Remote operation and maintenance management method and system of commercial kitchen fume cleaning and disinfecting equipment

Abstract

This invention relates to the field of remote operation and maintenance management technology, specifically to a remote operation and maintenance management method and system for commercial kitchen fume cleaning and disinfection equipment. The specific steps are as follows: acquiring equipment execution data during fume cleaning and disinfection, and controlling a liftable camera to extend to the observation position for panoramic image acquisition, obtaining image data of the inner wall of the flue; dividing the panoramic field of view of the inner wall of the flue at the observation position into a set of spatial partitions covering the inner wall of the flue; determining the visibility statistics of each partition, reflecting image clarity and imaging effectiveness, and judging whether a partition is a visible partition based on the visibility statistics; for visible partitions, determining the cleaning effect based on image data and performing anomaly judgment on the spraying and washing units associated with the partition; for invisible partitions, performing anomaly judgment on the spraying and washing units based on execution data and image data. This solves the technical problem in existing technologies where uneven image evidence due to differences in flue viewing distance leads to unstable remote cleaning acceptance conclusions.

Description

Technical Field

[0001] This invention relates to the field of remote operation and maintenance management technology, and in particular to a remote operation and maintenance management method and system for commercial kitchen fume cleaning and disinfection equipment. Background Technology

[0002] Existing commercial kitchen fume cleaning and disinfection equipment is usually used in conjunction with a remote operation and maintenance platform. The remote end can view and manage the equipment's operating status and remotely issue cleaning / disinfection tasks when needed. After receiving control commands, the equipment drives the supply of cleaning media, the switching of pipelines and valves, and the operation of spraying actuators (such as rotating nozzles) according to the preset operation process, so that the cleaning media covers and washes the inner wall of the flue. At the same time, high-temperature steam and other methods can be used to carry out disinfection treatment inside the flue, and key operating data, alarm information and process records during the operation are transmitted back to the platform for record keeping and subsequent operation and maintenance management.

[0003] For example, in their research paper "An Automatic Cleaning and Fire Prevention System for Catering Industry Oil Ducts" published in the August 2013 issue of the Journal of Shaanxi University of Technology (Natural Science Edition), Fu Mingxing and Zhao Feng disclosed a scheme that uses a spray system for both fire prevention and extinguishing and duct cleaning. When duct cleaning is required, the valve is opened and the pressurized pump is started, so that the hot water from the solar collector is mixed with the washing liquid in the siphon pipe and sprayed onto the four walls of the duct. The oil stains are cleaned under the dual action of the high-pressure water jet impact and the hot washing liquid. The paper also clarifies that the cleaning process can be carried out separately for the vertical and horizontal sections according to the duct structure. The vertical section can be cleaned by adjusting the spray angle of the sprayer and using multiple layers of spray holes at different angles to expand the water curtain contact range and reduce omissions or dead angles. The horizontal section can be cleaned by setting up short sections of duct that are easy to disassemble or replace parts of the structure. Meanwhile, the prior art also describes the mixing and self-priming mechanism of the washing liquid and hot water (a siphon self-priming structure based on a pressure difference formed by a variable diameter pipe) to support the supply and spraying of the cleaning medium.

[0004] It is evident that existing technologies focus on the implementation of the "cleaning execution side," such as cleaning medium supply, spray coverage, and flue structure adaptation. However, they do not provide a feasible mechanism for "acceptance judgment and anomaly location" in remote operation and maintenance scenarios, which involves the collaboration of image evidence and execution data.

[0005] In remote operation and maintenance scenarios, in order to determine whether cleaning is needed, whether the cleaning meets the standards, whether the spraying coverage is adequate, and whether the nozzles are abnormal without personnel entering the flue, a liftable camera is often configured to obtain image evidence inside the flue before and after cleaning. However, due to the long and narrow space of the flue, the large variation in viewing distance, and the presence of bends that obstruct the view, coupled with factors such as the adhesion of steam / water mist / oil droplets during cleaning, the images captured by the camera often show a clear view up close, but are unclear at a distance due to distance and obstruction. This leads to uncertainty in the image-based judgment of the cleaning effect (whether it is clean, whether there are localized stubborn stains) and the location of nozzle coverage abnormalities, which in turn affects the reliability of the remote acceptance conclusion and the accurate identification of nozzle abnormalities. Summary of the Invention

[0006] To address the technical problem of uneven image evidence due to differences in the viewing distance of the flue, leading to unstable remote cleaning acceptance conclusions in existing technologies, this invention provides a remote operation and maintenance management method and system for commercial kitchen fume cleaning and disinfection equipment. The technical solution is as follows: On the one hand, a remote operation and maintenance management method for commercial kitchen fume cleaning and disinfection equipment is provided, which includes: Step 1: Acquire equipment execution data during fume cleaning and disinfection, and control the liftable camera to extend to the observation position for panoramic acquisition, obtaining image data of the flue's inner wall. Divide the panoramic field of view of the flue's inner wall at the observation position into a set of spatial partitions covering the flue's inner wall. Step 2: Determine the visibility statistics of each partition, reflecting image clarity and imaging effectiveness, and determine whether a partition is a visible partition based on the visibility statistics. For visible partitions, determine the cleaning effect based on the image data and perform anomaly judgment on the associated spraying units of the partition. For invisible partitions, perform anomaly judgment on the spraying units based on the execution data and image data. Step 3: Based on the visibility statistics of each partition, fuse the anomaly judgment results of the visible partitions and the anomaly judgment results of the invisible partitions to obtain the cleaning acceptance conclusion of the entire flue.

[0007] On the other hand, a remote operation and maintenance management system for commercial kitchen fume cleaning and disinfection equipment is provided. This system applies methods for the remote operation and maintenance management of commercial kitchen fume cleaning and disinfection equipment. The system includes: The surround-view acquisition module is used to acquire equipment execution data during fume cleaning and disinfection, and control the liftable camera to extend to the observation position for surround-view acquisition, obtaining image data of the inner wall of the flue. The surround-view field of view of the inner wall of the flue at the observation position is divided into a set of spatial partitions covering the inner wall of the flue. The visual gating judgment module is used to determine the visibility statistics of each partition, reflecting the image clarity and imaging effectiveness, and to determine whether a partition is a visible partition based on the visibility statistics. For visible partitions, the cleaning effect is determined based on the image data, and anomaly judgment is performed on the spraying action unit associated with the partition. For invisible partitions, anomaly judgment is performed on the spraying action unit based on the execution data and image data. The weighted fusion acceptance module is used to fuse the anomaly judgment results of visible partitions and invisible partitions based on the visibility statistics of each partition to obtain the cleaning acceptance conclusion of the entire flue.

[0008] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: (1) By combining visibility gating and partitioned evidence paths, the same process can still be closed-loop accepted even in scenarios with unstable imaging conditions, avoiding the single-point dependence of existing technologies on clear images. Specifically, this solution first establishes a set of partitions within the field of view of the observation position, and calculates the visibility statistics for each partition to determine whether the partition has the imaging conditions to support the judgment of the cleaning effect; for visible partitions, the image residue characterization is used to complete the comparison before and after cleaning and output the cleaning effect conclusion; for invisible partitions, the execution data-driven judgment path is switched and supplemented with weak prompt correction. Unlike the existing technology that "uses images to judge the cleaning effect uniformly, and if the image is unusable, it cannot be accepted or is misjudged", this solution explicitly models image quality problems as two types of evidence conditions: visible and invisible, and provides corresponding judgment mechanisms at the process level, so that image unreadable caused by long-distance blurring, steam and water mist obstruction, reflection, etc., no longer directly causes acceptance interruption or nozzle misposition, thereby better solving the problem of unusable acceptance and misassignment caused by unstable shooting conditions during the cleaning stage.

