaining the category output after the image enters the three-dimensional convolutional network.

[0160] (4-3-2) Loss function calculation: The output of the three-dimensional convolutional classification network is the predicted classification of the combustion state of the boiler flame. The cross-entropy loss function can be used to measure the classification error.

[0161] (4-3-3) Backpropagation for parameter update: The backpropagation algorithm calculates the gradient of the loss function with respect to the model parameters, thereby guiding the update of the model parameters. Then, the optimization algorithm (such as stochastic gradient descent and Adam optimizer) updates the model parameters based on the calculated gradient to minimize the loss function;

[0162] (4-3-4) Training iteration: Repeat the above process for several rounds (called epochs) until the loss function converges or the predetermined number of training iterations is reached.

[0163] After the S5 is trained, the real-time dual-light fusion image sequence is input into the trained three-dimensional convolutional neural network to classify the flame combustion state and output the boiler flame combustion state.

[0164] S6 via infrared image I i The sequence characterizes the rate of change of boiler temperature, obtains the airflow velocity v, and uses optical flow to track consecutive frames of infrared images I. i The velocity vector field of the airflow is obtained by measuring the change in the gas temperature pixel. Obtain the boiler airflow status;

[0165] (6-1) Through continuous infrared light images I i The airflow velocity v is estimated by the infrared imaging temperature change in the sequence, satisfying the following conditions:

[0166]

[0167] Where v(x,y) is the airflow velocity at position (x,y), ΔT(x,y) is the temperature change at position (x,y), Δt is the time interval, and α is the thermal diffusivity of the gas.

[0168] (6-2) Using optical flow method to process continuous infrared light images I i The sequence is estimated to estimate the velocity vector field of the airflow. satisfy,

[0169]

[0170] in for The airflow velocity vector field at the location, This represents the gradient of the pixel at position (x, y) in the spatial direction of the image. This represents the temporal gradient of the image.

[0171] S7 uses the characteristics of the boiler airflow state f airflow Characteristics of boiler flame combustion state f flame Weighted fusion is performed to obtain the boiler combustion state characteristics f. combined This allows us to determine the boiler combustion status.

[0172] In practical applications,

[0173] f combined =w1·f airflow +w2·f flame

[0174] Here, w1 and w2 are weights, representing the importance of airflow features to the flame image classification results. By adjusting the weights, the contribution of different features can be optimized according to the actual application.

[0175] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for detecting the combustion state of a boiler of a thermoelectric system based on dual light fusion, characterized by: The method comprises the following steps: S1, acquiring boiler flame combustion images respectively by a visible light camera and an infrared camera; A visible light camera and an infrared camera are arranged to capture visible light images and infrared images of the boiler flame at the same sampling frequency and infrared images ; S2 feature points are acquired on the visible light image and the infrared image and matched, the infrared image is aligned to the visible light image based on the feature points; S3, fusing the images after processing by a dual-light fusion algorithm; aligning the infrared image and the visible light image using multi-resolution weighted fusion using Laplacian pyramids to obtain a dual-light fused image ; S4, segmenting a collection of continuous dual-light fusion images into feature subsequences, and inputting the historical acquisition feature subsequences into a three-dimensional convolutional neural network for training; S5, inputting the fused continuous images as a group of feature sequences into the three-dimensional convolutional neural network for classification, and obtaining a boiler flame combustion state through classification; S6, inputting a real-time dual-light fusion image sequence into the trained three-dimensional convolutional neural network, classifying the flame combustion state, and outputting a boiler flame combustion state; S6, detecting the temperature distribution and radiation characteristics of the gas in the boiler through the acquired infrared images, and obtaining a boiler gas flow state; The rate of change of the temperature of the boiler is characterized by a sequence of infrared images The rate of change of the temperature of the boiler is characterized by a sequence of infrared images The rate of change of the temperature of the boiler is characterized by a sequence of infrared images The rate of change of the temperature of the boiler is characterized by a sequence of infrared images S7, combining the boiler flame combustion state and the boiler gas flow state to obtain a boiler combustion state; By weighting fusion of the features of the boiler gas stream state and the features of the boiler flame combustion state a boiler combustion state feature is obtained a boiler combustion state is derived.

