Image processing method for reconnaissance of heating and ventilation equipment

Abstract

The invention discloses an image processing method for reconnaissance of heating and ventilation equipment, and particularly relates to the technical field of image processing of reconnaissance of heating and ventilation equipment, comprising the following steps: acquiring a field image shot for target heating and ventilation equipment, performing equipment positioning processing on the field image, and determining an equipment area corresponding to the target heating and ventilation equipment, extracting a state object region in the equipment region, and outputting a state object region set; a state object area in a heating and ventilation equipment image is positioned, a peripheral evidence area of the state object area is positioned according to a state forming mechanism corresponding to an object type, and attachment change, wear change, corrosion change, connection change or blocking change in the peripheral evidence area are identified and merged; and the states of filth blockage of the filter screen, clamping stagnation of the valve, aging of parts and the like are judged according to peripheral change evidences.

Description

Technical Field

[0001] This invention relates to the field of image processing technology for HVAC equipment surveying, and more specifically, to an image processing method for HVAC equipment surveying. Background Technology

[0002] In the existing HVAC equipment survey and energy-saving diagnosis work, the mainstream practice in the industry is to solve the problem of the difficulty in quickly obtaining the status of on-site equipment. Usually, staff use mobile phones or handheld terminals to take photos of the equipment, and then perform target positioning and image recognition on the filters, valves, dials and electrical components in the photos to give conclusions such as whether the filters are dirty and clogged, whether the valves are stuck, whether the dial parameters are abnormal, and whether the components are aging. Taking the inspection of air conditioning rooms in shopping malls or office buildings as an example, staff often need to quickly take pictures of multiple devices along narrow passages without shutting down the machines. There are hard constraints such as insufficient light, obvious reflections, frequent obstructions, difficulty in ensuring the shooting angle, and inability to disassemble and verify the equipment. In addition, the image results formed by a single shooting are required to be directly used for subsequent operation judgment and energy-saving strategy generation. Under the aforementioned constraints, mainstream practices consistently reveal observable flaws. While photos may capture the valve body, filter surface, or component appearance, the identification results often show inconsistent conclusions for the same object in different photos. For example, a filter may appear black but cannot be consistently identified as being clogged due to long-term dirt, a valve may appear to be at a certain angle but cannot be consistently identified as being stuck, or a component may appear old but cannot be consistently identified as being overly aged. This is because the key evidence for these conditions is not always entirely concentrated on the object itself, but rather appears more often in changes at the connection points, limit contacts, edge attachments, adjacent air duct inlets, and related areas such as fasteners and nameplates around the object. Existing methods often only directly identify the object itself, lacking the identification and merging of this surrounding evidence, resulting in unstable condition determinations that are difficult to verify. The technical problem this application aims to solve is: how to reliably identify conditions such as clogged filters, stuck valves, and aging components in HVAC equipment survey images, under conditions where shooting is limited and disassembly and inspection are not possible, by utilizing visible changes in the surrounding area of ​​the object. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an image processing method for HVAC equipment surveying. By locating the state object area in the HVAC equipment image and locating its surrounding evidence area according to the state formation mechanism corresponding to the object type, the method identifies and merges the adhesion changes, wear changes, corrosion changes, connection changes, or obstruction changes in the surrounding evidence area, thereby using the surrounding change evidence to determine the state of filter clogging, valve jamming, and component aging.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an image processing method for HVAC equipment surveying, comprising: S1. Acquire on-site images of the target HVAC equipment, perform equipment positioning processing on the on-site images, determine the equipment area corresponding to the target HVAC equipment, extract the status object area within the equipment area, and output the status object area set. The state object region represents the process of performing object validity determination on the candidate object region set, retaining candidate object regions that meet the object shape conditions and positional relationship conditions; S2. Perform object type identification processing on each state object region in the state object region set, determine the state object type corresponding to each state object region, and locate the surrounding evidence region carrying residual traces within the outer adjacent range according to the state formation mechanism corresponding to each state object type, and output the surrounding evidence region set. S3. Perform image recognition processing on each set of surrounding evidence regions to identify changes in attachment, wear, corrosion, connection or obstruction in the surrounding evidence regions, and merge them into the corresponding state object regions according to the adjacency relationship, and output the set of evidence of surrounding changes. S4. Perform state determination processing on each set of surrounding change evidence according to the corresponding state object type. Based on the set of surrounding change evidence, determine the state of filter clogging, valve jamming, or electrical component aging, and output the object state result. S4 includes: S4-1. Obtain the surrounding change evidence set and object type results corresponding to each state object area. Calculate the change intensity, change distribution range and change continuity for each surrounding change evidence set, including attachment change, wear change, corrosion change, connection change or obstruction change. Output the change judgment feature set corresponding to each state object area. S4-2. Based on the object type result, call the state determination rule corresponding to the object type identifier, perform dirt and blockage matching processing on the change determination feature set corresponding to the filter, perform jamming matching processing on the change determination feature set corresponding to the valve, perform aging matching processing on the change determination feature set corresponding to the electrical components, and output the state candidate results corresponding to each state object area. S4-3. Perform conflict verification processing on each state candidate result, determine whether the change judgment features in the same state object area meet the preset co-occurrence condition and exclusion condition, retain the state candidate result that meets the preset co-occurrence condition and does not meet the exclusion condition as the object state result, and output the object state result. S5. Perform equipment aggregation processing on the status results of each object, summarize the status results of objects belonging to the same target HVAC equipment, form the survey status results corresponding to the target HVAC equipment, and output the operating status identification results of the target HVAC equipment based on the survey status results.