[0009] (2) By combining the binding of partitions and spray washing units with the feedback evidence summary of abnormal results, the abnormality location is improved from the entire section of uncleanliness to the location of the coverage direction or specific spray washing unit, and the handling strategy is transformed from experience-based dispatch to verifiable dispatch. Specifically, the scheme solidifies the partition identifier and establishes the correspondence between the partition and the spray washing unit identifier during the partition generation stage. Subsequently, regardless of whether the image judgment of the visible partition or the execution side judgment of the invisible partition is used, the output is fed back to the corresponding spray washing unit with the partition as the carrier, and the partition image evidence summary or trigger basis summary is generated simultaneously. Compared with the existing technology, which mostly outputs the cleaning failure as a global image or global alarm and is difficult to map to the specific nozzle / coverage direction, this scheme binds the partition result, spray washing unit and evidence summary into an integrated whole, so that the abnormality location has directionality and objectivity, which can better solve the problems of inadequate spray washing coverage, local stubborn residue and difficulty in locating abnormal spray washing units, while reducing the cost of repeated cleaning and invalid dispatch.

[0010] (3) By combining the event-following consistency hit rate and the execution completeness judgment, the abnormal judgment of the invisible partition is realized. The execution is upgraded from being considered normal as long as there is data to the physical consistency verification of whether the action-response is self-consistent, which significantly reduces the false judgment and false omission of relying solely on threshold alarms. Specifically, in the invisible partition, this solution no longer takes whether the monitored quantities such as pressure and flow rate fall within the range as the sole basis. Instead, it transforms the following relationship between valve opening events and flow response into a consistency index, and generates execution-side risks together with the completeness index of process completion. When valve jamming, blockage, condensation accumulation, or insufficient supply causes the valve to open but the flow rate does not increase or the response is discontinuous, even if some instantaneous values ​​are still within the threshold range, the consistency hit rate will drop significantly, thereby triggering an anomaly. Compared with the single variable threshold or simple timing rules commonly used in the prior art, this solution establishes cross-variable consistency judgment with control events as anchor points, which can better solve the problems of substandard cleaning effect and mis-assignment caused by similar execution data, difficulty in distinguishing alarms, and inaccurate anomaly location.

[0011] (4) By combining weak prompts with conservative tightening, capping the score for invisible areas, and weighted fusion based on visibility, the safety-side decision-making for flue-level acceptance is achieved under conditions of uneven evidence strength, avoiding the overly optimistic approval or overly pessimistic rejection of existing technologies in invisible areas. Specifically, this scheme extracts directional weak prompt features such as fogging, brightness fluctuations, and dynamic responses from images for invisible partitions, and limits the risk of weak prompts to tightening only when the execution side has not triggered anomalies, and does not allow the reduction of conclusions for triggered anomalies; at the same time, in flue-level fusion, the partition fusion weight is generated based on the visibility statistics, so that the partitions with more reliable image evidence contribute more, and a capping constraint is imposed on the score of invisible partitions to limit their support for the pass conclusion. Compared to existing technologies that either ignore the unvisible, leading to increased release risks, or fail to accept the entire segment due to partial unvisibility, this solution forms a synergy between the partitioning layer, risk layer, and fusion layer: weak prompts are responsible for reducing missed judgments, capping constraints are responsible for suppressing optimistic biases caused by the unvisible, and visibility weighting is responsible for letting reliable evidence dominate the conclusion, thereby better solving the problem of acceptance reliability and safe release boundary under unstable images during the cleaning stage. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of the remote operation and maintenance management method for commercial kitchen fume cleaning and disinfection equipment provided in this embodiment of the invention; Figure 2 This is a schematic diagram of the remote operation and maintenance management system for commercial kitchen fume cleaning and disinfection equipment provided in this embodiment of the invention; Figure 3 This is a flowchart of the cleaning effect and spray washing unit anomaly determination based on the partition visibility statistics gating provided in this embodiment of the invention; Figure 4 This is a schematic diagram of the structure of a commercial kitchen fume duct steam cleaning and purification recovery system provided in an embodiment of the present invention. Detailed Implementation

[0014] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0015] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0016] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0017] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] In a specific embodiment of the present invention, a remote operation and maintenance management method for commercial kitchen fume cleaning and disinfection equipment is provided, such as... Figure 1 The flowchart shown is for the remote operation and maintenance management of commercial kitchen fume cleaning and disinfection equipment. The process of this method may include the following steps: Step 1: Obtain the equipment execution data during oil fume cleaning and disinfection, and control the liftable camera to extend to the observation position to collect panoramic images of the flue wall. Divide the panoramic field of view of the flue wall at the observation position into a set of spatial partitions covering the flue wall.

[0020] This embodiment provides a remote operation and maintenance management method for commercial kitchen fume cleaning and disinfection equipment. The equipment includes a flue, a steam generation and delivery component, one or more spray cleaning units, a liftable camera, an operation monitoring and control component, and a remote operation and maintenance platform. The operation monitoring and control component collects operational data during task execution, including pressure, temperature, flow rate, valve opening / closing or opening degree, nozzle rotation speed or motor speed, motor current, camera lifting position, and task start and end times, forming execution data. The liftable camera is used to extend into the flue and collect images during critical moments of the task. The steam generation and delivery component generates the cleaning medium and delivers it to the spray cleaning units via a network of pipes and valves. The component includes a steam generator and steam delivery pipelines, which include a main pipeline, branch pipelines, and regulating valves, providing the spray cleaning units with the required medium pressure and flow rate during the task. The spraying unit is located inside the flue and is used to spray and scour the inner wall of the flue under steam pressure. The spraying unit includes at least a nozzle, a nozzle rotation drive, and a medium channel connected to the nozzle. The nozzle rotation drive is used to drive the nozzle to rotate around a preset axis to form a circumferential coverage. The medium channel controls the spraying enable state through the opening and closing or opening degree of the valve, thereby forming a controllable spraying coverage direction and coverage intensity within the task window.

[0021] like Figure 4 As shown in the schematic diagram of the commercial kitchen fume duct steam cleaning and purification recovery system provided in this embodiment of the invention, the fumes generated by the stove enter the flue under the suction of the exhaust fan and are transported along a predetermined flow direction. Purification units are arranged along the flue, and the fumes undergo particulate oil mist capture and deep purification as they pass through these units: High-voltage electric field purification charges the oil mist particles and captures them under the action of the electric field; wet scrubbers wash and absorb residual oil mist and soluble components through a water film or spray, further reducing emissions. The purified gas is discharged by the exhaust fan. When cleaning is required, the wall-mounted steam cleaning equipment generates high-temperature steam and delivers it through pipelines to the cleaning nozzles inside the flue. The cleaning nozzles spray and flush the inner wall of the flue, softening and peeling off the attached oil stains, which then form oily wastewater with condensate and flushing liquid. The oily wastewater is collected in an oil stain recovery tank or wastewater recovery device through a wastewater recovery pipeline. During the cleaning process, the visual lifting camera extends to the observation position at critical moments of the task to collect images around the room. Combined with operational monitoring data such as temperature sensors, it provides evidence for pre-cleaning assessment, in-cleaning coverage inspection, and post-cleaning acceptance, supporting remote operation and maintenance dispatch, review, and audit record keeping.

