A method and system for calculating the ash shrinkage rate of a cigarette

Abstract

The present application provides a cigarette ash packaging shrinkage rate calculation method and detection system, which belongs to the technical field of cigarette detection. The cigarette ash packaging shrinkage rate calculation method comprises: acquiring multi-angle images before and during cigarette combustion, and pre-processing the multi-angle images; collecting a cigarette image area before cigarette combustion, and automatically labeling the cigarette area boundary using an image processing algorithm based on adaptive boundary tracking; collecting an ash packaging image area after cigarette combustion, and ensuring the correspondence of the image areas before and after combustion through image feature point matching; based on a convolutional neural network, segmenting the cigarette image area and the ash packaging image area respectively and obtaining the segmented feature map; in the segmented feature map, identifying the boundaries of the cigarette area and the ash packaging area, and extracting the effective connected domains in the cigarette area and the ash packaging area; and based on the extracted effective connected domains, calculating the cigarette ash packaging shrinkage rate and outputting the calculation result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cigarette detection, in particular to a cigarette ash shrinkage rate calculation method and detection system. BACKGROUND

[0002] In the quality control process of cigarette production, the burning performance of cigarettes is one of the important indicators for evaluating cigarette products. The burning process of cigarettes directly affects the experience of consumers, and the ash cracking rate is one of the key parameters for measuring the burning quality of cigarettes. The ash cracking rate is usually determined by calculating the ratio of the cracking area in the ash column area after the cigarette is burned to the total ash area. The lower the cracking rate, the higher the integrity of the ash column, and the stronger the ability of the cigarette to maintain the structure during the burning process. Therefore, the detection and analysis of the cracking rate can not only be used for quality control, but also provide data support for the optimization of cigarette materials and formulations.

[0003] Traditional cracking rate detection methods rely on manual detection or simple image analysis techniques, usually based on visual judgment or basic image processing algorithms such as threshold segmentation and edge detection. These methods have problems such as low detection accuracy, detection results easily affected by light, and unclear cracking features. In addition, due to the complex surface morphology of the ash column, manual detection is inefficient and the objectivity of the results is difficult to guarantee, especially in large-scale production detection. With the development of deep learning and image processing technology, detection methods based on deep learning have begun to show significant advantages in industrial detection. These methods can automatically extract features and accurately identify images through deep learning models, allowing for automated and precise detection and segmentation of complex ash column cracks.

[0004] However, in actual cigarette detection, the cracking area of the ash column often has diversity, including irregular shape, uneven area, and fuzzy edge. In addition, the light contrast between the cracking and non-cracking areas is different, and the degree of burning also affects the clarity of the image. These factors pose a great challenge to the detection of cigarette cracking rate. Existing image segmentation methods based on convolutional neural networks (CNN) and attention mechanisms, such as YOLO, ResNeXT, Swin-Transformer, etc., have been used in similar industrial detection scenarios. However, how to further improve the detection accuracy and robustness to achieve automatic detection and accurate segmentation of cigarette cracks in complex backgrounds is still a research focus in this field.

[0005] In view of this, the present inventors, based on years of production and design experience in this field and related fields, have designed a cigarette ash shrinkage rate calculation method and cigarette ash shrinkage rate detection system through repeated trials, in order to solve the problems existing in the prior art. SUMMARY

[0006] The application aims to provide a cigarette ash shrinkage rate calculation method and detection system, which can effectively improve the detection efficiency and accuracy of the cigarette ash shrinkage rate.

[0007] To achieve the above-mentioned application purposes, the application provides a cigarette ash shrinkage rate calculation method and detection system, wherein, as shown in the accompanying drawings, Figure 1 The cigarette ash shrinkage rate calculation method comprises:

[0008] acquiring multi-angle images before and during the combustion of the cigarette, and pre-processing the multi-angle images;

[0009] collecting a cigarette image area before the combustion of the cigarette, and automatically labeling the boundary of the cigarette area by using an image processing algorithm based on adaptive boundary tracking;

[0010] collecting an ash image area after the combustion of the cigarette, and ensuring the correspondence of the image areas before and after the combustion by image feature point matching;

[0011] segmenting the cigarette image area and the ash image area based on a convolutional neural network respectively and obtaining a segmented feature map;

[0012] identifying the boundary of the cigarette area and the boundary of the ash area in the segmented feature map, and extracting effective connected domains in the cigarette area and the ash area;

[0013] calculating the cigarette ash shrinkage rate based on the extracted effective connected domains and outputting the calculation result.

[0014] The application also provides a cigarette ash shrinkage rate detection system, characterized in that the system realizes the method for achieving the above-mentioned purposes, and comprises the following modules:

[0015] An image acquisition module is used to acquire multi-angle images before and after the combustion of the cigarette;

[0016] An ash area segmentation module is used to segment the ash area based on a parallel convolutional neural network and an adaptive gating mechanism;

[0017] A feature extraction and fusion module is used to perform multi-scale feature fusion on the ash area and remove background noise;

[0018] An area calculation and shrinkage rate detection module is used to calculate the shrinkage rate based on a shrinkage rate formula;

[0019] A data analysis and output module is used to generate a shrinkage rate trend chart and a comprehensive detection report.

[0020] The application also provides a computer readable storage medium, which stores a computer program, characterized in that the computer program is executed by a processor to realize the steps of the above-mentioned calculation method.

[0021] The application further provides a computer device, comprising a memory and a processor, wherein the computer program capable of running on the processor is stored on the memory, and the processor implements the steps of the calculation method when executing the computer program.

[0022] Compared with the prior art, the application has the following characteristics and advantages:

[0023] The cigarette ash shrinkage rate calculation method and detection system provided by the application collect image data in the cigarette combustion process through a multi-angle imaging system, automatically label the ash column region boundary before combustion by using an adaptive boundary tracking algorithm, and ensure the region consistency of the images before and after combustion by using an image registration algorithm. In combination with a parallel convolutional neural network (CNN) and an adaptive gating mechanism, the ash column region is automatically segmented, non-target regions are removed through multi-scale feature fusion and connected domain analysis technology, and cracks or irregular edges are repaired. Finally, the shrinkage rate is obtained by calculating the area ratio of the ash region before and after combustion, a shrinkage rate trend chart is automatically generated, dynamic analysis of the cigarette ash shrinkage rate is realized, the automation level of cigarette combustion quality detection is improved, human factors are reduced, the stability and reliability of the detection result are improved, and important technical support is provided for cigarette production quality management. BRIEF DESCRIPTION OF DRAWINGS

[0024] The drawings described herein are only for illustrative purposes, and are not intended to limit the scope of the present application in any way. In addition, the shapes and scale sizes of the components in the drawings are only illustrative, and are used to help understand the present application, and are not specific limitations on the shapes and scale sizes of the components of the present application. Those skilled in the art can select various possible shapes and scale sizes to implement the present application according to specific circumstances under the guidance of the present application.