2. The method according to claim 1, wherein the method is a dual optical fusion based thermoelectric system boiler combustion state detection method. S2 comprises the following steps: S2.1 detect feature point sets in the visible light image and the infrared image respectively using a SIFT algorithm , ; S2.2 using the corresponding feature point set , the feature descriptor set of the feature point and calculate the similarity of the feature points, and obtain the most matched feature point pair; S2.3 Based on the matched feature points, the infrared image is affine transformed to align with the visible light image to the visible light image above.

3. The dual optical fusion based thermoelectric system boiler combustion state detection method of claim 2, wherein: In S2.1, , wherein n and m are the number of elements in the set of feature points , , respectively. In S2.2, , , obtaining a matching pair set M, satisfying ， wherein, is a similarity threshold, and and denote pairs of matching feature points in the visible light image and the infrared image.

4. The dual optical fusion based thermoelectric system boiler combustion state detection method of claim 3, wherein: S2.3 comprises the following steps: S2.3.1, extracting the coordinates of the matched feature point pairs to obtain the coordinates of the feature points on the visible light image and the coordinates of the feature points on the infrared image; S2.3.2, defining an affine transformation model T, satisfying, ， where a and b are scaling factors in the x and y directions, respectively, and c and d are rotation factors, and are translation factors in the x and y directions, respectively; Let the original coordinates of the matching points in the infrared image be (x, y), and the points after the model T affine transformation be (x', y'), satisfying , ; S2.3.

3. Calculate a, b, c, d in model T by solving the following equations simultaneously with at least three sets of matched point pairs, and ; S2.3.4 Aligning the infrared image Applying the computed parameters to the model T to align it to the visible light image The aligned infrared image The pixel value at each pixel position (x', y') is equal to the pixel value at the corresponding position (x, y) in the original infrared image The pixel value at each pixel position (x', y') is equal to the pixel value at the corresponding position (x, y) in the original infrared image 。 5. The dual optical fusion based thermoelectric system boiler combustion state detection method of claim 1, wherein: S3 comprises the following steps: S3.1 To the visible image and the aligned infrared image Constructing a Gaussian pyramid, satisfying ， ， ， ， Wherein, the superscript of G is the number of layers of the Gaussian pyramid image, and the 0th layer is the original image; pyrDown is a downsampling operation; S3.2 Constructing a Laplacian pyramid from the obtained Gaussian pyramid and the aligned infrared image Constructing a Laplacian pyramid, satisfying, ， ， Wherein, pyrUp is an upsampling operation; S3.3, weighted fusion is performed on each layer of Laplacian image, satisfying, ， wherein and are weighting coefficients satisfying ; S3.4 Reconstructing dual light fusion images based on fused Laplacian pyramid , meet, 。 6. The dual optical fusion based thermoelectric system boiler combustion state detection method of claim 1, wherein: In S4, a collection of 50 continuous dual-light fusion images within 2 seconds is segmented into feature subsequences.

7. The dual optical fusion based thermoelectric system boiler combustion state detection method of claim 6, wherein: In S4, training the three-dimensional convolutional neural network model comprises the following steps: S4.1, acquiring historical acquisition dual-light fusion images, and obtaining feature subsequences, and pre-processing the feature subsequences to meet the input requirements; S4.2, labeling the feature subsequences according to the historical operation state of the boiler; S4.3, inputting the feature subsequences into the three-dimensional convolutional neural network to obtain the forward propagation prediction results, and updating the network weights through back propagation until the training iteration ends.

8. The dual optical fusion based thermoelectric system boiler combustion state detection method according to claim 1 or 7, characterized in that: The three-dimensional convolutional neural network comprises an input layer, a three-dimensional convolutional layer, a three-dimensional pooling layer, a residual block group, a global average pooling layer, a full connection layer and a Softmax classifier arranged in sequence; The residual block group comprises a plurality of residual blocks, and any residual block comprises two three-dimensional convolutional layers, two three-dimensional batch normalization layers and a ReLU activation function arranged in sequence.

9. The dual optical fusion based thermoelectric system boiler combustion state detection method of claim 1, wherein: S6 comprises the following steps: S6.1 The air flow velocity v is acquired by a sequence of successive infrared images satisfying v = v0+ v1t ， wherein is the air flow velocity at the position, is the change in temperature at the position, is the time interval, is the thermal diffusion coefficient of the gas; S6.2 Adopting the optical flow method, based on successive infrared light images The sequence acquires the velocity vector field of the air flow , meet, ， in, for The airflow velocity vector field at the location, and For the image in The gradient of a pixel at a given location in the spatial direction. This represents the temporal gradient of the image.

Images