[0022] The remote operation and maintenance platform generates a task number upon task initiation and continuously receives operational monitoring data and control feedback uploaded from the equipment during task execution, forming an execution data set. This execution data includes at least: steam pressure sequence, steam temperature sequence, steam flow sequence, valve status sequence, nozzle speed or motor speed sequence, motor current sequence, camera position sequence, task start and end time windows, and spraying duration. This execution data originates from the equipment's existing sensor and controller interfaces and is acquired in real-time.

[0023] During non-shooting phases, the camera is in a retracted position to reduce long-term oil buildup. When the evidence collection phase begins, the platform issues a downward command, and the camera descends to at least one observation position, acquiring a sequence of panoramic images. Observation positions are identified by numbers k (k=1, 2, ..., K), where K is the total number of observation positions. Each observation position corresponds to a stable stopping position, confirmed by feedback from the lifting mechanism's encoder or limit switch. The j-th frame of the original image obtained at observation position k is denoted as I. k，j (j=1, 2, ..., F), the frame rate and frame count F are configured by the platform or the device, and the source is a fixed configuration. Surround view can be achieved by directly covering the area with a wide-angle / panoramic lens, or by rotating a camera or gimbal around the entire screen; either method can be chosen for engineering implementation.

[0024] The field of view of the flue's inner wall from the observation position is divided into a set of spatial partitions covering the flue's inner wall. In this embodiment, the field of view is divided into N non-overlapping spatial partitions, with partition numbers represented by n (n=1, 2, ..., N). The partitioning process can be implemented by dividing the angle domain and constraining the spatial domain boundaries.

[0025] Angle domain segmentation refers to dividing the 360° viewing angle into P angle sectors according to preset rules, with the set of dividing angles denoted as {θ0, θ1, ..., θ...}. P}, where θ0=0, θ P =360. The number and value of the dividing angles are fixed by the configuration. Spatial domain boundary constraints determine the boundary of the effective inner wall region in the image, excluding lens distortion edges, occluding component regions, and non-inner wall regions. This boundary can be obtained and fixed through one-time calibration during the installation and debugging phase, and its source is obtained during installation and debugging. Based on the dividing angles and boundary constraints, a partition mask is generated for each partition, so that in image I... k，j The partition map I can be obtained by cropping. n，k，j .

[0026] To achieve anomaly localization, each partition n is bound to a spray washing unit number, denoted as u(n). This correspondence is recommended to be obtained through installation, commissioning, and calibration: In commissioning mode, each spray washing unit is turned on individually and short-sequence panoramic images are collected. The platform detects the intensity distribution of the spray washing effect on the inner wall based on the changes in the images before and after the start-up, and then superimposes and statistically analyzes it with the partition mask to determine the partition with the greatest impact contribution, thereby establishing the correspondence between the partition number and the spray washing unit number, and finally solidifying the u(n) mapping table; In another implementation method, the u(n) mapping table is directly given and solidified by the equipment structure design.

[0027] Step 2: Determine the visibility statistics of each zone, which reflect the image clarity and imaging effectiveness, and determine whether a zone is a visible zone based on the visibility statistics. For visible zones, determine the cleaning effect based on the image data and make anomaly judgments on the spray washing action units associated with the zone. For invisible zones, make anomaly judgments on the spray washing action units based on the execution data and image data.

[0028] The visibility statistics reflecting image sharpness and imaging effectiveness for each partition are determined. The specific determination process is as follows: sharpness, exposure effectiveness, and contrast are extracted from the image data of each partition; the sharpness, exposure effectiveness, and contrast are normalized to obtain the corresponding sub-scores; and the sub-scores are fused according to the preset fusion weights to obtain the frame-level visibility score sequence for each partition.

[0029] For the partitioned image I at observation position k n，k，j First convert to grayscale image G n，k， The resolution metric is denoted as A. n，k，j The Laplace variance is used as a feasible and commonly used evaluation method; the exposure effectiveness characterization quantity is denoted as B. n，k，j First calculate the grayscale value. In calculating the saturation ratio s of the grayscale image of the j-th frame at observation position k in partition n. n，k，j It is defined as the proportion of pixels within a partition whose grayscale value is close to the upper or lower limit, for example, the proportion of pixels whose grayscale value is less than or equal to δ; δ is the saturation judgment width, which is derived from the fixed configuration. The exposure effectiveness characterization quantity is expressed in a practical gating form as follows: Where μ0 is the desired center value of brightness, c2 and c3 are coefficients, and clip(x, 0, 1) represents truncation to [0, 1]. The recommended sources for μ0, c2, and c3 are offline calibrations of historical samples: obtaining a reasonable range from normally available images and then fixing it; alternatively, initial values ​​can be provided for the project and then fine-tuned and fixed during the installation and debugging phase. The contrast characterization is denoted as C. n，k，j The grayscale standard deviation is used as an feasible and commonly used evaluation method. This method is sensitive to the distinguishability of texture and boundary and is simple to implement in engineering.

[0030] To eliminate differences in equipment and ambient lighting, the three raw quantities were normalized to obtain individual scores. , , Where A0, A1, B0, B1, C0, and C1 are the normalization upper and lower bounds, and the source is recommended to be offline calibration of historical samples: take the upper and lower bounds according to the quantiles from the available imaging samples and fix them to avoid interference from extreme outliers. Normalize the three component scores according to their weights, ensuring that the sum of the three weights is 1, thus guaranteeing that the fused frame-level visibility score still falls within the range of 0 to 1, resulting in the frame-level visibility score v. n，k，j The weights are derived from offline calibration.

[0031] Considering the inter-frame fluctuations caused by steam mist, reflection, and transient occlusion, robust statistical aggregation is performed on the frame-level visibility score sequence to obtain the partition visibility statistics at observation position k. When multiple observation positions exist, to improve coverage, the visibility of the same partition at different observation positions is aggregated to obtain the final visibility statistics for the partition. This definition states that if there is an observation position that can reliably see the partition, then the partition is considered more usable. In engineering, this can significantly reduce the number of missed judgments caused by occlusion.

[0032] The determination of whether a zone is visible is based on visibility statistics. The specific process is as follows: The visibility statistics of each zone are compared with a visibility threshold. The visibility threshold is used to gate the visibility statistics of a zone, distinguishing whether the zone image meets the imaging conditions to support the determination of cleaning effect and the anomaly of the spray washing unit. When the visibility statistics of a zone are not lower than the visibility threshold, the zone is marked as visible; when the visibility statistics of a zone are lower than the visibility threshold, the zone is marked as invisible. It is recommended that the visibility threshold be obtained through offline calibration and fine-tuning during installation and debugging. Initial values ​​are determined offline using historical samples to achieve the target accuracy for acceptance judgment of visible zones, followed by fine-tuning and solidification using a small number of on-site samples.