[0025] Figure 1 A schematic diagram of the ash shrinkage rate detection process of the present application;

[0026] Figure 2 A schematic diagram of the CSegNet network structure of the present application;

[0027] Figure 3 A schematic diagram of the ResNeXTR network structure of the present application;

[0028] Figure 4 A schematic diagram of the residual block of the present application;

[0029] Figure 5 A schematic diagram of the encoder structure of the Swin-Transformer of the present application;

[0030] Figure 6a A schematic diagram of the ECBAM module structure of the present application;

[0031] Figure 6b Structure diagram of the ECAM module of the present application;

[0032] Figure 6c Structure diagram of the SAM module of the present application;

[0033] Figure 7 Structure diagram of the JEUNet network of the present application;

[0034] Figure 8 Structure diagram of the joint excitation upsampling JEU module of the present application;

[0035] Figure 9 Cigarette ash shrinkage rate calculation method of the present application. DETAILED DESCRIPTION

[0036] The details of the present application can be more clearly understood with reference to the drawings and the description of specific embodiments of the present application. However, the specific embodiments of the present application described herein are for the purpose of explanation and should not be considered in any way limiting of the present application. Based on the teachings of the present application, a person skilled in the art can conceive any possible modification of the present application, which should be considered as falling within the scope of the present application.

[0037] The present application provides a cigarette ash shrinkage rate calculation method and a cigarette ash shrinkage rate detection system, wherein the cigarette ash shrinkage rate calculation method comprises:

[0038] Obtaining multi-angle images before and during cigarette combustion, and pre-processing the multi-angle images;

[0039] Collecting a cigarette image area before cigarette combustion, and automatically labeling the cigarette area boundary using an image processing algorithm based on adaptive boundary tracking;

[0040] Collecting an ash image area after cigarette combustion, and ensuring the correspondence of the image areas before and after combustion through image feature point matching;

[0041] Segmenting the cigarette image area and the ash image area based on a convolutional neural network, and obtaining a segmented feature map;

[0042] In the segmented feature map, identifying the boundaries of the cigarette area and the ash area, and extracting the effective connected domains in the cigarette area and the ash area;

[0043] Calculating the cigarette ash shrinkage rate based on the extracted effective connected domains, and outputting the calculation result.

[0044] The cigarette ash shrinkage rate calculation method and the cigarette ash shrinkage rate detection system provided by the application ensure the quality and consistency of image data through multi-angle imaging and image preprocessing, provide a reliable basis for subsequent image analysis, improve the accuracy and efficiency of cigarette ash shrinkage rate calculation, improve the detection efficiency, can realize real-time dynamic analysis of cigarette combustion quality, and provide important technical support for cigarette production quality management.

[0045] In an optional embodiment of the application, multi-angle images before and during cigarette combustion are obtained through a multi-angle imaging system, and the multi-angle images at least include front view images, side view images and top view images. By using the multi-angle imaging system technology, more comprehensive and accurate image data can be obtained when calculating the cigarette ash shrinkage rate, effectively capturing various details of the ash column during cigarette combustion, including length changes, surface cracks and color changes of the ash column.

[0046] In an optional embodiment of the application, the cigarette area and the ash area are segmented based on a convolutional neural network, including using a multi-scale convolution processing model combined with ResNeXT50 and Swin-Transformer network structure, combining a Squeeze-and-Excitation module for feature channel enhancement, and realizing the separation of the cigarette area and the ash area from the background area. The cigarette ash shrinkage rate calculation method using ResNeXT50 and Swin-Transformer network structure combination and combining the Squeeze-and-Excitation module significantly improves the accuracy and efficiency of segmentation. By combining the traditional convolutional neural network and the self-attention mechanism, not only local features can be captured, but also long-distance dependencies in the image can be identified, so that the network can more comprehensively understand the image content. The introduction of the Squeeze-and-Excitation module further enhances the network's control ability over the feature channel, so that the model can pay more attention to the key information for distinguishing the cigarette area and the ash area.

[0047] In an optional embodiment of the present application, the cigarette area and the ash area are segmented based on a convolutional neural network, which includes, in the encoder part, using a convolutional neural network module to extract multi-scale features from the image, dividing the feature map into multiple stages according to different scales, and enhancing the feature channel correlation through a Squeeze-and-Excitation module; in the decoder part, the feature map is upsampled through a joint excitation upsampling module, and the spatial and channel attention mechanisms are combined to improve the segmentation accuracy. By using multi-scale feature extraction and Squeeze-and-Excitation module in the encoder part to enhance the feature channel correlation, and by using JEU module for upsampling in the decoder part and combining spatial and channel attention mechanisms, the accuracy of cigarette ash area segmentation is significantly improved. This method not only can capture the subtle features in the image, but also can understand the image content at different scales, so as to more accurately separate the cigarette area and the ash area. This precise segmentation capability ensures that background interference can be excluded when calculating the ash reduction rate, improving the accuracy of ash reduction rate calculation and providing more reliable data support for cigarette quality control. In addition, this method also improves the calculation efficiency of the segmentation process, making the calculation of cigarette ash reduction rate faster, which helps to realize real-time or near real-time quality monitoring.

[0048] In an optional embodiment of the present application, the cigarette area and the ash area are segmented based on a convolutional neural network, which includes, based on a parallel convolutional neural network, using an adaptive gating mechanism to dynamically enhance the boundaries of the cigarette area and the boundaries of the ash area, and identifying multi-scale region features through a multi-scale feature extraction module. The cigarette ash reduction rate calculation method using parallel convolutional neural network and adaptive gating mechanism significantly improves the accuracy of boundary recognition and the quality of segmentation. The adaptive gating mechanism enables the network to dynamically adjust according to the actual image features, effectively enhancing the boundary features and reducing the influence of noise and interference. The application of the multi-scale feature extraction module enables the network to consider features at different scales at the same time, which not only improves the recognition ability of small-scale cracks and defects, but also enhances the ability to capture large-scale structural changes. Through the comprehensive segmentation method, more accurate region definition can be obtained when calculating the ash reduction rate, thereby improving the accuracy of ash reduction rate calculation and providing more reliable data support for cigarette quality control.