[0005] In a preferred embodiment, S1 includes: S1-1. Perform equipment contour recognition and structural boundary recognition processing on the on-site image to determine the outer contour range and internal structure distribution range of the target HVAC equipment in the on-site image, and output the equipment boundary results. S1-2. Perform region constraint decomposition processing on the equipment boundary results. According to the preset state objects in the target HVAC equipment, decompose the equipment boundary results into multiple candidate object regions and output the candidate object region set. S1-3. Perform object validity determination processing on the candidate object region set, retain the candidate object regions that meet the object shape conditions and positional relationship conditions as state object regions, and output the state object region set.

[0006] In a preferred embodiment, S2 includes: S2-1. Obtain the target state object region from the state object region set, perform object type recognition processing on the target state object region, generate an object type identifier corresponding to the target state object region, and output the object type result. S2-2. Based on the object type result, call the state formation mechanism mapping relationship that corresponds one-to-one with the object type identifier, solve the set of residual trace carrying elements corresponding to the object type identifier, and generate candidate surrounding areas within the outer adjacent range of the target state object area with each residual trace carrying element as a constraint, and output the set of candidate surrounding areas. S2-3. Perform region filtering processing on each candidate surrounding region in the candidate surrounding region set, calculate the region spacing, relative orientation and boundary connection relationship between each candidate surrounding region and the target state object region, retain the candidate surrounding regions whose region spacing falls within the preset adjacency distance range, whose relative orientation conforms to the preset orientation rule corresponding to the object type and whose boundary connection relationship conforms to the structural connection rule corresponding to the object type as the surrounding evidence region, and output the surrounding evidence region set.

[0007] In a preferred embodiment, the process of outputting the candidate surrounding region set in S2-2 further includes: S2-21. Obtain the object type result and read the state formation mechanism mapping relationship that corresponds one-to-one with the object type identifier. Perform element calculation on the state formation mechanism mapping relationship to obtain the set of residual trace bearing elements. Generate a constraint set containing spatial orientation constraints, structural connection constraints and visibility constraints for each bearing element in the set of residual trace bearing elements. Output the bearing element constraint set. S2-22. Obtain the target state object region and limit its peripheral adjacent range. Perform boundary and structural primitive extraction on the image regions within the peripheral adjacent range to form a candidate region map. Calculate the orientation deviation, connectivity inconsistency, and visibility loss of each candidate region in the candidate region map relative to the target state object region to form the constraint violation cost. When performing constrained search on the candidate region map, determine the search expansion order from low to high constraint violation cost and generate candidate surrounding regions item by item under the premise of satisfying spatial orientation constraints and structural connectivity constraints. Output the initial candidate surrounding region set sorted by constraint violation cost. S2-23. Perform iterative confidence update processing on the initial candidate surrounding area set, convert the degree of satisfaction between each initial candidate surrounding area and the constraint set of carrying elements into confidence, and reuse the confidence of the previous round to prune the candidate area map in the iteration to control the consumption of computing resources. When the confidence ranking remains unchanged for two consecutive rounds of iteration and the candidate set is no longer added or deleted, output the candidate surrounding area set.

[0008] In a preferred embodiment, S3 includes: S3-1. Obtain the set of surrounding evidence regions and perform image standardization processing on each of the surrounding evidence regions, including brightness normalization for uneven illumination, suppression and separation for reflective highlights, and sharpness compensation for motion blur, and output the standardized set of surrounding evidence regions. S3-2. Perform change type identification processing on each of the surrounding evidence regions in the standardized surrounding evidence region set, including calculating the texture density increment of attachment change, the edge continuity break of wear change, the color diffusion connectivity of corrosion change, the structural alignment offset of connection change, and the morphological passage gap of obstruction change, and output the surrounding change candidate set according to the relationship between each change quantity and the corresponding change type discrimination condition. S3-3. Perform adjacency merging processing on the candidate set of surrounding changes, including calculating the regional distance and relative orientation between each candidate of surrounding changes and its corresponding state object region and verifying its structural adjacency relationship. Retain the candidates of surrounding changes that simultaneously meet the requirements of regional distance within the adjacency range, relative orientation conforming to the orientation rules of object type and structural adjacency relationship and merge them into the corresponding state object region, and output the evidence set of surrounding changes.