[0033] like Figure 3The flowchart of the cleaning effect and spray washing unit anomaly determination based on the visibility statistics gating of the partition provided by this embodiment of the invention is shown. The process begins by acquiring equipment execution data and image data of the inner wall of the flue. After acquiring the relevant data, the inner wall of the flue is divided into spaces. After the space division is completed, the visibility statistics of each partition are further acquired. Then, the judgment step is entered to determine whether the visibility statistics of the partition is not lower than the visibility judgment threshold. When the judgment result is yes, the partition is marked as a visible partition, and the cleaning effect is determined based on the image data and the anomaly determination of the spray washing unit associated with the partition is performed. When the judgment result is no, the partition is marked as an invisible partition, and the anomaly determination of the spray washing unit is performed based on the execution data and image data.

[0034] For visible partitions, the cleaning effect is determined based on image data, and anomalies are judged for the partition-associated spray washing units. The specific process is as follows: determine the residual characterization before cleaning for the pre-cleaning image data belonging to the visible partition, and determine the residual characterization after cleaning for the post-cleaning image data belonging to the visible partition.

[0035] For each visible region, an image sequence is acquired before and after cleaning, and the residual mask q is extracted from the image frames. n，j (x, y), where pixel (x, y) belongs to the pixel set Ω of partition n. n The mask is defined as: when a pixel is determined to be residual, q n，j (x, y) = 1, otherwise 0. Residual extraction can be implemented using a rule-based approach: combining local texture intensity and reflection suppression rules in the luminance and chrominance spaces to form residual candidates, and then using morphological opening and closing operations for denoising to obtain a stable mask. This implementation does not rely on a depth model and is easy to review and deploy in engineering. Define the percentage of residuals per frame. ;in This represents the number of pixels in the partition.

[0036] The residual characterization amount before cleaning is denoted as R. 0，n The residual characterization amount after cleaning is denoted as R. 1，n Robust aggregation is employed. , where r (0) From the pre-cleaning sequence, r (1) From the cleaned sequence; define the characterization of the cleaning effect. ε1 is a small constant to prevent the denominator from being zero, and its source belongs to the configuration solidification.

[0037] The cleaning effect of a zone is determined based on the residual characterization before and after cleaning. The cleaning effect characterization of the zone is then compared with the preset acceptance criteria to generate the cleaning effect judgment result of the visible zone. The cleaning effect characterization must be greater than or equal to the minimum value of the cleaning effect characterization within the specified range, and the residual characterization after cleaning must be less than or equal to the maximum value of the residual characterization after cleaning within the specified range. If the cleaning effect of the visible zone is deemed to be up to standard, then the cleaning effect of the visible zone is deemed to be down to standard, and all other cases are deemed to be down to standard, which means that the conditions for triggering an anomaly are met.

[0038] When the cleaning effect assessment meets the abnormal triggering conditions, the spray cleaning unit number u(n) bound to the partition is located. The output includes: anomaly marker of the spray cleaning unit, partition number, key indicators, and evidence summary. The evidence summary must include at least: the indices of two representative images, one before cleaning and one after cleaning, and the indicator value R. 0，n R 1，n G n V n The relationship u(n) and the partition location number; the above content can be directly used for remote display, review and work order dispatch.

[0039] For unseen zones, anomaly detection of the spray washing unit is performed based on execution data and image data. Specifically, the execution completeness index is determined based on the execution data associated with the unseen zone, and the execution consistency index is determined based on the execution data. The execution completeness index characterizes the degree to which the spray washing unit associated with the unseen zone meets the preset process requirements during the cleaning task. The execution consistency index characterizes whether the execution data of the same spray washing unit satisfies the preset physical self-consistency relationship.

[0040] For the invisible partition n, first take its bound spray washing unit number u(n). The platform reads the process requirements of the spray washing unit from the configuration (sourced from configuration fixing or installation and commissioning fixing), such as: minimum duration t0(u), minimum pressure p0(u), minimum flow rate q0(u), minimum rotation speed s0(u). Statistically calculate the actual measurements of the spray washing unit within the task window from the execution data: actual duration t(u), statistical pressure p(u), statistical flow rate q(u), and statistical rotation speed s(u). The statistical form can be median or mean; median is recommended for greater robustness. Separate scores are generated: duration-based score. ; Pressure in place score Traffic arrival score Speed ​​reached Weights are assigned to the scores for duration, pressure, flow rate, and rotation speed, respectively. Each weight is non-negative, and the sum of the weights is 1, ensuring that the execution completeness index remains within the range of 0 to 1, thus obtaining the execution completeness index K. n The weights can be derived from offline calibration or fixed configuration.

[0041] In the anomaly detection of invisible partition n, the completeness index K is executed. n In addition to characterizing the degree to which the spray washing unit meets the process requirements, the consistency index J is also calculated. n Consistency is used to characterize whether the control actions and monitored responses of the spray washing unit maintain a stable following relationship. The spray washing unit number bound to partition n is denoted as u(n). Within the task time window, the valve opening / closing indication sequence gu(t) and the flow monitoring sequence qu(t) of the spray washing unit are extracted from the execution data; where gu(t) takes a value of 1 to indicate that the valve is in the open state and a value of 0 to indicate that the valve is in the closed state, and qu(t) represents the flow measurement value at the corresponding time. If the equipment only provides common flow monitoring, the common flow sequence can be used instead of qu(t), while the control event positioning of gu(t) is still used to maintain the association with the spray washing unit u(n). To reduce the impact of sampling frequency differences and spike noise, the valve opening / closing indication and flow sequence can be resampled to a fixed time step Δts, and the flow sequence can be de-spiked; where Δts is fixed by configuration or determined by installation and commissioning.

[0042] The instant a valve changes from closed to open is defined as an opening event. The timestamps of these opening events are sequentially denoted as t1, t2, ..., tM, where M represents the number of opening events detected within the task window. For each opening event, a pre-event time window and a post-event time window are used to determine whether the flow rate follows the increase: the pre-event time window is a short period before the opening event, and the post-event time window is a short period after the opening event; the length of these two time windows is uniformly set to Δ, which is either fixed in the configuration or determined during installation and commissioning based on the pipeline response delay. To avoid the influence of instantaneous fluctuations, the median flow rate within the pre-event time window and the median flow rate within the post-event time window are used as representative values. If the increment of the post-event representative value relative to the pre-event representative value is not less than the minimum response threshold qmin, the opening event is considered a hit event, indicating that the control action produced an observable normal flow response; if the increment is less than qmin, the opening event is considered a miss event, indicating possible valve jamming, pipeline blockage, condensation buildup, insufficient supply, or abnormal flow measurement leading to a failure to follow the flow. The minimum response threshold qmin is the minimum flow increment that should result from opening the valve under normal spraying conditions. It is recommended that qmin be obtained and solidified offline from historical normal samples, but it is also permissible to fine-tune and solidify it using a small number of field samples during the installation and commissioning phase.

[0043] Execution Consistency Indicator J nThe hit rate is defined as the ratio of the number of hit events to the total number of open events: the hit rate equals the number of hit events divided by M. When M is 0, Jn is directly set to 0, which reflects that no valid open events have occurred within the task window. A conservative judgment is adopted when there is a lack of consistent evidence.