[0049] In an optional embodiment of the present application, in the segmented feature map, the boundaries of the cigarette area and the boundaries of the ash area are identified, including fine identification of the boundaries of the cigarette area and the boundaries of the ash area through processing by a depth estimation and adaptive weight allocation image enhancement module. By applying the depth estimation and adaptive weight allocation image enhancement module in the segmented feature map, the identification accuracy of the boundaries of the cigarette area and the ash area is improved. It can automatically adapt to different image features and environmental conditions, provide more accurate boundary positioning, and thus provide a solid foundation for subsequent area measurement and ash reduction rate calculation. Through this fine identification, errors caused by inaccurate boundary identification can be effectively reduced, and the reliability and accuracy of the ash reduction rate calculation are improved. In addition, the adaptability of the system to different brands and types of cigarettes is also enhanced

[0050] In an optional embodiment of the present application, the effective connected domain in the cigarette area and the ash area is extracted, including using multi-scale feature fusion and connected domain analysis technology to eliminate non-target regions in the background. By using multi-scale feature fusion and connected domain analysis technology, the extraction accuracy of the effective connected domain in the cigarette ash area is improved. Non-target regions in the background are effectively removed, reducing irrelevant information interference, thereby ensuring the accuracy and reliability of the ash reduction rate calculation. By accurately identifying and extracting the region directly related to the cigarette burning quality, the present application not only improves the precision of the ash reduction rate calculation, but also enhances the adaptability of the system to different burning stages and different types of cigarettes.

[0051] In an optional example of this embodiment, multi-scale feature fusion is used, including through a multi-scale feature fusion module, the features extracted by multi-scale convolution kernels are fused into a unified feature map, and the weights of each feature channel are adaptively adjusted by combining a Squeeze-and-Excitation module to highlight the ash column region features. The cigarette ash reduction rate calculation method using multi-scale feature fusion and Squeeze-and-Excitation module effectively improves the quality of feature extraction and the accuracy of segmentation. By integrating feature maps of different scales, the network can capture both local details and global context information, which helps to better understand the image content. The adaptive weight adjustment of the SE module further enhances the network's ability to identify ash column region features, making the feature map more prominent in features related to the ash column, thereby improving the accuracy of segmentation.

[0052] In an optional example of the embodiment, the effective connected domain in the cigarette area and the ash area is extracted, and the crack or irregular edge is repaired by a region filling algorithm. By introducing the region filling algorithm in the process of extracting the effective connected domain, the accuracy and reliability of the cigarette ash shrinkage rate calculation can be improved, and the crack or irregular edge in the image can be effectively repaired, avoiding the negative impact of these defects on the shrinkage rate calculation. By filling the crack, the present application ensures the integrity of the ash column area, making the calculation result closer to the actual burning quality, and providing more accurate data support for quality control in the cigarette production process.

[0053] In an optional example, based on the adaptive gating mechanism and the connected domain analysis of the convolutional network, the crack removal is performed through edge detection and connected domain filling. By combining the adaptive gating mechanism and the connected domain analysis technology, the crack in the cigarette ash area can be efficiently removed, not only improving the accuracy of crack detection, but also repairing the image discontinuity caused by the crack through intelligent filling technology, thereby ensuring the integrity of the ash column area and the accuracy of the shrinkage rate calculation.

[0054] Preferably, the crack removal includes: using a Sobel operator or a Canny edge detection algorithm to extract the crack edge; embedding an adaptive gating module in a deep convolutional network for enhancing crack features and suppressing non-crack features; extracting the connected domain of the crack area and screening the feature area according to area and shape; filling the crack area and performing color measurement. By comprehensively using the edge detection algorithm, the adaptive gating module and the connected domain analysis technology, the crack in the cigarette ash area can be efficiently and accurately removed. This method not only improves the accuracy of crack detection, but also repairs the image discontinuity caused by the crack through intelligent filling technology, thereby ensuring the integrity of the ash column area and the accuracy of the shrinkage rate calculation.

[0055] In an optional embodiment of the present application, the cigarette ash shrinkage rate calculation method further includes: performing integrity detection on the cigarette area and the ash area based on color channel analysis. By introducing the integrity detection step based on color channel analysis, the cigarette ash shrinkage rate calculation method can more comprehensively evaluate the burning quality of the cigarette. Color differences caused by uneven burning or ash column structure defects can be identified, thereby providing important information about the burning characteristics of the cigarette. Through color channel analysis, the present application not only improves the accuracy of shrinkage rate calculation, but also enhances the comprehensive evaluation ability of the quality of the cigarette product.

[0056] In an optional embodiment of the present application, the formula for calculating the cigarette ash shrinkage rate is:

[0057] SR = (1-S1 / S0) x 100

[0058] In the formula, S0 is the area of the cigarette before burning, with the unit of square millimeter, and S1 is the area of the cigarette after burning, with the unit of square millimeter. By accurately measuring the area of the ash-covered region before and after the cigarette is burned and applying a specific calculation formula, this method can reliably reflect the integrity and stability of the ash column during the burning process of the cigarette. Using a quantitative evaluation method not only improves the accuracy of quality control, but also helps manufacturers optimize the design and production process of cigarettes, thereby improving the overall quality of the product and the consumer experience.

[0059] In an optional embodiment of the present application, the output calculation result includes: outputting the shrinkage rate calculation result and generating a shrinkage rate change trend chart during the burning process. By outputting the shrinkage rate calculation result and generating the change trend chart, a comprehensive and intuitive cigarette burning quality analysis tool is provided for the user. This output method not only makes complex data information easy to understand and analyze, but also helps cigarette manufacturers and researchers identify and analyze the performance of cigarettes at different burning stages. Through this information, manufacturers can improve product design and optimize burning characteristics to improve the overall quality of cigarettes.

[0060] The present application also proposes a cigarette ash covering shrinkage rate detection system, wherein the system implements the above method and includes the following modules:

[0061] An image acquisition module is used to acquire multi-angle images before and after the cigarette is burned.

[0062] An ash-covered region segmentation module is used to segment the ash-covered region based on a parallel convolutional neural network and an adaptive gating mechanism.

[0063] A feature extraction and fusion module is used to perform multi-scale feature fusion on the ash-covered region and remove background noise.

