Bag-type dust collector leakage detection method based on background difference method

Abstract

The invention relates to a bag-type dust collector leakage detection method based on a background differencing method, which comprises the following steps: firstly, acquiring a pattern card image in real time through an industrial camera, preprocessing through Gaussian filtering and ROI (Region of Interest) cutting, and then separating a static background from a potential leakage region by adopting a dynamic background modeling technology, so as to effectively overcome dust accumulation interference; calculating the weighted information entropy of each cloth bag region, and fusing the texture features and the motion features to enhance the leakage recognition capability; in the time sequence analysis stage, leakage continuity is evaluated through entropy change stability, an airflow space distribution mode is analyzed in combination with a vector convergence degree, the detection accuracy is remarkably improved, finally, a leakage judgment standard is dynamically adjusted by adopting a self-adaptive threshold mechanism, and the robustness of the system under different working conditions is ensured. According to the scheme, automation and intelligentization of the whole leakage detection process are achieved, the advantages of being high in anti-interference capacity, low in false alarm rate, wide in adaptability and the like are achieved, and the operation and maintenance efficiency of the bag-type dust collector can be remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of fault detection methods, specifically a leak detection method for bag filters based on the background difference method. Background Technology

[0002] A baghouse dust collector is a dry, high-efficiency dust removal device that filters dust-laden gas through fibrous fabric. Its core component is the filter bag (made of natural fibers, synthetic fibers, or glass fibers). When dust-laden gas enters the dust collector, large dust particles settle into the ash hopper due to gravity, while fine dust particles are intercepted, adsorbed, or electrostatically captured by the fibers on the surface of the filter bag. The purified gas escapes from the inside of the filter bag. The dust layer (initial layer) accumulated on the surface of the filter bag significantly improves filtration efficiency, but after long-term operation, it needs to be cleaned by pulse jet cleaning or mechanical vibration to maintain equipment performance. If the filter bag is burned through by high-temperature particles, punctured by mechanical friction, or corroded by chemical corrosion, dust will directly penetrate the filter bag, resulting in excessive emission concentrations; therefore, it is extremely necessary to perform filter bag fault detection on baghouse dust collectors. As a core piece of equipment for industrial waste gas treatment, the leakage detection of baghouse dust collectors directly affects the environmental compliance rate. Current technologies mostly use image processing and local information entropy-based detection methods to detect baghouse faults. However, this method has the following problems: relying on a fixed background or a single local information entropy makes it difficult to cope with interference from dust accumulation and changes in lighting in industrial scenarios, resulting in a high false detection rate; furthermore, it easily overlooks the temporal continuity and spatial pattern of leaks, making it susceptible to transient interference such as mechanical vibration; and fixed thresholds are a common defect of existing technologies, unable to adapt to changes in scene caused by dust concentration fluctuations or equipment aging, easily leading to poor detection results and a high false detection rate.