[0044] The execution-side anomaly triggering rules employ a two-layer gating system: an anomaly is directly triggered when the execution completeness falls below the lower completeness threshold; an anomaly is triggered when the execution completeness is not lower than the lower completeness threshold but the execution consistency falls below the lower consistency threshold. Both thresholds are calibrated and solidified offline using historical samples, or solidified through configuration, installation, debugging, and fine-tuning. Through these rules, the platform can still output interpretable execution-side anomaly judgment results based on execution evidence even under conditions of visual limitations, and pinpoint the anomaly conclusion to the spray washing action unit number u(n).

[0045] After obtaining the anomaly determination result on the execution side, the conclusion is further corrected using weak cue information in the unseen partition image, thereby generating the final anomaly determination result for the unseen partition. The image of the unseen partition is not suitable for residual detail judgment, but directional cue information can still be stably extracted. In this embodiment, this type of information is defined as weak cue features. Weak cue features are used to reflect abnormal signs at the image level during the spraying process, such as the intensity of fogging obstruction, the intensity of brightness fluctuation, and the intensity of dynamic response. The calculation of weak cue features adopts lightweight statistics, and the threshold or reference range is calibrated and solidified offline by historical samples, or finely adjusted and solidified by collecting a small number of on-site samples during installation and debugging.

[0046] Weak cue risk values ​​are calculated based on images from unseen partitions. These risk values ​​quantify the strength of weak cue support for anomalies. Specifically, the platform selects image sequences belonging to unseen partitions within the task time window, calculates weak cue features that can be stably obtained under low-resolution conditions for each frame, and performs robust statistical analysis on the features of each frame to obtain partition-level weak cue feature values. Weak cue features include at least one or more of the following: fogging intensity, brightness fluctuation intensity, and dynamic response intensity. Fogging intensity reflects the contrast reduction caused by water mist or oil stains; brightness fluctuation intensity reflects abnormal fluctuations in overall brightness during the spraying process; and dynamic response intensity reflects the strength of image changes caused by the spraying medium. The platform sets corresponding risk judgment thresholds or threshold ranges for each type of weak cue feature and compares the weak cue feature values ​​with the thresholds to obtain a sub-risk score for that type of weak cue. The thresholds are obtained through offline calibration of historical samples and fixed in the configuration, or fine-tuned based on on-site samples during the installation and debugging phase and then fixed. Subsequently, weights were assigned to each type of weak alert sub-risk score and normalized so that the sum of the weights was 1. Then, a weighted average was calculated for each type of sub-risk score according to its weight to obtain the weak alert risk value. When a certain type of sub-risk score became unavailable, its weight was reset to zero and the remaining weights were re-normalized. The fusion weights were either calibrated offline using historical samples and then fixed, or initially set by configuration and fine-tuned during the installation and debugging phase. The weak alert risk value is limited to the range of 0 to 1; a higher weak alert risk value indicates a stronger support for anomalies.

[0047] Simultaneously, execution-side risk values ​​are determined based on execution completeness and execution consistency metrics. These risk values ​​serve as a measure of the degree of abnormality corresponding to the execution-side anomaly assessment results. Specifically, the platform sets a lower threshold for execution completeness and a lower threshold for execution consistency. Both thresholds are calibrated offline using historical samples and embedded in the configuration, or initial values ​​are provided in the configuration and then fine-tuned during the installation and debugging phase. The platform compares the execution completeness metrics with the lower thresholds. If the execution completeness metrics are lower than the lower thresholds, a completeness risk is triggered and a value of 1 is assigned; otherwise, a value of 0 is assigned. Similarly, the platform compares the execution consistency metrics with the lower thresholds. If the execution consistency metrics are lower than the lower thresholds, a consistency risk is triggered and a value of 1 is assigned; otherwise, a value of 0 is assigned. The platform combines completeness risk and consistency risk into an execution-side risk value according to a preset fusion rule. The preferred preset fusion rule is to take the larger of the two values ​​to ensure that the execution-side risk value remains triggered and is not weakened when either risk is triggered. In another implementation, the preset fusion rule can be implemented by weighted fusion and binarization. That is, a fusion weight is set for completeness risk and consistency risk respectively. The fusion weight is non-negative and normalized so that the sum of the two fusion weights is 1, thus forming a fusion risk value. Then, the fusion risk value is compared with a preset binarization threshold. When the fusion risk value is not lower than the binarization threshold, the execution-side risk value is set to 1. When the fusion risk value is lower than the binarization threshold, the execution-side risk value is set to 0 to ensure that the execution-side risk value is still a binary result. The fusion weights and binarization thresholds are preferably calibrated offline from historical samples and fixed into the configuration. Alternatively, initial engineering values ​​can be provided by the configuration and fine-tuned during the installation and debugging phase. The fusion weights are used to express the relative importance of completeness risk and consistency risk in the execution-side anomaly determination, while the binarization thresholds are used to limit the minimum gating conditions for fusion risk values ​​to trigger anomalies. Through the above-mentioned gated rule scoring method, the execution-side risk value can represent the degree of execution-side anomaly in a very simple and interpretable form, and can participate in subsequent correction risk value fusion and conservative tightening determination together with the weakly indicated risk value.

[0048] The weak cue risk value and the execution-side risk value are fused to obtain a corrected risk value. Based on this corrected risk value and preset correction judgment rules, a correction anomaly judgment result for the invisible partition is generated. The fusion method can employ segmented enhancement, which is used to generate the corrected risk value. The weak cue risk value ranges from 0 to 1, representing the strength of the image's weak cue support for anomalies. The execution-side risk value is either 0 or 1, representing whether the execution evidence has reached the anomaly triggering level. The platform uses the execution-side risk value as a gating mechanism for segmented enhancement: when the execution-side risk value is 1, the corrected risk value is directly set to 1; when the execution-side risk value is 0, the corrected risk value is set to the weak cue risk value, causing the corrected risk value to change with the strength of the weak cue support.

[0049] The preset correction judgment rules are used to generate correction anomaly judgment results from correction risk values, and limit the role of weak prompts to conservative tightening. Specifically: when the execution-side risk value is 1, the correction anomaly judgment result remains in the triggered state, and the weak prompt risk value is only used to supplement the summary of trigger basis, not to reduce the degree of anomaly or cancel the anomaly trigger; when the execution-side risk value is 0, the platform compares the correction risk value with the tightening trigger threshold and the strong tightening threshold to generate a correction anomaly judgment result: if the correction risk value is lower than the tightening trigger threshold, it remains untriggered; if it is between the tightening trigger threshold and the strong tightening threshold, it is tightened to trigger review; if it is not lower than the strong tightening threshold, it is tightened to trigger anomaly. The tightening trigger threshold and the strong tightening threshold are calibrated and solidified offline by historical samples, or fine-tuned and solidified during the installation and debugging phase after configuration solidification, and the strong tightening threshold is not less than the tightening trigger threshold. Through the above rules, weak prompts only provide tightening basis when no anomaly is triggered on the execution side, and do not relax the conclusion of already triggered anomalies, thereby ensuring that the correction judgment of invisible partitions follows the security side strategy.

[0050] Output the final anomaly determination result of the invisible partition and its trigger basis summary. The final anomaly determination result shall include at least the partition number, the associated spraying action unit number, the final anomaly trigger status, and the anomaly degree. The trigger basis summary shall include the correlation information of execution completeness indicators, execution consistency indicators, and weak prompt features. The correlation information shall include at least the key statistics of completeness and consistency, the start event hit statistics, the key statistics of weak prompt features, the weak prompt risk value, the execution side risk value, and whether weak prompt tightening has occurred, thereby supporting remote review, work order dispatch, and audit trail recording.