[0064] An area calculation and shrinkage rate detection module is used to calculate the shrinkage rate based on the shrinkage rate formula.

[0065] A data analysis and output module is used to generate a shrinkage rate change trend chart and a comprehensive detection report.

[0066] The present application also proposes a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the above calculation method.

[0067] The present application also proposes a computer device including a memory and a processor, wherein the memory stores a computer program capable of running on the processor, and the processor implements the steps of the above calculation method when executing the computer program.

[0068] Specifically, as shown in the accompanying drawings, Figure 1 a cigarette ash covering shrinkage rate calculation method includes the following steps:

[0069] 1) Input the sample image of the cigarette burning process to be detected, and obtain the ash cover image at different angles and burning stages through a multi-angle imaging system to ensure the comprehensiveness and accuracy of the image data;

[0070] 2) Obtain the image area before the ash cover area burns, automatically label the ash cover area boundary using an image processing algorithm based on adaptive boundary tracking, and ensure the area consistency of the images before and after burning through an image registration algorithm;

[0071] 3) After the cigarette burns, re-collect the ash cover area image, and ensure the correspondence of the images before and after burning through image feature point matching to improve the accuracy of the ash cover rate calculation;

[0072] 4) Automatically segment the ash cover area through a parallel convolutional neural network (CNN) and an adaptive gating mechanism, use a multi-scale convolution processing model combined with the ResNeXT50 and Swin-Transformer network structures, and combine the Squeeze-and-Excitation (SE) module for feature channel enhancement to ensure accurate separation of the ash column area and the background area;

[0073] 5) In the segmented feature map, refine the ash column area boundary, process it through a depth estimation and adaptive weight distribution image enhancement module to achieve fine identification of the ash cover area boundary, and ensure the accuracy of the area measurement;

[0074] 6) Use multi-scale feature fusion and connected domain analysis technology to remove non-target areas in the background, extract the effective connected domain of the ash cover area, and repair cracks or irregular edges through a region filling algorithm;

[0075] 7) After segmenting the ash cover area before and after burning, calculate the ash cover rate SR through SR = (1-S1 / S0) x 100;

[0076] Where S0 is the area of the ash cover area before burning, with the unit of square millimeter (mm 2 ), and S1 is the area of the ash cover area after burning, with the unit of square millimeter (mm 2 );

[0077] 8) Output the ash cover rate detection result and generate the ash cover rate trend chart during the burning process for dynamic analysis of the cigarette ash cover ash cover rate.

[0078] 1.1 Image processing

[0079] 1.1.1 Image acquisition module

[0080] A high-resolution static imaging camera is used to acquire static images of the carbon line during the cigarette burning process.

[0081] A dynamic imaging camera is used to capture the real-time changes in the carbon line image produced during cigarette combustion.

[0082] By using dual-camera cooperative imaging, the carbon line features of the cigarette combustion can be captured at different time nodes and perspectives, ensuring clear images even in complex background conditions.

[0083] 1.1.2 Image preprocessing

[0084] The collected raw image data is processed by the preprocessing module, including:

[0085] Color calibration: Use standard color cards to calibrate the color of the image to eliminate color deviation under different lighting conditions.

[0086] Light uniformity adjustment: According to the change of environmental light, adjust the light uniformity of the image to ensure the clarity and consistency of the cigarette carbon line image during the combustion process.

[0087] Image denoising: Through noise filtering algorithm, remove the interference factors such as smoke and ash produced during the combustion process of cigarette, enhance the quality and effectiveness of the image.

[0088] 1.2 Image area acquisition and labeling of ash area before combustion

[0089] Before the cigarette is lit, the initial image of the ash area is acquired. Through the adaptive boundary tracking algorithm, the boundary of the ash area is automatically labeled. This algorithm detects the boundary of the initial image to accurately mark the outline of the ash area before combustion, ensuring effective differentiation between the ash area and the background in subsequent analysis. Then, image registration algorithm is used to align the images before and after combustion, ensuring the consistency of the image area before and after combustion.

[0090] 1.3 Image acquisition and feature point matching of ash area after combustion

[0091] After the cigarette combustion ends or at a specific stage, the image of the ash area is collected again. Through image feature point matching technology, the images before and after combustion are compared to ensure the alignment and correspondence of the image area. The purpose of this step is to reduce the image deviation caused by angle change or ash column shrinkage, and improve the accuracy of shrinkage rate calculation.

[0092] 1.4 CSegNet network

[0093] In order to realize accurate segmentation of the package ash crack, the application proposes a crack segmentation algorithm (Crack Segmentation Network, CSegNet) combining convolutional neural network (CNN) and Transformer self-attention mechanism. The CSegNet network adopts an Encoder-Decoder coding and decoding framework similar to DeepLab V3+, and includes the following steps:

[0094] 1) Obtain the crack image in the cigarette combustion process, and pre-process the image, including noise removal, gray calibration and contrast enhancement;

[0095] 2) In the encoder, a ResNeXt-Transformer (ResNeXTR) module is used, which includes a ResNeXt50 convolutional network and a Swin-Transformer network, for extracting local and global features in the image;

[0096] 3) In the decoder, an efficient convolution block attention module (ECBAM) is combined, including an efficient channel attention module (ECAM) and a spatial attention module (SAM), for strengthening the salient features of the crack region;

[0097] 4) Use average pooling and point-wise convolution operations to reduce the dimension of the encoder output to reduce the cost of self-attention calculation;

[0098] 5) Use a combined loss function of Binary Cross-Entropy with Logits and Dice loss, the Binary Cross-Entropy with Logits loss and the Dice loss function are combined together with a weight of 0.5 to optimize the segmentation accuracy and edge detection effect.

[0099] The main innovations of CSegNet are as follows:

[0100] (1) In the encoder Encoder part, a ResNeXt-Transformer (ResNeXTR) feature extraction module is designed as the backbone network, ResNeXTR combines the flexibility of Transformers in modeling long-term dependencies and the ability of CNNs to capture local context information through local receptive fields, weight sharing and spatial subsampling, enhancing the feature extraction capability of the crack region in the package ash image. In addition, in order to reduce the computational cost of Transformers, an average pooling layer is introduced to downsample the dimension of the encoded features.

[0101] (2) In the decoder part, the Efficient Channel Attention Module (ECAM) and the Spatial Attention Module (SAM) are combined to design an Efficient Convolutional Block Attention Module (ECBAM) to further optimize feature representation and focus on the key information of the crack under different background colors.