[0003] Therefore, this solution addresses the aforementioned issues by introducing the background difference method and combining it with multi-dimensional parameter analysis to construct a leak detection system with environmental adaptability. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a leak detection method for baghouse dust collectors based on the background difference method. By introducing the background difference method and combining it with multi-dimensional parameter analysis, a leak detection system with environmental adaptability is constructed. This system has advantages such as accurate judgment, minimal environmental influence, and strong adaptability. It solves the problems of existing detection methods that rely on fixed backgrounds or single local information entropy, making it difficult to cope with interference such as dust accumulation and light changes in industrial scenarios, resulting in high false detection rates. Furthermore, existing methods tend to ignore the temporal continuity and spatial pattern of leaks during detection, making them susceptible to transient interference such as mechanical vibration. At the same time, fixed thresholds are a common defect of existing technologies, which cannot adapt to changes in scene caused by dust concentration fluctuations or equipment aging, easily leading to poor detection results and high false detection rates.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for detecting leakage in a baghouse dust collector based on the background difference method, comprising the following steps: S1. Acquire raw image sequences of the bag filter tube sheet at a fixed frame rate. The original image sequence is then denoised to obtain the image sequence. ; S2, the denoised image sequence The images of individual fabric bags were obtained by cropping. ; S3, Image sequence The input is fed into a Gaussian mixture model, and the image sequence is processed according to a multimodal Gaussian distribution. Each pixel in To build a background model and derive an image sequence Overall probability density of the background model ; S4. Based on the current frame pixel value The foreground residual map is obtained by performing background subtraction with the corresponding background model parameters. and foreground mask and foreground mask Morphological optimization is performed to obtain the foreground mask. ; S5. Local images of each bag Calculate its local information entropy For residual plots Calculate its residual entropy Combined with local information entropy and residual entropy To set up dynamic weight adjustment The weighted entropy is obtained by weighted fusion of local information entropy and residual entropy. ; S6. Select the weighted entropy of the most recent N frames. The sequence is obtained by calculating the entropy change between adjacent frames to get the first-order difference sequence, and the entropy stability is obtained by calculating the reciprocal of the standard deviation of the first-order difference sequence. To measure the persistence of the leak; S7. Foreground Residual Map Calculate its gradient Spatial vector convergence of adjacent bags is used to assess the degree of convergence. ; S8. Based on weighted entropy Persistent leakage degree of convergence The three characteristics determine the degree of leakage. ; S9. Use a sliding window to calculate the probability of leakage in all fabric bags in the most recent W frames. Values ​​and calculate their mean. and its standard deviation And introduce regulatory factors This is to achieve dynamic threshold adjustment, based on the dynamic threshold to assess the likelihood of leakage. Determine if a leak exists. If so, mark the fault point and output the fault result. Otherwise, execute S1 again to continue the detection.

[0006] Furthermore, S3 employs a Gaussian mixture model to establish a probability distribution for each pixel, and its calculation formula is as follows: ; in, It is the current pixel value. The Gaussian distribution is used to describe the complexity of the background. These are the weights of the k-th distribution. and These are the mean and covariance of the Kth distribution. It is a Gaussian probability density function.

[0007] Furthermore, S4 also includes the following sub-steps: S4.1 The formula for calculating the residual plot is as follows: ; in The background model is in pixels The predicted value at that location, The current pixel value; S4.2 Thresholding is performed on the residual map to obtain the foreground mask. : ; in It is a binary foreground mask. These are the parameters of the best-fit Gaussian distribution, where 1 represents foreground and 0 represents background. Typical threshold; S4.3, Foreground Mask Morphological optimization involving erosion followed by expansion was performed to remove subtle noise. .

[0008] Furthermore, S5 also includes the following sub-steps: S5.1, Local images of each bag Calculate its local information entropy The formula is as follows: For each bag area The local information entropy is calculated using the following formula: ; Where pk is the grayscale value K in The probability of it; S5.2, Residual Plot Calculate its residual entropy The formula is as follows: ; Where pk is the grayscale value K in The probability of it; S5.3. Combining the original entropy and residual entropy, the weights are dynamically adjusted to obtain the weighted information entropy Hw formula as follows: ; in; ; That is, the higher the proportion of the prospect, the greater the weight of the residual.

[0009] Furthermore, S6 also includes the following sub-steps: S6.1 Select the weighted entropy sequence of the most recent N frames. ; in For reference frame count, The weighted information entropy of the i-th bag in frame k; S6.2 Calculate the entropy changes of k adjacent frames to obtain the first-order difference sequence. ; in This represents the change in entropy of the i-th bag in the k-th frame, where k is the frame index. arrive ,like Small fluctuations indicate stable entropy changes, which may indicate leakage. Large fluctuations indicate the presence of random noise interference. S6.3 Calculate the reciprocal of the standard deviation of the first-order difference sequence to measure leakage persistence. The calculation formula is as follows: ; in It is a difference sequence The standard deviation of entropy is used to measure the degree of fluctuation in entropy value. To prevent the denominator from being divided by zero, in order to find the minimum value, This represents the entropy stability of the i-th bag in frame t.