[0009] In a preferred embodiment, the process of outputting the state candidate results corresponding to each state object region in S4-2 further includes: S4-21. Obtain the object type result and change judgment feature set, call the corresponding state judgment rule according to the object type identifier, and decompose the state judgment rule into feature satisfaction condition, feature exclusion condition and feature co-occurrence condition, and output the rule constraint set corresponding to each state object region; S4-22. Perform candidate solution processing on the change judgment feature set of each state object region. Calculate the support, conflict degree and missing degree of each change judgment feature to each state candidate category in the rule constraint set. Sort each state candidate category in the order of support from high to low, conflict degree from low to high and missing degree from low to high. Retain the state candidate categories with the highest support ranking and which simultaneously satisfy the conflict degree and missing degree of no higher than the other state candidate categories in the corresponding ranking relationship as the initial state candidate results. Output the initial state candidate result set. S4-23. Perform iterative correction processing on the initial state candidate result set. Use the support, conflict and missing values ​​in each initial state candidate result as confidence update input. Combine the co-occurrence consistency between the surrounding change evidence adjacent to the current state object region to perform confidence correction on each initial state candidate result. Retain the initial state candidate results whose confidence ranking remains unchanged for two consecutive rounds and whose state candidate category no longer changes as state candidate results. Output the state candidate results corresponding to each state object region.

[0010] In a preferred embodiment, S5 includes: S5-1. Obtain the status results of each object and its corresponding status object area location information. Perform intra-device classification processing on the status results of each object according to the location belonging relationship and object type belonging relationship within the device area, and output the device status unit set. S5-2. Perform association and aggregation processing on the equipment status unit set, calculate the co-occurrence relationship, conflict relationship and supplementary relationship between each equipment status unit, and perform combination reconstruction on each equipment status unit according to the co-occurrence relationship, conflict relationship and supplementary relationship, and output the exploration status result set. S5-3. Perform operational status identification processing on the survey status result set, determine the operational status category corresponding to the target HVAC equipment based on the combination results of each equipment status unit in the survey status result set, and output the operational status identification result of the target HVAC equipment.

[0011] The technical effects and advantages of this invention are as follows: 1. This solution first locates the state object area, then locates the surrounding evidence area according to the state formation mechanism corresponding to the object type, and makes state determination based on surrounding change evidence rather than just the appearance of the object itself. Under the conditions of limited shooting and inability to disassemble and verify, it can relatively improve the stability of state identification such as filter clogging, valve jamming and component aging, and better meet the judgment needs when the object body evidence is incomplete. 2. The mapping relationship of the state formation mechanism is solved into a set of residual trace carrying elements, and the surrounding evidence area is screened in the adjacent range based on this. This can make the area involved in the identification consistent with the actual structural relationship and abnormal formation path of the object, thereby relatively suppressing the situation where the background adjacent area and structurally similar area are mistakenly introduced into the judgment process. 3. Brightness normalization, reflection suppression and clarity compensation are performed on the surrounding evidence area, and then the adhesion changes, wear changes, corrosion changes, connection changes and obstruction changes are identified. Under conditions of uneven lighting, obvious reflection and blurry shooting, the distinguishability of the surrounding residual traces can be relatively improved, and the impact of shooting condition fluctuations on the consistency of subsequent state judgment can be reduced. 4. Merging candidate changes in the surrounding area into the corresponding state object area according to the regional spacing, relative orientation and structural adjacency relationship can establish a traceable attribution relationship between change evidence and target object, thereby relatively improving the problem of inconsistent conclusions for the same object in different photos, and providing a clearer source of evidence for subsequent state determination. 5. Extracting the intensity, distribution range, and continuity of change from the surrounding change evidence set, and combining it with the satisfying conditions, exclusion conditions, and co-occurrence conditions in the state determination rules to generate state candidate results, enables the object state determination to be based on the joint constraints of multiple types of change evidence, thereby relatively reducing the possibility of misjudgment or omission caused by a single local phenomenon. Attached Figure Description