[0102] (3) The ReLU activation function, SGD gradient descent and Binary Cross Entropy with Logits are used to accelerate the convergence speed and improve the segmentation accuracy.

[0103] 1.4.1 Overview of the CSegNet Algorithm

[0104] The CSegNet network model adopts the basic encoder-decoder architecture of DeepLab V3+, such as Figure 2 As shown. Given the flexibility of Transformer in long-term dependency modeling, and the ability of CNN to acquire local contextual information using local receptive fields, shared weights, and spatial subsampling, a ResNeXt-Transformer (ResNeXTR) feature extraction module was designed as the backbone network in the encoder to enhance the feature extraction capability of the input image. The CNN part of the ResNeXTR feature extraction module uses a ResNeXt50 network as the convolutional feature extractor, and the Transformer part uses a Swin-Transformer network based on a self-attention mechanism. To reduce the computational cost of self-attention calculation, average pooling is introduced to downsample the encoded dimension. For example... Figure 5 As shown, after the image passes through the encoder, in order to further enrich the feature information and enhance the feature extraction capability of the model, the shallow features from ResNet50 are concatenated with the output of the Swin Transformer to form a shallow output, which is then input into the decoder and fused with the deep feature map output by ResNet50 to form the comprehensive output of the encoder.

[0105] The decoder retains the structure of DeepLab V3+, but designs an ECBAM attention module before outputting the classification label, which enhances the long-range interaction between low-level feature maps without significantly increasing the parameter size, thereby further optimizing the feature representation. After the decoder obtains the multi-scale semantic feature map, it performs 4 times upsampling, then fuses the shallow semantic information output by the ResNeXTR module, and after 3x3 convolution and 4 times upsampling, the features pass through the ECBAM attention module to obtain the semantic segmentation result.

[0106] 1.4.2 ResNeXTR network structure

[0107] The ResNeXTR feature extraction module structure for feature extraction in the encoder is shown in Figure 3 ResNeXTR uses ResNeXt50 network as the CNN feature extractor on one hand, which can learn more extensive and rich local features from the package gray crack image; on the other hand, it uses Swin-Transformer network based on self-attention mechanism as the Transformer component in the module, which enables the network to capture the long-distance correlation between different regions in the crack image. By combining the features output by the two networks, ResNeXTR constructs a feature engineering with rich shallow features and strong local feature correlation, which balances the segmentation efficiency and accuracy to a certain extent.

[0108] 1.4.2.1 ResNeXt

[0109] The use of ResNeXt network is mainly based on the following factors.

[0110] (1) ResNeXt50 introduces "cardinality" as a new dimension, which is on par with depth and width, and improves the representation ability of the model without increasing the complexity of the model.

[0111] (2) The modular design of ResNeXt50 makes the model more flexible, easy to extend, and can be trained more efficiently.

[0112] (3) ResNeXt50 performs better than Xception, the initial encoder network of DeepLab V3+, on the ImageNet dataset, and has excellent generalization ability on multi-task learning and different datasets. Therefore, using ResNeXt50 as part of the feature extraction network can improve the model performance and enhance the stability and robustness of the model. The ResNet network of ResNeXt50 is composed of multiple residual blocks (Residual Blocks), as shown in Figure 4The ResNeXt50 network is shown in FIG. 2, and is distributed in 4 stages according to the configuration of [3, 4, 6, 3]. Through these residual blocks, the ResNeXt50 network conducts in-depth feature analysis on the input image, generates multi-scale feature maps, and ensures accurate and rich semantic representation of the input image. In actual training, we first use the ResNeXt50 network pre-trained on the public dataset ImageNet to perform preliminary feature extraction on the input image.

[0113] 1.4.2.2 Average Pooling Layer Embedded Transformer

[0114] In addition to the above-mentioned ResNeXt module, the Swin Transformer V2, which is also pre-trained on ImageNet, is introduced to analyze the shallow features of the input image and construct detailed feature maps. Swin-Transformer V2 is an efficient visual Transformer network structure, unlike traditional Transformer networks, Swin-Transformer V2 processes images by using windowed self-attention mechanism and introduces local information exchange to reduce computational complexity. In our model, the self-attention mechanism of Swin-Transformer helps to capture more local and global details, thereby providing a more fine-grained feature representation for the semantic label generation of the crack. Since the original attention has relatively large amount of calculation and high requirements for equipment, in order to reduce the amount of model calculation, the CSegNet improves the basic attention calculation, and after calculating the K, V vectors, it is down-sampled, and the specific structure is as shown in FIG. 3. Figure 4

[0115] In order to reduce the amount of operation in the self-attention calculation, the average pooling layer O S The dimension of the encoded is down-sampled. In order to further reduce the amount of calculation, point-wise convolution is also used to further reduce the feature dimension. The specific calculation process of self-attention SA is as follows:

[0116] SA(X) = Attention(X·W Q ,P S (X·W K ),P S (X·W V )) (2.1)

[0117] where W Q , W K , W V represent the linear layers used for encoding, and Attention is the attention calculation method, and its calculation formula can be represented as:

[0118] ​Attention(Q, K, V) = softmax(SIM(Q, K))V (2.2)

[0119] SIM(Q, K) is the similarity calculation of Q and K, using cosine similarity attention, and its calculation formula is as follows:

[0120]

[0121] where B ij is the relative position bias between pixel i and pixel j; τ is a learnable scaling parameter; B is obtained by the continuous relative position bias method, which is not directly optimized parameterized bias, but uses a small meta-network to update on the relative coordinates:

[0122]

[0123] where, denotes a small network, for example, a 2-layer MLP with ReLU activation by default. The Transformer block is composed of self-attention (SA) and MLP modules, and the calculation process is as follows:

[0124]

[0125]

[0126]

[0127]

[0128] where, denotes the result of layer normalization (LN) of the output of self-attention SA, denotes the result of layer normalization (LN) of the output of MLP, z l denotes the final output result of the Transformer block.

