A monitoring method and system for hazardous waste treatment

Abstract

The present application belongs to the technical field of image processing, and particularly relates to a monitoring method and system for hazardous waste treatment, which method comprises: acquiring a continuous image sequence of a spraying system, and pre-processing images in the image sequence; analyzing the pre-processed images based on superpixel segmentation and density clustering algorithm to generate instantaneous cavity confidence, thereby constituting an instantaneous cavity confidence graph sequence corresponding to the image sequence; fusing the instantaneous cavity confidence graph sequence to calculate a time accumulation stability index for each pixel point representing the continuity of its cavity state in the time dimension, thereby generating a time accumulation stability graph; calculating a final cavity confidence based on the time accumulation stability graph, and judging the final cavity confidence based on a preset threshold to identify a spraying failure. The present application effectively suppresses instantaneous disturbance interference, significantly reduces the false positive rate, and realizes accurate identification of a real spraying failure.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing. More particularly, the present application relates to a monitoring method and system for hazardous waste treatment. BACKGROUND

[0002] In the hazardous waste incineration treatment process, the quench tower is the key equipment for controlling the re-synthesis of dioxin, a highly toxic substance. Its core function is to atomize water through high-pressure lances and uniformly spray it into high-temperature flue gas, causing the flue gas to rapidly cool down in a very short time, crossing the dangerous temperature zone of dioxin re-synthesis. Therefore, the atomization effect of the spraying system directly determines the environmental safety level of the entire incineration system. However, the traditional monitoring method mainly relies on temperature sensors at the tower outlet. This method measures the average temperature, which not only has significant hysteresis, but also cannot reflect the local spraying state abnormalities caused by problems such as lance blockage in real time, which may delay the discovery of the risk of excessive emission of pollutants.

[0003] Currently, existing technologies use online visual monitoring based on image processing techniques. For example, by analyzing a single frame of image, using a superpixel algorithm to divide the image into several small region blocks with similar features, then extracting the brightness and texture features of each region block, and finally identifying the sparse region blocks deviating from the normal dense fog field as voids to determine possible faults in the spraying system.

[0004] However, the static analysis based on single-frame images is too sensitive to the instantaneous airflow disturbance inside the quench tower, which easily leads to false positives. The high-speed turbulent flow of flue gas in the tower may temporarily blow away a small patch of fog field at a certain moment, forming a non-fault temporary pseudo-void. Since the existing method only analyzes the image frame in isolation, it cannot effectively distinguish between this instantaneous pseudo-void caused by airflow disturbance and the persistent real-time image caused by physical faults such as lance blockage in the time dimension. This lack of time information leads to insufficient algorithm reliability and easy false alarms. SUMMARY

[0005] To solve the technical problem of high false positive rate of the spraying diagnosis method based on single-frame images, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present application provides a monitoring method for hazardous waste treatment, comprising: acquiring a continuous image sequence of a spray system inside a quench tower in a hazardous waste treatment process, and pre-processing each frame image in the image sequence; analyzing each pre-processed frame image based on superpixel segmentation and density clustering algorithm, generating an instantaneous hollow confidence degree for each pixel point to represent its possibility of becoming a spray hole in the current frame, thereby constituting an instantaneous hollow confidence degree image sequence corresponding to the image sequence; fusing the instantaneous hollow confidence degree image sequence, calculating a time accumulation stability index for each pixel point to represent the continuity of its hollow state in the time dimension, thereby generating a time accumulation stability image; based on the time accumulation stability index of each pixel point and the time accumulation stability index of the pixel points in the spatial neighborhood thereof in the time accumulation stability image, calculating a final hollow confidence degree, and judging the final hollow confidence degree based on a preset threshold to identify a spray failure.

[0007] The present application effectively filters instantaneous disturbances through time accumulation analysis, and enhances the identification of real failure areas by combining spatial aggregation analysis, fundamentally solves the false alarm problem of single-frame static analysis, significantly improves the reliability and accuracy of diagnosis, and can realize accurate positioning of failure.