[0010] Furthermore, S7 includes the following sub-steps: S7.1, Foreground Residual Map Calculate its gradient , ; in For residual plot The gradient in the x-direction, It is a residual plot Gradient in the Y direction; S7.2 The formula for calculating the degree of convergence of spatial vectors is as follows: ; Where Ni is the set of adjacent bags, It is the angle between the gradient direction and the line connecting the centers of the two bags. This indicates that the gradient direction is consistent with the direction of the adjacent bags, indicating noise interference. This indicates that there is leaking airflow because the gradient direction is perpendicular to the adjacent filter bag. The lower the value, the more dispersed the gradient direction.

[0011] Furthermore, the formula for the potential leakage level in S8 is as follows: ; in This represents the degree of vector dispersion.

[0012] Furthermore, S9 includes the following sub-steps: S9.1 Calculate the probability of leakage within the sliding window. The mean is calculated using the following formula: ; S9.2 Calculate the probability of leakage within the sliding window. The standard deviation of is given by the following formula: ; When the possibility of leakage within the window The degree of fluctuation, when the dust concentration is high Increase; S9.3, Based on the degree of leakage possibility mean and its standard deviation And an adjustment factor is introduced to construct a dynamic threshold judgment formula: ; in For dynamic thresholds, As a dynamic adjustment factor; S9.4 The leakage determination criteria are as follows: ; Where f is the number of continuous leaked frames, to avoid misjudgment caused by transient noise.

[0013] Furthermore, the formula for calculating the dynamic adjustment factor k in S9.3 is as follows: in Based on sensitivity, It is the reference standard deviation.

[0014] Compared with the prior art, the technical solution of this application has the following beneficial effects: 1. This leak detection method for baghouse dust collectors based on background difference offers a significant improvement over traditional methods that rely on a fixed background or single local information entropy, which struggle to handle interference from dust accumulation and lighting changes in industrial settings, leading to high false detection rates. This solution innovatively introduces Gaussian mixture background modeling (MOG2), effectively separating static dust accumulation from dynamic leak airflow by dynamically updating the background model, thus reducing the false detection rate. Furthermore, a weighted information entropy fusion mechanism is proposed, combining local texture entropy and residual motion entropy to dynamically distinguish between leaks and dust adhesion, significantly improving the discriminative power of leak detection.

[0015] 2. This method for detecting leaks in baghouse dust collectors based on the background difference method addresses the common problem that existing technologies often neglect the temporal persistence and spatial patterns of leaks, making them susceptible to transient interference such as mechanical vibration. This innovative approach proposes an entropy variation stability index, identifying persistent leaks and filtering transient noise by analyzing the first-order difference stability (reciprocal of the standard deviation) of the weighted entropy sequence. Simultaneously, a vector convergence algorithm is designed, quantifying airflow divergence characteristics based on the spatial vector relationship between the residual gradient direction and adjacent bags. The combined judgment of these two methods can reduce false alarms due to vibration, further improving detection accuracy.

[0016] 3. The existing baghouse dust collector leakage detection method based on background difference relies on a fixed threshold, a common drawback of current technologies, which cannot adapt to changes in dust concentration or equipment aging. This solution innovatively employs a sliding window dynamic threshold mechanism. Based on the mean and standard deviation of the leakage probability in the most recent consecutive frames, the threshold is adjusted in real time, with the coefficient k automatically adjusting with dust concentration. This design ensures a low false alarm rate even when dust concentration suddenly increases. Furthermore, the introduction of a continuous frame confirmation condition further filters transient noise. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 This is a diagram illustrating the detection logic operation of the present invention; Figure 3 This is a line graph showing the accuracy test results of this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1-3 The leakage detection method for bag filters based on background difference method in this embodiment includes the following steps: S1. Acquire raw image sequences of the dust collector tube sheet at a fixed frame rate. The original image sequence is then denoised to obtain the image sequence. ; In actual testing, an industrial-grade CCD camera (such as Basler ace acA2000-50gm) can be used with a ring light to ensure clear imaging even under low light conditions. Industrial sites have problems such as dust and uneven lighting, resulting in a lot of noise in the original images. Noise reduction filtering is required. Gaussian filtering is preferred for noise reduction. Set the Gaussian filter kernel size to 5×5 and σ=1.5 to eliminate high-frequency noise.