[0129] 1.4.4 Efficient Convolution Block Attention Module ECBAM

[0130] In order to enable the network to effectively focus on the key information of the crack in the complex background while suppressing irrelevant information from different channels, an efficient convolution attention module (Efficient Convolutional Block Attention Module, ECBAM) is designed in the output part of the decoder of CSegNet. The structure of ECBAM is as follows: Figure 6aAs shown in Fig. 2. The ECBAM first inputs the feature map into the ECAM (Efficient Channel Attention Module) module as shown in Fig. 3 to reduce the increase in model parameter count caused by the introduction of the mixed attention mechanism module. The ECAM uses 1D convolution to effectively realize the cross-channel interaction of local information. Then, the feature map is input into the SAM (Spatial Attention Module) module as shown in Fig. 4. The SAM simultaneously performs maximum pooling and average pooling on the feature map, and performs overall weighting operation on the feature map after splicing the results. This design can make ECBAM pay attention to the global semantic information of the defect area while introducing the channel correlation of the feature map. Figure 6b Figure 6c

[0131] 1.4.5 Loss function

[0132] In the training and verification stage of the model, a joint combination loss function combining the Binary Cross-Entropy with Logits (BCE With Logits Loss) loss function and the Dice loss function is designed.

[0133] (1) Binary Cross-Entropy with Logits loss function is a loss function commonly used in binary classification problems, which combines Sigmoid activation function and binary cross-entropy loss. In the crack segmentation task, the BCE With Logits function provides explicit feedback for the model, and its definition is shown in formula 2.9:

[0134]

[0135] In the formula, N represents the total number of samples; y i represents the true label of the i-th sample, 1 represents that the pixel is a crack, and 0 represents that the pixel does not belong to the crack; p i is the original output of the model prediction for the i-th pixel. σ(p i ) represents the result obtained by processing the original output of the model prediction for the i-th pixel through the sigmoid function.

[0136] (2) Dice loss function is different from Binary Cross-Entropy. Dice loss function measures the overlap between labels and predictions. Assuming that X represents the true label of the real crack segmentation image, and Y represents the pixel category of the model prediction segmentation image, the definition of the Dice loss function is shown in formula 2.10 and formula 2.11:

[0137] L Dice ​​= 1 - Dice Coefficient (2.10)

[0138]

[0139] where |X∩Y| represents the number of intersection elements between X and Y, and |X| and |Y| represent the number of elements in X and Y, respectively. The final loss function used by the model is shown in Equation 2.12:

[0140] L = 0.5L BCEwtihlogits + 0.5L Dice (2.12)

[0141] This formula combines the BCE With Logits loss and the Dice loss function with a weight of 0.5 each. The combination of these two loss functions helps to balance the accuracy of crack detection and boundary delineation in the segmentation task.

[0142] 1.5 Research on JEUNet-based Crack Semantic Segmentation Algorithm

[0143] The research motivation of CSegNet is to achieve accurate segmentation of the ash-laden crack area. However, due to the embedding of ResNeXt50, Swin-Transformer V2, ECBAM, and other modules, the parallel operation of these modules requires a large amount of computing resources, which is not conducive to the engineering of the algorithm. Therefore, research on efficient semantic segmentation algorithms is needed to provide a fast and accurate algorithm for the engineering implementation of ash-laden crack detection.

[0144] To address the problem of existing semantic segmentation algorithms requiring large computing resources while achieving accurate segmentation, this section designs a joint excitation up-sampling network (JEUNet) for crack segmentation. JEUNet is a network that contains an encoder-decoder symmetric structure. In the decoder stage, spatial attention is integrated with joint up-sampling to replace the general layer-by-layer up-sampling method. It also considers the correlation of multi-scale feature information and channel direction, which enhances the accuracy of semantic segmentation while improving the speed of up-sampling.

[0145] 1.5.1 Overview of JEUNet Algorithm

[0146] Ash-laden crack detection can be described as a pixel-level anomaly prediction task. The proposed joint excitation up-sampling network JEUNet has a classic encoder-decoder symmetric structure, and its network framework is as follows: Figure 7The yellow arrows represent direct input; the black arrows represent 3x3 convolution operations with a step size of 1, which use a Zero-padding strategy to keep the feature maps of the same level at the same size; the gray arrows represent the cut and connection operations, which connect the left feature maps after cutting them to the size corresponding to the right side; the blue arrows represent the 1x1 operation for the final classification, and the last two layers output are the results and the background; the red arrows represent the down-sampling of the feature maps through 2x2 max pooling; and the green arrows represent the 2x up-sampling of the feature maps using the 2x2 deconvolution operation. k represents the number of basic channels of the convolution feature map (in this paper, k = 32).

[0147] The encoder part of the JEUNet is consistent with the backbone network of the UNet. The original input image is encoded through successive convolution layers in 4 stages, and is transmitted between every 2 stages through 2x max pooling down-sampling. Therefore, there are a total of 5 scales including the original image, and the feature maps of each scale contain information with different receptive fields, with the shallow feature maps mainly being the detailed texture information of local pixels, and the deep feature maps containing the semantic information of the local region in the image. In the decoding process of the JEUNet, the outputs of the 2nd, 3rd and 4th stages are input into the designed JEU (Joint Excitation Up-sampling block) module, which replaces the three-scale convolution continuous up-sampling in the original UNet. The output of the JEU module is up-sampled once and concatenated with the output of the 1st stage of the encoder, and then two 3x3 convolution operations are performed in the same layer, and finally up-sampled to the original image size and connected with the initial convolution result of the original image, and the prediction result is obtained after two convolution operations. In the decoder part, the up-sampled feature map and the shallow feature of the encoder part are concatenated to improve the prediction accuracy of the local pixels.

[0148] 1.5.2 JEUNet module

[0149] The joint excitation up-sampling JEU module is a kind of calculation unit that integrates spatial attention into the joint up-sampling network. On the one hand, 3 feature maps with different receptive fields are adjusted to the same size and connected as a joint feature map T; on the other hand, a transformation vector W is constructed to excite each channel in the tensor T. In the up-sampling process, the JEU module simultaneously considers the correlation between multi-scale feature information and channel direction, improves the accuracy of semantic segmentation, and reduces the complexity of up-sampling calculation. Figure 8 The structure diagram of the JEU block is shown, the purple arrow represents the joint up-sampling path, and the red arrow represents the feature map excitation path.

[0150] The present application JEUNet uses joint upsampling to replace the initial two layer-by-layer upsampling of the decoder link in UNet. Through embedding vector extraction, the multi-level feature maps output by the 2nd, 3rd and 4th stages are input into the designed JEU module, and the feature maps of stage 4 (w x w) and stage 3 (2w x 2w) are upsampled to the same size as stage 2, 4w x 4w. The concatenated operation is performed on the feature maps after joint upsampling, and a multi-level feature map T with a size of 4w x 4w x X (X = 4k + 8k + 16k = 28k = 896) is obtained. After joint upsampling, the new 3 feature maps are combined into a multi-level feature map T, which has a size of 4w x 4w x X, where X = 4k + 8k + 16k = 28k = 896. The multi-level feature map T contains 4w x 4w embedding vectors with a size of 1 x X, each embedding vector contains semantic information of 3 different receptive fields, and each embedding vector is one-to-one corresponding to each position (x, y) in the original image.