[0008] Preferably, the acquisition process of the instantaneous hollow confidence degree is: performing superpixel segmentation on each frame image to obtain a plurality of superpixel blocks; extracting a multi-dimensional feature vector including average brightness, brightness standard deviation and centroid coordinates for each superpixel block; using a density clustering algorithm to cluster the multi-dimensional feature vector, and determining the superpixel blocks not belonging to the core cluster as noise points; calculating the minimum feature distance of each noise point to all core clusters, and performing normalization processing on the minimum feature distance to obtain the instantaneous hollow confidence degree corresponding to the noise point.

[0009] By first segmenting the image into superpixel blocks, then extracting the multi-dimensional features and using the density clustering algorithm for analysis, the calculation is promoted from the pixel level to the region level, which not only reduces the computational complexity, but also smooths the noise by using regional features.

[0010] Preferably, the instantaneous hollow confidence degree satisfies the relationship: ; wherein, represents the instantaneous hollow confidence degree of pixel point o in the kth frame image, and pixel point o belongs to the superpixel block determined as a noise point , is the minimum feature distance of the ith superpixel block to all core clusters, and respectively represent the maximum and minimum values of the minimum feature distances corresponding to all noise points in the current frame.

[0011] The normalization eliminates the influence of distance fluctuation caused by changes in illumination, fog concentration and the like between frames of images, ensuring that the generated instantaneous void confidence value has consistent comparability in the entire image sequence.

[0012] Preferably, the time accumulation stability indicator satisfies the relationship: ; wherein, represents the time accumulation stability indicator of the pixel point o at the current time t, is the instantaneous void confidence of the pixel point o at the current time t, is the time accumulation stability indicator of the pixel point o at the previous time t-1, is a smoothing factor.

[0013] By adopting an exponentially weighted moving average, higher weight can be given to recent data, enabling the system to respond more quickly to new emerging and persistent fault signals, while also forgetting transient disturbances more quickly.

[0014] Preferably, the value of the smoothing factor is in the range [0.2, 0.4].

[0015] Preferably, the final void confidence satisfies the relationship: ; wherein, is the final void confidence of the pixel point o at the current time t, represents the time accumulation stability indicator of the pixel point o at the current time t, is the total number of pixels in a preset spatial neighborhood window centered on the pixel point o, and q is the index of the pixel point in the spatial neighborhood window, is a spatial enhancement coefficient.

[0016] By introducing spatial aggregation analysis, confidence enhancement can be performed on persistent suspected void regions with spatial aggregation characteristics, while stable abnormal points that are isolated and may be caused by camera noise and the like are suppressed, enabling the algorithm to focus more accurately on real void regions formed by physical reasons such as nozzle blockage, thereby improving the sensitivity and robustness of detection.

[0017] Preferably, the preset threshold includes a final confidence threshold, an area threshold and a duration threshold, and the final void confidence is judged based on the preset threshold to identify a sprinkler fault, including: marking the pixel point whose final void confidence is greater than the final confidence threshold as a void pixel point to form a void region; and if the area of the void region continuously exceeds the area threshold for a time reaching the duration threshold, determining that there is a dry sprinkler fault and issuing an alarm.

[0018] Preferably, the pre-processing of each frame of the image sequence comprises: adopting median filtering to perform denoising processing on each frame of image; and adopting a limited contrast self-adaptive histogram equalization algorithm to perform contrast enhancement on the denoised image.

[0019] In a second aspect, the present application provides a monitoring system for hazardous waste treatment, comprising a processor and a memory, wherein the memory stores computer program instructions which, when executed by the processor, implement the above-mentioned monitoring method for hazardous waste treatment.

[0020] By using the above technical solution, the above-mentioned monitoring method for hazardous waste treatment is generated into a computer program and stored in the memory to be loaded and executed by the processor, so that a terminal device is made according to the memory and the processor, and use is facilitated.

[0021] The present application can effectively filter out pseudo cavities caused by instantaneous air flow disturbance by introducing time cumulative stability analysis, and only responds to abnormal regions that exist continuously in the time dimension, thereby solving the problem that single-frame static analysis is easily disturbed and improving the accuracy and reliability of fault monitoring.