[0051] Step 3: Based on the visibility statistics of each zone, the anomaly judgment results of the visible zone and the anomaly judgment results of the invisible zone are merged to obtain the cleaning acceptance conclusion of the entire flue.

[0052] After determining the cleaning effect and identifying any abnormalities in the spraying units for each zone, the process proceeds to the flue-level fusion and acceptance phase. A fusion weight is generated for each zone, representing its contribution proportion to the flue-level conclusion. The fusion weight is constructed based on the zone's visibility statistics, ensuring that zones with more reliable image evidence have a greater impact on the final conclusion. To avoid some zones having extremely low visibility and thus zero weight, affecting normalization, a small bias is superimposed on the visibility statistics and normalized: first, the visibility statistics of each zone are added to a preset bias to obtain the original weight value; then, the original weight values ​​of each zone are divided by the sum of the original weight values ​​to obtain the fusion weight. The bias is fixed by the configuration to ensure the numerical stability of the weight calculation. If the equipment has multiple observation positions, the optimal value among all observation positions is prioritized for the zone's visibility statistics, ensuring that if an observation position provides relatively reliable image evidence for that zone, that zone receives a relatively higher weight.

[0053] Once a zone is identified as visible, the platform calculates a score for that zone based on the image. This score characterizes the cleaning effect and is consistent with the anomaly assessment results of the associated spray washing unit. The visible zone score is constructed using a combination of post-cleaning residue level and improvement degree constraints to ensure that the score reflects both the final cleanliness and any significant improvement before and after cleaning. The platform obtains representative values ​​for the percentage of post-cleaning residue and the degree of improvement from the visible zone. The smaller the percentage of post-cleaning residue and the greater the degree of improvement, the higher the score. When the percentage of post-cleaning residue exceeds the acceptance limit or the degree of improvement falls below the improvement limit, the zone score is lowered to a low score range, triggering a zone anomaly. The relevant residue calculations, representative value statistics, and thresholds all originate from the visible zone process in step two. The thresholds are either calibrated and solidified offline using historical samples or fine-tuned and solidified during the installation and debugging phase after configuration solidification.

[0054] The representative value of the residual proportion after cleaning is used to characterize the proportion of area within a partition that is still identified as residual after cleaning. The platform generates a residual binary mask frame by frame for the cleaned image sequence and calculates the residual proportion for each frame. Then, robust statistical aggregation is performed on the residual proportion sequence after cleaning to obtain the representative value of the residual proportion after cleaning. The robust statistical aggregation preferably uses the median or quantile to reduce the impact of short-term disturbances such as water mist and reflection on the results. The representative value of the degree of improvement is used to characterize the decrease in the residual level before and after cleaning in the same partition. The platform calculates the normalized decrease ratio using the representative values ​​of the residual proportion before cleaning and the representative values ​​of the residual proportion after cleaning and truncates it to 0 to 1 to make the representative value of the degree of improvement interpretable and numerically stable.

[0055] For partitions deemed invisible, a score is generated based on execution data and weak cue features. The score for invisible partitions is consistent with the correction anomaly determination result of the associated spraying unit. The score for invisible partitions is constructed using a risk-reverse mapping method: the higher the execution-side risk and the weak cue correction risk, the lower the score. When the correction anomaly determination result is a triggered anomaly, the score for invisible partitions is directly set to a low value or zero. When the correction anomaly determination result is a triggered review, the score for invisible partitions is limited to a low to medium range to reflect uncertainty. Since invisible partitions lack verifiable detailed image evidence, to avoid assigning excessive credibility to invisible partitions in the overall flue score, a capping constraint is imposed on the score of invisible partitions: even if neither the execution-side nor the weak cue triggers anomalies, the score for invisible partitions is not allowed to exceed a preset upper limit. This upper limit is fixed by configuration or by offline calibration of historical samples, with the design goal of maintaining a conservative credibility boundary under invisible conditions.

[0056] The platform calculates the total flue gas score by weighting and aggregating the scores of each partition based on the partition fusion weight. Each partition's score is then weighted and summed according to its fusion weight, with the weights normalized to ensure the total flue gas score reflects the combined contribution of each partition's score and remains within a preset score range. Specifically, the platform first obtains the fusion weight of each partition, takes non-negative values, and normalizes them so that the sum of all partition fusion weights is 1. Then, each partition's score is multiplied by its corresponding fusion weight and summed to obtain the total flue gas score. If any partition score is deemed invalid by quality control, its corresponding fusion weight is reset to zero, and the remaining fusion weights are re-normalized before weighted aggregation to ensure the continuous calculation of the total flue gas score and prevent it from being affected by missing items. After weighted aggregation, the platform uses a strong veto, total score threshold, and review branch acceptance strategy to generate a flue gas cleaning acceptance conclusion, ensuring both safety and maintainability in the project.

[0057] A strong veto rule ensures that no process is approved without addressing any critical anomalies: if any visible zone shows residual cleaning exceeding the acceptance limit, improvement falling below the improvement threshold, or the associated cleaning unit being flagged as abnormal, the platform directly outputs a "fail" conclusion for flue cleaning, listing the triggering zone number and cleaning unit number. Similarly, if any unvisible zone's correction anomaly assessment results in a triggered anomaly, a "fail" conclusion is also output, listing the corresponding cleaning unit number. The thresholds and rules related to the strong veto are configured or calibrated offline using historical samples, and can be fine-tuned during the installation and commissioning phase if necessary.

[0058] The total score threshold rule is used to make a comprehensive judgment on the flue level without triggering a strong veto: the platform compares the total score of the flue with the acceptance threshold. When the total score of the flue is not lower than the acceptance threshold, it outputs "pass"; when the total score of the flue is lower than the acceptance threshold, it outputs "fail" or suggests re-washing. The acceptance threshold is calibrated and solidified offline by historical samples, or it is solidified by configuration and then fine-tuned and solidified during the installation and debugging stage.

[0059] The review branch is used to express the uncertainty brought about by the lack of visibility: when a strong veto is not triggered but there is an invisible partition that triggers review, or when the proportion of invisible partitions exceeds the preset upper limit, the platform outputs the conclusion that review is required or recommends rewashing, and marks the partition that caused the review and the spraying unit number in the conclusion; the upper limit of the proportion of invisible partitions and the review triggering conditions are fixed by configuration or offline calibration.

[0060] When the platform outputs the flue cleaning acceptance conclusion, in addition to the conclusion status, it also outputs the total score of the flue, the list of key triggering zones, the list of corresponding spraying action units, and the evidence summary index. The evidence summary index points to the image evidence summary of the visible zone and the trigger basis summary of the invisible zone, so as to support one-click backtracking and auditing traceability at the remote end.

[0061] Specific Embodiment Two is an improvement on the remote operation and maintenance management method of Embodiment One. The improvement is concentrated on the surrounding field of view partitioning method in Step One: In the process of mapping the surrounding field of view into a set of mutually distinguishable partition units, the spraying execution trajectory is introduced to participate in partition generation, so that the coverage direction of the partition and the spraying action unit is consistent during the cleaning task, thereby improving the reliability and interpretability of the partition-spraying action unit correspondence; the visibility determination, anomaly determination and flue-level fusion process in Steps Two and Three can still be implemented using Embodiment One.