[0151] 1.5.3 Multi-level feature map spatial channel excitation

[0152] After joint upsampling, the JPU block then uses four different expansion rates of dilated convolution to extract features and perform splicing, with the purpose of capturing feature information at different scales in the multi-level feature map to ensure the accuracy of the semantic segmentation upsampling decoder. However, each scale of feature map in the multi-level feature map itself contains semantic information of different receptive fields, and the use of multiple dilated convolutions may repeat the multi-level embedding vector itself, resulting in no obvious effect on the improvement of semantic segmentation accuracy. In addition, the multi-level feature map T contains X channels, each channel is obtained by convolution kernel operation with different parameters, and the relevance between these channels has not been paid attention to, so its accuracy improvement for semantic segmentation prediction results is limited.

[0153] In order to adaptively recalibrate the feature response of the channel direction and model the mutual dependence between different channels. We introduce the SE (Squeeze-and-Excitation) module into the JEU block, which integrates spatial attention into the structure of the upsampling network, which is also a significant difference between JEU and JPU modules. As shown in formula 2.13, for the Tensor T of 4w x 4w x X, a global average pooling function is used to compress it into a 1 x X feature vector F, and the xth element F x of F is calculated as follows:

[0154]

[0155] In the formula, F(x) corresponds to the xth channel of Tensor T, and I and J are the xth channel feature map T xwhere I and J are the height and width of the feature map (I = J = 4w), and i and j represent the coordinates of each point on the feature map. To establish the relevance between different channel features, F is connected with two fully connected layers to calculate the contribution weight of different channels during the learning process, and to activate the corresponding channel in the feature map during prediction, as shown in equation (2.14).

[0156]

[0157] where δ() represents the ReLU activation function and σ() represents the Sigmoid activation function.

[0158] During the learning process, the feature vector F is mapped through two fully connected layers to form the final weight vector W, and the weight of each channel in W is x The importance of the corresponding channel x in the multi-level feature map T is predicted, thereby modeling the correlation between feature channels. In the operation of the two fully connected layers, the vector F is scaled (1 / 2) and restored once to reduce the amount of calculation, w1 is the mapping weight vector of the first fully connected layer, and w2 is the mapping weight vector of the second fully connected layer. represents the feature map T corresponding to the xth channel x and the weight vector W x The weighted multiplication of the excitation feature map, the JEU block gives each channel c in the multi-level feature map T obtained by joint upsampling a new response weight through the weight vector W, and finally obtains the multi-level feature map after spatial channel excitation

[0159] 1.6 Crack region removal

[0160] 1) Crack edge detection: Apply Sobel operator or Canny edge detection algorithm in the gray column segmentation map to extract the edge information in the image. By analyzing the connectivity of the edge image, the edge features of the crack region are preliminarily identified.

[0161] 2) Adaptive gating mechanism application: Introduce an adaptive gating module (Adaptive Gating Mechanism) in the feature extraction layer of the convolutional neural network. This module adjusts the feature points through a dynamic gating function (such as the Sigmoid function), gives higher weights to the edge features of the crack region, and attenuates the noise interference of the gray column, to improve the accuracy of crack detection.

[0162] 3) Connected component analysis: Perform connected component analysis on the image after adaptive gating processing. First, extract all connected components that may belong to cracks, and filter these connected components by area size, shape contour, etc. to remove small or irregular connected components, and ensure that only the true crack region is retained.

[0163] 4) Crack filling: After determining the crack area, fill the crack area by morphological closing operation or other region filling algorithms. If the crack area is large, use a multi-scale filling strategy, i.e. fill from the center to the edge layer by layer, to ensure that the filled ash cover image is complete and intact, so as to facilitate subsequent color detection.

[0164] 1.7 Color feature extraction

[0165] 1) Color space conversion: Convert the ash cover area image after removing the crack from the RGB color space to the CIELAB color space. This conversion process can better simulate the human eye's perception of color and adapt to changes in different lighting environments.

[0166] 2) Color channel separation: In the CIELAB space, extract the L (brightness), a (red-green component) and b (yellow-blue component) values. On this basis, calculate the average value of L values of all pixels in the ash cover area to represent the color gray information of the ash column. The average value of L value can accurately reflect the depth of ash cover color.

[0167] 1.7 Color calibration and gray balance

[0168] 1) D65 light source calibration: Based on the D65 light source of the CIELAB color space standard, calibrate the image color to ensure that the detection result is not affected by the change of external light, especially to maintain consistent color measurement under different environmental lighting.

[0169] 2) Gray balance: Use the gray histogram equalization method to process the ash cover image after removing the crack, dynamically adjust the image brightness and contrast, and make the gray distribution more uniform, so as to ensure the stability of ash cover color detection in different combustion stages.

[0170] 1.8 Detection of shrinkage rate

[0171] After completing the ash column area segmentation and removing the crack, calculate the shrinkage rate according to the segmented ash cover area before and after combustion. The specific calculation formula (2.15) is used for calculation:

[0172]

[0173] Where S0 is the area of ash cover area before combustion, with the unit of square millimeter (mm 2 ), S1 is the area of ash cover area after combustion, with the unit of square millimeter (mm 2 ).

[0174] The above detailed description of the embodiments is only intended to explain the present application, so that the present application can be better understood, but these descriptions cannot be interpreted as limiting the present application in any way, and in particular, the features described in different embodiments can be arbitrarily combined with each other to form other embodiments, and these features should be understood as being applicable to any one embodiment, and not limited to the described embodiments, except for the explicitly opposite description.