[0022] Further, by introducing spatial aggregation analysis, the confidence of suspected cavity regions with spatial aggregation characteristics and persistence can be enhanced, and isolated stable abnormal points can be suppressed, so that the algorithm can focus more accurately on real cavity regions formed by nozzle blockage and other physical reasons, and accurate positioning of faults is realized. BRIEF DESCRIPTION OF DRAWINGS

[0023] The above and other objects, features and advantages of the exemplary embodiments of the present application will be readily understood through reading the following detailed description, taken in conjunction with the accompanying drawings. In the drawings, several embodiments of the present application are illustrated by way of example and not limitation in which like reference numerals refer to like elements, in which:

[0024] Figure 1 is a flow chart schematically showing a monitoring method for hazardous waste treatment in the present application;

[0025] Figure 2 is a schematic diagram schematically showing an image containing real faults and instantaneous disturbances;

[0026] Figure 3 is a schematic diagram schematically showing a transient cavity confidence map;

[0027] Figure 4 is a schematic diagram schematically showing a time cumulative stability map;

[0028] Figure 5is a schematic diagram schematically showing a final hole confidence map. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0030] The specific embodiments of the present application will be described in detail below with reference to the drawings.

[0031] The embodiments of the present application disclose a monitoring method for hazardous waste treatment, referring to Figure 1 , comprising steps S1-S4:

[0032] S1, acquiring a continuous image sequence of a spraying system inside a quench tower in a hazardous waste treatment process, and pre-processing each frame of image in the image sequence.

[0033] In an optional embodiment, an observation window capable of fully covering all spray gun fog field areas inside the quench tower can be selected, and an industrial camera with a pneumatic cleaning device is installed to ensure that the lens remains clean in a high-temperature and high-humidity environment. The camera has a high dynamic range function to capture details of both the bright area of high-temperature flue gas and the dark area of dense water mist at the same time, continuously collects a video stream at a frame rate of 10 frames per second, and real-time intercepts an image sequence containing 10 continuous frames from the video stream as the continuous image sequence.

[0034] Further, due to the poor environment in the tower, salt and pepper noise caused by splashing large water droplets or smoke dust particles may exist in the original image, and the dark area details are submerged due to uneven overall illumination, so it is necessary to pre-process each frame of image in the continuous image sequence.

[0035] Specifically, first, each frame of image is denoised by median filtering, because median filtering can better protect the clarity of the fog field edge while removing impulse noise compared with Gaussian filtering; then a limited contrast adaptive histogram equalization algorithm can be used to enhance the contrast of the denoised image, by dividing the image into multiple small grid regions, histogram equalization is independently performed on each region, which can effectively enhance the local contrast between the fog field and the background, and make the fog field shape clearer.

[0036] As Figure 2The image diagram shown is a schematic diagram showing the image containing real faults and transient disturbances, and it can be seen that when the system captures this frame of image, there is a real fault in the upper left corner area caused by the blockage of the spray gun and a temporary false cavity in the lower right corner area caused by a transient air flow disturbance.

[0037] In this way, through the acquisition and preprocessing of the image sequence, stable and clear visual data can be obtained, providing high-quality data support for subsequent feature analysis and fault diagnosis.

[0038] S2, based on superpixel segmentation and density clustering algorithm, analyze each preprocessed frame of image, generate a transient cavity confidence for each pixel point representing its possibility of becoming a spray cavity in the current frame, and thus form a transient cavity confidence map sequence corresponding to the image sequence.

[0039] In an optional embodiment, a simple linear iterative clustering algorithm can be used to segment each frame of image into N superpixel blocks, and then a multi-dimensional feature vector including but not limited to average brightness, brightness standard deviation and centroid coordinates can be extracted for each superpixel block. The average brightness is used to reflect the density of the fog, the brightness standard deviation is used to reflect the texture complexity, and the centroid coordinates are used for spatial positioning.