[0062] Within the cleaning task time window, the execution data of the spray washing unit is acquired, and the spray washing phase trajectory is constructed. The execution data includes at least rotation state, spray washing enable state, and timestamp information. The rotation state can be derived from the angle value output by the rotary encoder or from the motor speed signal. The spray washing enable state indicates whether the spraying medium is in an effective spraying state at the corresponding moment. The timestamp aligns the rotation state and the spray washing enable state to the same time axis. The spray washing phase trajectory characterizes the trajectory of the spray washing coverage direction changing over time. Its phase can be represented in angular form: when the rotary encoder directly outputs the angle value, the angle value is normalized to 0°–360° over one rotation cycle to obtain the phase; when only the speed signal is available, the platform integrates the speed based on adjacent timestamp intervals to obtain the angle increment, accumulates it to form the phase, and then normalizes it to 0°–360° over one rotation cycle. The spray washing enable state is used to gate the phase trajectory, ensuring that only the spray washing enable state contributes to the effective time slice statistical coverage, thereby avoiding interference from "rotation without spraying" in partition generation.

[0063] Subsequently, the platform statistically analyzes the dwell time contribution of the spray phase trajectory within the phase domain and determines the phase domain boundary set accordingly. The phase domain is defined as the phase angle space from 0° to 360°. The platform discretizes the phase domain into several phase intervals and accumulates the effective dwell time of the spray in each phase interval. The accumulation method is as follows: when the spray enable state is valid, the time interval between adjacent timestamps is added to the phase interval. To more closely approximate the actual spray coverage intensity, the platform can also introduce execution-side intensity weights to weight the time intervals. For example, the dwell time can be weighted using the normalized value of pressure or flow rate, so that time slices with higher pressure or flow rate contribute more to coverage. After completing the phase domain dwell contribution statistics, the platform determines the phase domain boundary set based on the preset number of partitions or the preset minimum coverage contribution condition. The generation of the phase domain boundary set can adopt an equal contribution division strategy: the cumulative contribution of the phase domain is evenly divided according to the target number of partitions, and the boundary is set at the position where the cumulative contribution reaches the target value of each segment; or a gap division strategy can be adopted: when there are continuous low contribution or near-zero contribution intervals in the phase domain, the boundary of the interval is used as the boundary, so as to explicitly isolate the direction of possible insufficient coverage or intermittent spraying. The target number of partitions, minimum coverage contribution condition and gap judgment threshold are fixed by configuration or fixed after offline calibration of historical samples, and fine-tuning and fixing are allowed in combination with the on-site spraying characteristics during the installation and commissioning stage.

[0064] After obtaining the phase domain boundary set, the platform maps the phase domain boundary set to the panoramic field of view to form a spatial partition set. To achieve a consistent mapping from phase to image angle, an alignment relationship between the phase zero point and the image zero orientation is established during the installation and debugging phase. That is, the corresponding angular position of the spray direction in the panoramic image is determined when the spray phase is 0°, and this alignment offset is fixed into the configuration. When the camera captures panoramic images to form a panoramic unfolded image, the lateral position of the image has a monotonic correspondence with the panoramic angle. Based on this, the platform maps the phase domain boundary angle to the angle boundary line in the image, and combines it with the boundary constraints of the effective imaging area of ​​the inner wall (also calibrated and fixed during the installation and debugging phase) to generate a spatial mask for each partition. Ultimately, each partition in the spatial partition set corresponds to a phase interval defined by the phase domain boundary set, and the partition represents the coverage contribution of the spraying unit within that phase interval. The platform assigns a partition identifier to each partition and establishes a correspondence between the partition identifier and the spraying unit identifier. This correspondence includes not only which spraying unit is responsible but also the corresponding phase interval range, so that subsequent anomaly localization can simultaneously provide the spraying unit number and the coverage direction interval, facilitating remote verification and handling.

[0065] The zoning boundary is driven by the spraying execution trajectory and the effective spraying dwell time contribution. It can adapt to execution differences such as spraying speed fluctuations, intermittent enabling, and local stagnation, avoiding the mismatch between zoning and coverage direction caused by fixed angle division when there are uneven speeds or missed spraying areas in actual spraying. The zoning and phase interval correspond one-to-one, enabling execution evidence under invisible conditions and image evidence under visible conditions to be fused under the same coverage direction semantics. This improves the accuracy and interpretability of anomaly localization to the spraying action unit and provides a more stable zoning basis for flue-level acceptance.

[0066] In a specific embodiment three, this invention provides a remote operation and maintenance management system for commercial kitchen fume cleaning and disinfection equipment, such as... Figure 2 The diagram shows the structure of a remote operation and maintenance management system for commercial kitchen fume cleaning and disinfection equipment. The system includes: a surround view acquisition and zoning module, a visual gate control and judgment module, a weighted fusion acceptance module, and an operation and maintenance database.

[0067] The surround view acquisition zoning module is connected to the visual gate control judgment module, and the visual gate control judgment module is connected to the weighted fusion acceptance module. The surround view acquisition zoning module, the visual gate control judgment module, and the weighted fusion acceptance module are all connected to the operation and maintenance database. The operation and maintenance database is used to store various parameters involved in the remote operation and maintenance management system of commercial kitchen fume cleaning and disinfection equipment.

[0068] The surround-view acquisition module is used to acquire equipment execution data during fume cleaning and disinfection, and control the liftable camera to extend to the observation position for surround-view acquisition, obtaining image data of the inner wall of the flue. The surround-view field of view of the inner wall of the flue at the observation position is divided into a set of spatial partitions covering the inner wall of the flue. The visual gating judgment module is used to determine the visibility statistics of each partition, reflecting the image clarity and imaging effectiveness, and to determine whether a partition is a visible partition based on the visibility statistics. For visible partitions, the cleaning effect is determined based on the image data, and anomaly judgment is performed on the spraying action unit associated with the partition. For invisible partitions, anomaly judgment is performed on the spraying action unit based on the execution data and image data. The weighted fusion acceptance module is used to fuse the anomaly judgment results of visible partitions and invisible partitions based on the visibility statistics of each partition to obtain the cleaning acceptance conclusion of the entire flue.

[0069] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0070] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0071] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0072] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0073] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0074] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, systems, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0075] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A remote operation and maintenance management method for commercial kitchen fume cleaning and disinfection equipment, characterized in that, The method includes: Step 1: Obtain the equipment execution data during oil fume cleaning and disinfection, and control the liftable camera to extend to the observation position to collect images of the inner wall of the flue. Divide the field of view of the inner wall of the flue at the observation position into a set of spatial partitions covering the inner wall of the flue. Step 2: Determine the visibility statistics of each zone, which reflect the image clarity and imaging effectiveness, and determine whether a zone is a visible zone based on the visibility statistics. For visible zones, determine the cleaning effect based on the image data and make an anomaly judgment on the spray washing unit associated with the zone. For non-visible zones, make an anomaly judgment on the spray washing unit based on the execution data and image data. Step 3: Based on the visibility statistics of each zone, the anomaly judgment results of the visible zone and the anomaly judgment results of the invisible zone are merged to obtain the cleaning acceptance conclusion of the entire flue. For the unvisual partition, the anomaly determination of the spray washing unit is based on the execution data and image data, and the specific process is as follows: An execution completeness index is determined based on the execution data associated with the invisible partition, and an execution consistency index is determined based on the execution data. Based on the execution completeness index and the execution consistency index, the execution-side anomaly determination result of the spray washing action unit is generated. The execution completeness index is used to characterize the degree to which the spray washing action unit associated with the invisible partition meets the preset process requirements in the cleaning task. The execution consistency index is used to characterize whether the execution data of the same spray washing action unit meets the preset physical self-consistency relationship. Image data belonging to the invisible partition is extracted as weak cue features, and the execution-side anomaly determination result is corrected based on the weak cue features to generate anomaly determination results for the invisible partition. The weak cue features are used to characterize directional cue information that can still be stably extracted from the image data of the invisible partition. Output the anomaly determination result of the invisible partition and its corresponding trigger basis summary. The trigger basis summary includes the association information of the execution completeness index, the execution consistency index and the weak prompt feature.