[0020] S2, the denoised image sequence The images of individual fabric bags were obtained by cropping. The calibration board is used to pre-calibrate the flower board plane and establish a pixel coordinate mapping table for the bag area. During the cropping process, perspective transformation can be performed using OpenCV's cv2.getPerspectiveTransform to correct image distortion. Cropping the image can reduce the amount of computation by 80% and improve real-time performance.

[0021] S3, Image sequence The input is fed into a Gaussian mixture model, and the image sequence is processed according to a multimodal Gaussian distribution. Each pixel in To build a background model and derive an image sequence Overall probability density of the background model ; S4. Based on the current frame pixel value The foreground residual map is obtained by performing background subtraction with the corresponding background model parameters. and foreground mask and foreground mask Morphological optimization is performed to obtain the foreground mask. ; S4.1 The formula for calculating the residual plot is as follows: ; in The background model is in pixels The predicted value at that location, The current pixel value; S4.2 Thresholding is performed on the residual map to obtain the foreground mask. : ; in It is a binary foreground mask. These are the parameters of the best-fit Gaussian distribution, where 1 represents foreground and 0 represents background. Typical threshold; S4.3, Foreground Mask Morphological optimization involving erosion followed by expansion was performed to remove subtle noise. .

[0022] Leaking airflow can increase the complexity of the fabric bag's surface texture (such as eddies and grayscale fluctuations), while dust accumulation can also increase the entropy value, requiring differentiation based on motion information. In actual calculations, the fabric bag area can be divided into 8×8 sub-blocks to avoid global statistics masking local leaks. The entropy value is mapped to the [0,1] interval, facilitating weighted fusion with the residual entropy, thereby achieving the technical effect of reducing the misjudgment rate when dust accumulates.

[0023] S5. Local images of each bag Calculate its local information entropy For residual plots Calculate its residual entropy Combined with local information entropy and residual entropy To set up dynamic weight adjustment The weighted entropy is obtained by weighted fusion of local information entropy and residual entropy. ; S5.1, Local images of each bag Calculate its local information entropy The formula is as follows: For each bag area The local information entropy is calculated using the following formula: ; Where pk is the grayscale value K in The probability of it; S5.2, Residual Plot Calculate its residual entropy The formula is as follows: ; Where pk is the grayscale value K in The probability of it; S5.3. Combining the original entropy and residual entropy, the weights are dynamically adjusted to obtain the weighted information entropy Hw formula as follows: ; in ; That is, the higher the proportion of the prospect, the greater the weight of the residual.

[0024] In practical implementation, dust may accumulate on the surface of the filter bag, making direct analysis prone to misjudgment. Leakage manifests dynamically and needs to be extracted from the background. The above-mentioned technical solution, through MOG2 modeling, can adaptively update the background, classifying dust accumulation as background and avoiding false detection. Traditional static background methods cannot adapt to dust accumulation and have a high false detection rate. The residual map (foreground mask) directly reflects the airflow movement area and can suppress the interference of static dust. High local entropy but low residual entropy → dust; high local entropy and high residual entropy → leakage. Entropy values ​​are calculated only in the foreground area, reducing background interference. Weighted fusion can ensure the stability of the weighted entropy range, facilitating subsequent threshold determination. By dynamically adjusting the weights using α, it can adapt to different scenarios.

[0025] S6. Select the weighted entropy of the most recent N frames. The sequence is obtained by calculating the entropy change between adjacent frames to get the first-order difference sequence, and the entropy stability is obtained by calculating the reciprocal of the standard deviation of the first-order difference sequence. To measure the persistence of the leak; S6.1 Select the weighted entropy sequence of the most recent N frames. ; in For reference frame count, The weighted information entropy of the i-th bag in frame k; S6.2 Calculate the entropy changes of k adjacent frames to obtain the first-order difference sequence. ; in This represents the change in entropy of the i-th bag in the k-th frame, where k is the frame index. arrive ,like Small fluctuations indicate stable entropy changes, which may indicate leakage. Large fluctuations indicate the presence of random noise interference. S6.3 Calculate the reciprocal of the standard deviation of the first-order difference sequence to measure leakage persistence. The calculation formula is as follows: ; in It is a difference sequence The standard deviation of entropy is used to measure the degree of fluctuation in entropy value. To prevent the denominator from being divided by zero, in order to find the minimum value, This represents the entropy stability of the i-th bag in frame t.