[0040] Further, the feature vector set of all superpixel blocks in each frame of image is input into the DBScan density clustering algorithm. After the algorithm runs, most of the superpixel blocks representing normal dense fog field will be identified as one or more core clusters; the superpixel blocks not belonging to the core clusters are determined as noise points, and the outlying degree of each noise point is calculated, i.e. the minimum feature distance of each noise point to all core clusters is calculated The greater the distance, the farther the feature of the superpixel block deviates from the normal fog field.

[0041] In this optional embodiment, according to the obtained outlying degree, a transient cavity confidence for each pixel point in each frame of image can be generated to represent its possibility of becoming a spray cavity in the current frame. If the superpixel block to which the pixel point o belongs is a core cluster member, the corresponding transient cavity confidence is 0; if the superpixel block to which the pixel point o belongs is a noise point, its confidence can be obtained by the normalized minimum feature distance, and the transient cavity confidence satisfies the relationship:

[0042]

[0043] wherein, represents the transient cavity confidence of pixel point o in the k-th frame of image, and pixel point o belongs to the superpixel block determined as a noise point , is the minimum feature distance of the i-th superpixel block to all core clusters, and respectively represent the maximum and minimum of the minimum feature distances corresponding to all noise points of the current frame.

[0044] In this optional embodiment, an instantaneous hole confidence map is generated for each frame of image, thereby forming an instantaneous hole confidence map sequence corresponding to the image sequence. As shown in Figure 3 , which is a schematic diagram schematically showing the instantaneous hole confidence map, it can be seen that the real fault in the upper left corner and the instantaneous disturbance in the lower right corner are both marked out, and in the case of single-frame analysis, they are both regarded as highly suspicious holes, causing simultaneous identification and alarm of the two regions, while the alarm for the lower right corner region is a false alarm.

[0045] In this way, through static analysis of single-frame images, all suspected hole regions can be quickly identified, providing basic data for subsequent time sequence analysis.

[0046] S3, fuse the instantaneous hole confidence map sequence, calculate the time accumulation stability index of each pixel point representing the continuity of its hole state in the time dimension, thereby generating a time accumulation stability map.

[0047] In an optional embodiment, the holes caused by real physical faults such as nozzle blockage are continuously present in time, and their instantaneous hole confidence values remain at a high level for multiple frames; while the pseudo-holes caused by air flow disturbance are transient, and their instantaneous hole confidence values fluctuate sharply, so in order to effectively distinguish them, an exponential weighted moving average can be used to calculate the time accumulation stability of pixel point o at the current time t. The time accumulation stability index satisfies the relationship:

[0048]

[0049] wherein, represents the time accumulation stability index of pixel point o at the current time t, represents the instantaneous hole confidence of pixel point o at the current time t, represents the time accumulation stability index of pixel point o at the previous time t-1, is a smoothing factor, and its value range is [0.2, 0.4].

[0050] Specifically, the smoothing factor determines the decay rate of historical data, and in actual application, it can be set according to the requirements of the system on response speed and stability, and in this scheme, the value is 0.3. For scenes requiring fast response, the value can be taken as 0.4; for scenes requiring high stability and filtering out more fluctuations, the value can be taken as 0.2. ​​By limiting the range to a narrow range of [0.2, 0.4], a better balance can be struck between the sensitivity to real faults and the stability of filtering transient disturbances in practical applications.

[0051] For example, assuming the initial cumulative temporal stability of all pixels is 0, for pixel A representing a real hole, its five-frame instantaneous hole confidence map sequence is 0.9, 0.9, 0.9, 0.9, 0.9, with corresponding cumulative temporal stability updates to 0.27, 0.459, 0.591, 0.684, 0.749, continuously increasing and stabilizing at a high level. For pixel B representing a transient disturbance, its instantaneous hole confidence map sequence is 0.1, 0.9, 0.1, 0.1, 0.1, 0.1, with corresponding cumulative temporal stability updates to 0.03, 0.291, 0.234, 0.194, 0.165, showing one jump but quickly falling back to a low level. Therefore, the cumulative temporal stability index can successfully distinguish between persistent anomalies and transient anomalies.