2. The remote operation and maintenance management method for the commercial kitchen fume cleaning and disinfection equipment according to claim 1, characterized in that, The process of dividing the panoramic field of view of the flue inner wall at the observation position into a set of spatial partitions covering the flue inner wall is as follows: Based on the preset field of view space division rules, the boundary set of the surrounding field of view in the angular domain and the boundary constraints in the spatial domain are determined; Based on the boundary set and the boundary constraints, the surrounding field of view is mapped into a set of mutually distinguishable partitioned units; Each zone unit is assigned a zone identifier, and a correspondence is established between the zone identifier and the identifier of the spray washing unit.

3. The remote operation and maintenance management method for the commercial kitchen fume cleaning and disinfection equipment according to claim 1, characterized in that, The process for determining the visibility statistics of each partition, which reflect image clarity and imaging effectiveness, is as follows: Sharpness, exposure effectiveness, and contrast are extracted from the image data of each partition. Normalization is then performed on the sharpness, exposure effectiveness, and contrast to obtain the corresponding sub-scores. The scores of each component are fused according to the preset fusion weights to obtain the frame-level visibility score sequence of each partition; Robust statistical aggregation is performed on the frame-level visibility score sequence to obtain the visibility statistics for each partition.

4. The remote operation and maintenance management method for the commercial kitchen fume cleaning and disinfection equipment according to claim 1, characterized in that, The specific process for determining whether a partition is a visible partition based on the visibility statistics is as follows: The visibility statistics of each partition are compared with the visibility judgment threshold. The visibility judgment threshold is used to gating the visibility statistics of the partition to distinguish whether the partition image has the imaging conditions to support the judgment of cleaning effect and the judgment of abnormality of the spray washing unit. When the visibility statistics of a certain partition are not lower than the visibility determination threshold, the partition is marked as a visible partition. When the visibility statistics of a certain partition are lower than the visibility determination threshold, the partition is marked as an invisible partition.

5. The remote operation and maintenance management method for the commercial kitchen fume cleaning and disinfection equipment according to claim 1, characterized in that, For the visible partition, the cleaning effect is determined based on the image data, and anomalies are identified in the partition-associated spray cleaning unit. The specific process is as follows: Determine the residual characterization value before cleaning for the image data before cleaning belonging to the visible partition, and determine the residual characterization value after cleaning for the image data after cleaning belonging to the visible partition. Based on the residual characterization quantity before cleaning and the residual characterization quantity after cleaning, the characterization quantity of the cleaning effect of the zone is determined, and the characterization quantity of the cleaning effect of the zone is compared with the preset acceptance judgment conditions to generate the cleaning effect judgment result of the visible zone. When the cleaning effect determination result meets the abnormal triggering condition, the abnormal determination result of the spray washing unit associated with the visible partition is output, and a partition image evidence summary corresponding to the abnormal determination result is generated.

6. The remote operation and maintenance management method for the commercial kitchen fume cleaning and disinfection equipment according to claim 1, characterized in that, The correction process for the execution-side anomaly determination result based on the weak cue features is as follows: The weak prompt risk value is determined based on the weak prompt characteristics, and the execution-side risk value is determined based on the execution completeness index and the execution consistency index. The execution-side risk value is the degree of abnormality corresponding to the execution-side abnormality judgment result. The weak prompt risk value is fused with the execution-side risk value to obtain a correction risk value. Based on the correction risk value and the preset correction judgment rule, a correction anomaly judgment result for the invisible partition is generated. The preset correction judgment rule limits the effect of the weak prompt risk value on the anomaly judgment result to be conservative and tightened.

7. The remote operation and maintenance management method for the commercial kitchen fume cleaning and disinfection equipment according to claim 1, characterized in that, The process of fusing the anomaly detection results of visible and invisible partitions based on the visibility statistics of each partition is as follows: The partition fusion weight is determined based on the visibility statistics of each partition. For the visible partition, the visible partition score is obtained based on the image data. The visible partition score is used to characterize the cleaning effect and corresponds to the abnormal judgment result of the spray washing unit associated with that partition. For invisible partitions, an invisible partition score is obtained based on the execution data and image data. The invisible partition score corresponds to the abnormal judgment result of the spray washing action unit associated with the partition, and the invisible partition score is subject to a preset cap constraint to limit its upper limit. The scores of the visible and invisible partitions are weighted and aggregated according to the partition fusion weight to obtain the total score of the flue, and the flue cleaning acceptance conclusion is generated from the total score of the flue.

8. The remote operation and maintenance management method for the commercial kitchen fume cleaning and disinfection equipment according to claim 2, characterized in that, In the process of mapping the surrounding field of view into a set of mutually distinct partitioned units, partitioning can also be performed based on the spraying execution trajectory. The specific process is as follows: Based on the equipment execution data, the rotation state, spraying enable state and timestamp information of the spraying action unit during the cleaning task are obtained, and a spraying phase trajectory characterizing the change of the spraying coverage direction over time is constructed. The phase domain boundary set is determined based on the residence contribution of the spray phase trajectory in the phase domain. The phase domain boundary set is mapped to the surrounding field of view to form a spatial partition set, so that each partition in the spatial partition set corresponds to the coverage contribution of the spray phase trajectory within the phase interval defined by the phase domain boundary set, thereby making each partition correspond to the coverage direction of the spray action unit.

9. A system applying the remote operation and maintenance management method for commercial kitchen fume cleaning and disinfection equipment as described in any one of claims 1-8, characterized in that, include: Surround view acquisition zoning module, visual gate control judgment module, weighted fusion acceptance module; The surround view acquisition zoning module is used to acquire equipment execution data during oil fume cleaning and disinfection, and control the liftable camera to extend to the observation position to acquire surround view data, obtain image data of the inner wall of the flue, and divide the surround view field of view of the inner wall of the flue at the observation position into a set of spatial zoning zones covering the inner wall of the flue. The visual gating judgment module is used to determine the visibility statistics of each zone, which reflect the image clarity and imaging effectiveness, and to determine whether a zone is a visible zone based on the visibility statistics. For visible zones, the cleaning effect is determined based on the image data and anomaly judgment is made on the spray washing action unit associated with the zone. For non-visible zones, anomaly judgment is made on the spray washing action unit based on the execution data and image data. The weighted fusion acceptance module is used to fuse the anomaly judgment results of visible and invisible zones based on the visibility statistics of each zone, and obtain the cleaning acceptance conclusion of the entire flue.

Images