[0026] In actual testing, baghouse dust collectors are subject to interference such as mechanical vibration and transient airflow disturbances. Relying solely on single-frame entropy data can easily lead to misjudgments. It is necessary to distinguish between continuous leakage (stable airflow, continuous entropy changes) and transient interference (abrupt entropy changes). Therefore, a first-order difference sequence can be used to first calculate the change in weighted entropy of the most recent 5 frames to capture the trend of entropy changes. Stability quantification: The reciprocal of the standard deviation of the difference sequence is taken. The smaller the standard deviation (the smoother the change), the higher the stability. In the above scheme, time dimension analysis is introduced to avoid transient noise interference. The persistence of leakage is quantified through statistical characteristics (standard deviation), replacing the traditional fixed frame threshold method. When a real leakage occurs, the entropy value continuously increases (such as stable airflow disturbance), and the stability is >0.3. When a transient disturbance occurs, the entropy value abruptly changes and then recovers quickly (such as mechanical vibration), and the stability is <0.1. Compared with the single-frame entropy threshold method, the false detection rate is reduced by 40%.

[0027] Preferably, an embedded GPU (such as NVIDIA Jetson AGX Xavier) can be used to compute the difference sequence in real time, and a circular buffer in RAM can be maintained to store the entropy values ​​of the most recent 5 frames.

[0028] S7. Foreground Residual Map Calculate its gradient Spatial vector convergence of adjacent bags is used to assess the degree of convergence. ; S7.1, Foreground Residual Map Calculate its gradient , ; in For residual plot The gradient in the x-direction, It is a residual plot The gradient in the y-direction; S7.2 The formula for calculating the degree of convergence of spatial vectors is as follows: ; Where Ni is the set of adjacent bags, It is the angle between the gradient direction and the line connecting the centers of the two bags. This indicates that the gradient direction is consistent with the direction of the adjacent bags, indicating noise interference. This indicates that there is leaking airflow because the gradient direction is perpendicular to the adjacent filter bag. The lower the value, the more dispersed the gradient direction.

[0029] Because leaked airflow exhibits a radial divergence pattern on the tube sheet surface, while the gradient direction caused by mechanical vibration or dust falling is random, it is necessary to distinguish between leakage and vibration through spatial characteristics. The gradient direction is calculated by using the Sobel operator to extract the gradient direction from the foreground residual map. Furthermore, the cosine of the angle between the gradient direction of adjacent bag regions and the center line is calculated through spatial vector convergence. This transforms the physical airflow pattern (divergence) into a mathematically quantifiable vector consistency index. Combined with the topological structure (bag arrangement), this enhances the spatial correlation analysis.

[0030] Preferably, Sobel gradient calculation can be implemented via FPGA (such as Xilinx Zynq UltraScale+), and the bag coordinate matrix can be pre-stored for real-time vector calculation.

[0031] S8. Based on weighted entropy Persistent leakage degree of convergence The three characteristics determine the degree of leakage. The formula for the potential leakage level in S8 is as follows: ; in This represents the degree of vector dispersion.

[0032] S9. Use a sliding window to calculate the probability of leakage in all fabric bags in the most recent W frames. Values ​​and calculate their mean. and its standard deviation And introduce regulatory factors This is to achieve dynamic threshold adjustment, based on the dynamic threshold to assess the likelihood of leakage. Perform a judgment and output the judgment result.

[0033] In S3, a Gaussian mixture model is used to establish a probability distribution for each pixel, and its calculation formula is as follows: ; in, It is the current pixel value. The Gaussian distribution is used to describe the complexity of the background. These are the weights of the k-th distribution. and These are the mean and covariance of the Kth distribution. It is a Gaussian probability density function.