[0052] like Figure 4 The diagram shown is a schematic representation of the time-cumulative stability map. It can be seen that the real fault region in the upper left corner has very high cumulative stability because it persists throughout the entire image sequence; while the transient disturbance region in the lower right corner only appears in the last few frames and does not have enough time to accumulate, so the colors of the real fault region and the transient disturbance region begin to show obvious differences.

[0053] Thus, by accumulating the instantaneous void confidence over time, it is possible to effectively filter out false void signals caused by random airflow disturbances, while retaining and enhancing persistent void signals caused by real physical faults.

[0054] S4. Based on the time-cumulative stability index of each pixel in the time-cumulative stability map and the time-cumulative stability index of pixels in its spatial neighborhood, the final hole confidence is calculated, and the final hole confidence is judged based on a preset threshold to identify sprinkler failures.

[0055] In an optional embodiment, since a real physical void is not only continuous in time but also spatially a clustered region with a certain range, this scheme introduces spatial neighborhood information to weight and enhance the temporal cumulative stability to obtain the final void confidence score, in order to improve the identification of the real void region. The final void confidence score satisfies the following relationship:

[0056]

[0057] in, Let be the final hole confidence score of pixel o at the current time t. a time accumulation stability index of the pixel point o at the current time t, is the total number of pixels in the preset spatial neighborhood window centered on the pixel point o, and for example, the preset spatial neighborhood window is a 5*5 neighborhood window, that is, =25;q is a pixel point index in the spatial neighborhood window, is a spatial enhancement coefficient, and for example, the value of the spatial enhancement coefficient in the present scheme is 1.5. The relationship formula uses the average stability of the neighborhood of the pixel o to nonlinearly amplify the stability of the pixel o itself.

[0058] In the optional embodiment, the confidence of the final aggregated hollow region is significantly increased, and the isolated stable noise points are effectively suppressed, and the final effect is as shown in Figure 5 The brightness of the real fault region in the upper left corner is greatly enhanced again, and the brightness value is much higher than that of other regions, and there is a great difference from the dim disturbance region in the lower right corner.

[0059] Further, the final hollow confidence can be judged by a preset threshold to identify the spray failure, wherein the preset threshold includes a final confidence threshold, an area threshold and a duration threshold, and for example, the final confidence threshold is 1.5, the area threshold is 1000 pixels, and the duration threshold is 3 seconds. The purpose of setting the three thresholds for combined judgment is to not only require that the abnormal signal be strong enough, but also require that it form a meaningful region in space and stably exist in time, so that isolated interference points can be greatly excluded, and only real and persistent spray abnormalities can trigger an alarm, thereby greatly reducing the false alarm rate and improving the reliability of the monitoring system.

[0060] Specifically, all pixel points with a final hollow confidence greater than the final confidence threshold are marked as hollow pixel points to form a hollow region, the field of view of the camera is pre-divided into a plurality of monitoring regions (ROIs) corresponding to the spray guns, and the hollow area in each ROI is counted, and if the area of the hollow region in a certain ROI continuously exceeds the area threshold for a time reaching the duration threshold, it can be determined that the spray gun has an empty spray failure and an alarm information is sent.

[0061] In this way, by combining the spatial context information to enhance the stability of the time accumulation, the real aggregated hollow and the isolated stable noise points can be further distinguished, and combined with the multiple threshold judgment, the accurate diagnosis and positioning of the spray failure can be finally realized.

[0062] The embodiment of the present application also discloses a monitoring system for hazardous waste treatment, comprising a processor and a memory, and the memory stores computer program instructions, which realize a monitoring method for hazardous waste treatment according to the present application when executed by the processor.

[0063] The system also includes other components known to those skilled in the art such as a communication bus and communication interfaces, the arrangement and function of which are known in the art and thus will not be described herein.

[0064] In the description of the present specification, the meaning of "a plurality of" or "several" is at least two, for example, two, three or more, and the like, unless explicitly specifically defined otherwise.