Claims

1. A method for calculating the ash reduction rate of a cigarette ash filter, characterized by, The cigarette ash shrinkage rate calculation method comprises: Obtaining multi-angle images before and during cigarette combustion, and pre-processing the multi-angle images; Collecting a cigarette image area before cigarette combustion, and automatically labeling the cigarette area boundary using an image processing algorithm based on adaptive boundary tracking; Collecting an ash image area after cigarette combustion, and ensuring the correspondence of the image areas before and after combustion through image feature point matching; Based on the convolutional neural network, the cigarette image area and the ash image area are segmented respectively, and the segmented feature maps are obtained; in the segmented feature maps, the boundaries of the cigarette area and the ash area are identified, and the effective connected domains in the cigarette area and the ash area are extracted; Repairing cracks or irregular edges in the ash area through a region filling algorithm; Using JEUNet for crack detection; during the decoding process of JEUNet, the outputs of the 2nd, 3rd and 4th stages are input into the designed JEU module, which replaces the three-scale convolutional continuous up-sampling in the original UNet; the output of the JEU module is up-sampled once and concatenated with the output of the 1st decoding stage, and then two 3x3 convolution operations are performed in the same layer, and finally up-sampled to the original image size and connected with the initial convolution result of the original image, and the prediction result is obtained after two convolution operations; in the decoder decoding part, the up-sampled feature map and the shallow layer feature of the encoder part are concatenated to improve the prediction accuracy of the local pixels; Through embedded vector extraction, the multi-level feature maps output by the 2nd, 3rd and 4th stages are input into the designed JEU module, and the feature maps of the 4th stage and the 3rd stage are up-sampled to the same size as the 2nd stage, and the concatenated feature maps are concatenated to obtain a multi-level feature map T with a size of 4w×4w×X. For the multi-level feature map T with a size of 4w×4w×X, a global average pooling function is used to compress it into a 1×X feature vector F. The feature vector F is mapped through two fully connected layers to form the final weight vector W, and the weight value W of each channel in W x The importance of the corresponding channel x in the multi-level feature map T is predicted, thereby modeling the correlation between the feature channels. The JEU module gives each channel c in the multi-level feature map T obtained by joint up-sampling a new response weight through the weight vector W, and finally obtains a multi-level feature map after spatial channel excitation ; Based on the extracted effective connected domain, the cigarette ash shrinkage rate is calculated and the calculation result is output.

2. The method for calculating the ash shrinkage rate of cigarette packs as described in claim 1, characterized in that, Obtaining multi-angle images before and during cigarette combustion through a multi-angle imaging system, the multi-angle images at least including front view images, side view images and top view images.

3. The method for calculating the ash shrinkage rate of cigarette packs as described in claim 1, characterized in that, Based on the convolutional neural network, the cigarette area and the ash area are segmented respectively, including, A multi-scale convolution processing model combining ResNeXT50 and Swin-Transformer network structures is used to enhance the feature channel and realize the separation of the cigarette area and the ash area from the background area.

4. The method for calculating the ash shrinkage rate of cigarette packs as described in claim 1, characterized in that, Based on the convolutional neural network, the cigarette area and the ash area are segmented respectively, including, In the encoder part, a convolutional neural network module is used to extract multi-scale features from the image, and the feature maps are divided into multiple stages according to different scales, and the Squeeze-and-Excitation module is used to enhance the feature channel correlation; In the decoder part, the feature maps are up-sampled through the joint excitation up-sampling module, and the spatial and channel attention mechanisms are combined to improve the segmentation accuracy.

5. The method for calculating the ash shrinkage rate of cigarette packs as described in claim 1, characterized in that, The cigarette area and the ash area are segmented based on a convolutional neural network, including, The boundaries of the cigarette area and the ash area are dynamically enhanced using an adaptive gating mechanism based on a parallel convolutional neural network, and multi-scale region features are identified through a multi-scale feature extraction module.

6. The method for calculating the ash shrinkage rate of cigarette packs as described in claim 1, characterized in that, In the segmented feature map, the boundaries of the cigarette area and the ash area are identified, including, The boundaries of the cigarette area and the ash area are finely identified through a depth estimation and adaptive weight allocation image enhancement module.

7. The cigarette ash shrinkage rate calculation method according to claim 1, wherein The effective connected domains in the cigarette area and the ash area are extracted, including, Non-target regions in the background are removed using multi-scale feature fusion and connected domain analysis technology.

8. The method for calculating the ash shrinkage rate of cigarette packs as described in claim 7, characterized in that, Multi-scale feature fusion is used, including, Through a multi-scale feature fusion module, the features extracted by multi-scale convolution kernels are fused into a unified feature map, and the weights of each feature channel are adaptively adjusted by combining a Squeeze-and-Excitation module to highlight the ash column region features.

9. The method for calculating the ash shrinkage rate of cigarette packs as described in claim 1, characterized in that, Based on adaptive gating mechanism and convolutional network connected domain analysis, the removal of cracks is performed through edge detection and connected domain filling.

10. The method for calculating the ash shrinkage rate of cigarette packs as described in claim 9, characterized in that, The removal of cracks includes: Using a Sobel operator or Canny edge detection algorithm to extract crack edges; An adaptive gating module is embedded in a deep convolutional network to enhance crack features and suppress non-crack features; The connected domains of the crack region are extracted, and the feature regions are screened according to area and shape; The crack region is filled and color measurement is performed.

11. The method for calculating the ash shrinkage rate of cigarette packs as described in claim 1, characterized in that, The cigarette ash shrinkage rate calculation method further includes: The integrity of the cigarette area and the ash area is detected based on color channel analysis.

12. The method for calculating the ash shrinkage rate of cigarette packs as described in claim 1, characterized in that, The formula for calculating the cigarette ash shrinkage rate is: SR = (1-S1 / S0) × 100 In the formula, S0 is the area before the cigarette is burned, with a unit of square millimeters, and S1 is the area after the cigarette is burned, with a unit of square millimeters.

13. The method for calculating the ash shrinkage rate of cigarette packs as described in claim 1, characterized in that, The calculation result is output, including: The shrinkage rate calculation result is output and a shrinkage rate trend chart during the burning process is generated.

14. A cigarette pack ash shrinkage rate detection system, characterized in that, The system implements the method of any one of claims 1-13, including the following modules: An image acquisition module for acquiring multi-angle images before and after the cigarette is burned; An ash area segmentation module based on parallel convolutional neural network and adaptive gating mechanism for segmenting the ash area; A feature extraction and fusion module for multi-scale feature fusion of the ash area and removal of background noise; An area calculation and shrinkage rate detection module for calculating the shrinkage rate based on the shrinkage rate formula; A data analysis and output module for generating a shrinkage rate trend chart and a comprehensive detection report.

15. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the calculation method according to any one of claims 1 to 13.

16. A computer device comprising a memory and a processor, having stored on the memory a computer program capable of running on the processor, characterized in that, The processor executes the computer program to implement the steps of the calculation method according to any one of claims 1 to 13.

Images