[0034] S7 includes the following sub-steps: S9 includes the following sub-steps: S9.1 Calculate the probability of leakage within the sliding window. The mean is calculated using the following formula: ; S9.2 Calculate the probability of leakage within the sliding window. The standard deviation of is given by the following formula: ; When the possibility of leakage within the window The degree of fluctuation, when the dust concentration is high Increase; S9.3, Based on the degree of leakage possibility mean and its standard deviation And an adjustment factor is introduced to construct a dynamic threshold judgment formula: ; in For dynamic thresholds, As a dynamic adjustment factor; S9.4 The leakage determination criteria are as follows: ; Where f is the number of continuous leaked frames, to avoid misjudgment caused by transient noise.

[0035] The formula for calculating the dynamic adjustment factor k in S9.3 is as follows: ; in The baseline sensitivity set by the user. It is the reference standard deviation, obtained from historical experience.

[0036] Because the distribution range of characteristic values ​​varies greatly under different operating conditions, such as dust concentration and wind pressure, fixed thresholds are prone to failure. Therefore, the threshold needs to be adaptively adjusted according to the environment. The mean and standard deviation are calculated based on the P-value of the most recent 30 frames. By introducing a closed loop of operating condition feedback, such as dust concentration, and k-value, dynamic optimization of the threshold is achieved. Combined with continuous frame judgment (requiring 3 consecutive frames exceeding the threshold), transient noise is filtered out. Preferably, a dust sensor (such as a laser scattering PM2.5 sensor) can be used to input the k-value adjustment signal in real time. A DSP chip (such as TI TMS320C6748) can be used to quickly complete the sliding window statistics.

[0037] Please refer to Figure 3And as shown in Table 1 below, based on the existing local information entropy algorithm and this scheme, under conditions of high dust concentration, large changes in light intensity, and high on-site noise, 40 batches of bag filters were tested for fault detection, with 1000 frames per batch, to conduct fault accuracy experiments: Table 1 As can be clearly seen from the table above, this technical solution significantly improves the accuracy of determining leakage faults in baghouse dust collectors when the dust concentration is moderate, reaching an accuracy of approximately 95%. This method is more suitable for implementation under conditions of high dust concentration and complex operating conditions.

[0038] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0039] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for detecting leakage of a baghouse based on background subtraction method, characterized in that: Comprising the following steps: S1, collect the original image sequence of the cloth bag dust collector flower board at a fixed frame rate , and denoise the original image sequence to obtain an image sequence ; S2, on the denoised image sequence cropping to obtain the cropped local image of each bag ; S3, the image sequence is input into a mixture Gaussian model, and a background model is established according to a multi-modal Gaussian distribution of each pixel in the image sequence and an overall probability density of the background model of the image sequence is derived ; S4, according to the current frame pixel value and the corresponding background model parameters to carry out background difference to obtain the foreground residual graph and the foreground mask , and the foreground mask morphological optimization to obtain the foreground mask ; S5, for each cloth bag local image Calculate its local information entropy , for residual map Calculate its residual entropy , combined with local information entropy And residual entropy To set the dynamic adjustment weight , for local information entropy and residual entropy Weighted fusion to get weighted entropy ; S6, selecting the weighted entropy of the latest N frames The entropy value change of the adjacent frames is calculated to obtain a first-order difference sequence, and the reciprocal of the standard deviation of the first-order difference sequence is calculated to obtain the entropy change stability to measure the leakage persistence; S7, to the foreground residual map computing the gradient , to evaluate the degree of convergence of the spatial vector convergence of adjacent bags ; S8、According to the weighted entropy , leakage persistence , convergence degree , the three features get the leakage possibility degree ; S9. Use a sliding window to calculate the probability of leakage in all fabric bags in the most recent W frames. Values ​​and calculate their mean. and its standard deviation And introduce regulatory factors This is to achieve dynamic threshold adjustment, based on the dynamic threshold to assess the likelihood of leakage. Determine if a leak exists. If so, mark the fault point and output the fault result. Otherwise, execute S1 again to continue the detection.