[0065] While the present specification has shown and described a number of embodiments of the application, it is to be understood that such embodiments are merely illustrative of the many possible embodiments thereof. Numerous modifications, adaptations, and variations will be apparent to those skilled in the art in view of the foregoing description. It is to be understood that, in the course of practicing the present application, embodiments of the application described herein can be adopted to employ each of the alternatives, modifications and variations.

Claims

1. A monitoring method for hazardous waste treatment, characterized by, The method comprises the following steps: acquiring a continuous image sequence of a spraying system inside a quench tower in a hazardous waste treatment process, and pre-processing each frame of the image sequence; based on superpixel segmentation and density clustering algorithm, analyzing each pre-processed frame of the image, generating an instantaneous hollow confidence degree of each pixel point, which represents the possibility of the pixel point becoming a hollow in the current frame, thereby forming an instantaneous hollow confidence degree image sequence corresponding to the image sequence; the process of obtaining the instantaneous hollow confidence degree comprises: performing superpixel segmentation on each frame of the image to obtain a plurality of superpixel blocks; extracting a multi-dimensional feature vector comprising average brightness, brightness standard deviation and centroid coordinates for each superpixel block; using a density clustering algorithm to cluster the multi-dimensional feature vector, and determining the superpixel blocks not belonging to the core cluster as noise points; calculating the minimum feature distance of each noise point to all core clusters, and performing normalization processing on the minimum feature distance to obtain the instantaneous hollow confidence degree corresponding to the noise point; the instantaneous hollow confidence degree satisfies the relationship: wherein, represents the instantaneous hole confidence of the pixel point o in the k-th frame image, the pixel point o belongs to the superpixel block judged as the noise point , is the minimum feature distance of the i-th superpixel block to all core clusters, and respectively represent the maximum value and the minimum value of the minimum feature distances corresponding to all noise points of the current frame. fusing the instantaneous hollow confidence degree image sequence, calculating a time accumulation stability index of each pixel point, which represents the continuity of the hollow state of the pixel point in the time dimension, thereby generating a time accumulation stability image; the time accumulation stability index satisfies the relationship: wherein, represents a time cumulative stability index of the pixel point o at the current time t, represents an instantaneous hole confidence of the pixel point o at the current time t, represents a time cumulative stability index of the pixel point o at the previous time t-1, is a smoothing factor; based on the time accumulation stability index of each pixel point in the time accumulation stability image and the time accumulation stability index of the pixel points in the spatial neighborhood of the pixel point, calculating a final hollow confidence degree, and judging the final hollow confidence degree based on a preset threshold to identify a spraying failure; the final hollow confidence degree satisfies the relationship: wherein, is the final hole confidence of the pixel point o at the current time t, denotes the time accumulation stability index of the pixel point o at the current time t, is the total number of pixels in the preset spatial neighborhood window centered on the pixel point o, and q is the pixel point index in the spatial neighborhood window, is the spatial enhancement coefficient.

2. A monitoring method for hazardous waste treatment according to claim 1, characterized in that, The value range of the smoothing factor is [0.2, 0.4].

3. The monitoring method for hazardous waste treatment according to claim 1, wherein, the preset threshold comprises a final confidence degree threshold, an area threshold and a duration threshold, and the judgment of the final hollow confidence degree based on the preset threshold to identify the spraying failure comprises: marking the pixel points with a final hollow confidence degree greater than the final confidence degree threshold as hollow pixel points to form a hollow region; if the area of the hollow region continuously exceeds the area threshold for a duration reaching a duration threshold, determining that it is a hollow spraying failure and issuing an alarm.

4. The monitoring method for hazardous waste treatment according to claim 1, wherein the pre-processing of each frame of the image sequence comprises: performing denoising processing on each frame of the image using median filtering; performing contrast enhancement on the denoised image using a limited contrast adaptive histogram equalization algorithm.

5. A monitoring system for hazardous waste treatment, characterized by The method comprises the following steps: a processor and a memory, the memory storing computer program instructions, when the computer program instructions are executed by the processor, realizing a monitoring method for hazardous waste treatment according to any one of claims 1-4.

Images