2. A method for detecting a leak in a baghouse according to claim 1, wherein: The S3 adopts Gaussian mixture model to establish probability distribution for each pixel, and the calculation formula is as follows: ； where, is the current pixel value, is the number of Gaussian distributions used to describe the complexity of the background, is the weight of the kth distribution, and is the mean and covariance of the kth distribution, is the Gaussian probability density function.

3. A method for detecting a leak in a baghouse according to claim 1, wherein: The S4 further comprises the following sub-steps: S4.1, the formula for calculating the residual error map is as follows: ； wherein is a predicted value of the background model at pixel , is the current pixel value; S4.2, thresholding the residual map to obtain a foreground mask : ； wherein is a binary foreground mask, is the best matching Gaussian distribution parameter, where 1 is foreground and 0 is background, is a typical threshold value; S4.3, on the foreground mask A morphological opening with erosion followed by dilation is performed to remove small noise .

4. The method of claim 1, wherein the method further comprises: The S5 further comprises the following sub-steps: S5.1, for each cloth bag local image calculating its local information entropy , as follows: For each sack region , the local information entropy is calculated, as follows: ； where pkis the probability of the gray value K in the range [0, 255]. S5.2, on the residual map The residual entropy is calculated , as follows: ； where pkis the probability of the gray value K in the range [0, 255]. S5.3, the original entropy and residual entropy are combined to dynamically adjust the weight to obtain the weighted information entropy Hw, and the formula is as follows: ； Wherein ； That is, the higher the foreground proportion, the greater the residual weight.

5. A method for detecting a leak in a baghouse according to claim 1, wherein: The S6 further comprises the following sub-steps: S6.1, selecting the weighted entropy sequence of the last N frames, ； wherein is the number of reference frames, is the weighted information entropy of the ith bag at frame k. S6.2, calculating the entropy value change of k adjacent frames to obtain a first-order difference sequence, ； wherein represents the entropy value change of the i th cloth bag in the k th frame, k is the frame index, from to , if the fluctuation is small, it indicates that the entropy value changes stably, and then the leakage may occur, if the fluctuation is large, there is random noise interference; S6.3, the inverse of the standard deviation of the first difference sequence is computed to measure the leak persistence The formula is as follows: ； wherein is the standard deviation of the difference sequence for measuring the fluctuation degree of the entropy value change, to prevent the denominator from being zero, represents the entropy change stability of the ith cloth bag in the t frame.

6. A method for detecting a leak in a baghouse according to claim 1, wherein: The S7 comprises the following sub-steps: S7.1, to the foreground residual map computing the gradient , ； wherein is a residual map is a gradient in the x direction, is a residual map is a gradient in the Y direction; S7.2, the formula for calculating the spatial vector convergence degree is as follows: ； where Ni is the set of adjacent bags, is the angle between the gradient direction and the line connecting the centers of two bags, if , it means that the gradient direction is consistent with the direction of the adjacent bag, if , it means that the gradient direction is perpendicular to the adjacent bag, if , the smaller the value, the more dispersed the gradient direction.

7. A method for detecting a leak in a baghouse according to claim 1, wherein: The formula for calculating the leakage possibility in the S8 is as follows: ； wherein is the vector dispersion.

8. A method for detecting a leak in a baghouse according to claim 1, wherein: The S9 comprises the following sub-steps: S9.1, Calculate the degree of possible leakage within the sliding window The mean of the values of the formula as follows: ； S9.2, calculating the degree of possible leakage within the sliding window of the standard deviation, as follows: ； When the degree of leakage possibility of the window, the degree of fluctuation, the dust concentration is high increases; S9.3, according to the degree of possible leakage the mean value and its standard deviation and introduce a regulation factor to construct a dynamic threshold judgment formula: ； wherein is a dynamic threshold, is a dynamic adjustment factor; S9.4, the leakage judgment condition is as follows: ； Wherein f is the number of continuous frames of leakage, to avoid false judgment caused by transient noise.

9. A method of detecting a leak in a baghouse according to claim 8, wherein: The calculation formula of the dynamic adjustment factor k in the S9.3 is as follows: ； wherein is the base sensitivity, is the reference standard deviation.

Images