[0012] Figure 1 This is a flowchart outlining the method steps of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Refer to the instruction manual appendix Figure 1 The present invention provides an image processing method for surveying HVAC equipment, comprising: S1. Acquire on-site images of the target HVAC equipment, perform equipment positioning processing on the on-site images, determine the equipment area corresponding to the target HVAC equipment, extract the status object area within the equipment area, and output the status object area set. In this specific embodiment, S1 is used to first determine the actual image boundary of the target HVAC equipment from the on-site images taken by the staff, and then decompose the state object region related to subsequent state determination within the equipment boundary, thereby providing stable input for subsequent object type identification, surrounding evidence region location, and surrounding change identification. Its working mechanism is as follows: first, the target HVAC equipment in the on-site image is separated from the background environment by the equipment outline and internal structural boundaries; then, the equipment region is decomposed into candidate object regions that can be screened by utilizing the installation position relationships and structural adjacency relationships of different state objects inside the HVAC equipment; finally, through object shape and position relationship verification, regions that can stably represent filters, valves, dials, or electrical components are retained, avoiding the misidentification of background components, obstructed areas, or invalid structures as state object regions. This implementation process includes the following steps: The purpose of S1-1 is to extract the outer contour range and internal structure distribution range of the target HVAC equipment from the field images, so as to form the equipment boundary results required for subsequent region decomposition; the input is the field image, which can be a single image taken by the staff using a mobile terminal in the machine room, ceiling mezzanine or equipment platform, or it can be a target frame extracted from a continuous shooting sequence. The processing steps include performing brightness equalization and edge enhancement on the on-site image to reduce the impact of shadows, reflections, and local blurring on contour recognition. Then, based on the continuous edges, regular polylines, shell corners, and internal structure dividing lines of the equipment's outer contour, contour recognition and structural boundary recognition are performed to obtain the closed area of ​​the target HVAC equipment's outer contour and the distribution area of ​​its internal structure. The outer contour range is used to define the overall boundary of the equipment, and the internal structure distribution range is used to identify the location of internal structures such as filter frames, valve connection positions, panel mounting positions, and electrical component mounting areas. The output is the equipment boundary result, which is written to the equipment area cache corresponding to the current image for S1-2 to read. If there is occlusion, reflection, or edge breakage in the on-site image that causes the outer contour to not be closed, the broken boundary is repaired according to the continuity of adjacent edge directions. If multiple equipment contours are detected in the same image, the equipment boundary result corresponding to the target HVAC equipment is determined by combining the shooting center position, equipment area ratio, and the equipment boundary result of the previous frame. The purpose of S1-2 is to decompose the equipment boundary results into multiple candidate object regions with type orientation based on the installation rules and structural adjacency rules of each state object inside the target HVAC equipment, thereby narrowing the scope of subsequent object validity determination; the input is the equipment boundary results, and the processing actions include reading the outer contour range and internal structure distribution range in the equipment boundary results, calling the equipment structure template corresponding to the type of the target HVAC equipment, and the equipment structure template pre-stores the installation position relationships and structural adjacency relationships such as filters are usually located in the return air side or air inlet side frame area, valves are usually adjacent to the pipeline connection area, dials are usually located in the pipeline or unit surface mounting position, and electrical components are usually located in the electrical control box or wiring cavity area; Then, based on the installation location relationship and structural adjacency relationship, the device boundary results are decomposed into region constraints, dividing the regions that satisfy the location constraints and adjacency constraints into multiple candidate object regions. The source of the location constraints is the preset configuration of the device structure template, and the source of the adjacency constraints is the actual connection relationship between the internal structural boundaries of the device. The output is a set of candidate object regions, and the region coordinates, relative device orientation, and adjacent structure identifiers corresponding to each candidate object region are written into the candidate region index table for S1-3 to read. If the device structure template is inconsistent with the device orientation in the current image, the template is first oriented according to the direction of the long side of the outer contour and the interface orientation before decomposition. If no candidate object region is segmented under a certain location constraint, the location is marked as a region to be supplemented, and subsequent steps are allowed to supplement and generate candidate object regions from adjacent structural regions. The purpose of S1-3 is to filter out the truly usable state object regions from the candidate object region set, thereby avoiding irrelevant regions from entering the subsequent object type recognition. The input is the candidate object region set and the candidate region index table. The processing actions include calculating the object morphology conditions and positional relationship conditions for each candidate object region. The object morphology conditions are used to check whether the region has the basic morphological characteristics of the corresponding object. For example, the degree of closure of the frame and the continuity of the mesh texture in the filter area, the valve body outline and connector combination shape in the valve area, the circular or near-circular dial structure and scale distribution characteristics in the dial area, and the shell boundary and terminal distribution characteristics in the electrical component area. The positional relationship conditions are used to check whether the region is still located in the corresponding installation position and maintains the correct orientation relationship with the adjacent structure. For example, whether the filter area is located adjacent to the air inlet channel, whether the valve area is adjacent to the pipeline connection structure, whether the dial area is located in the installation position of the pipeline or equipment panel, and whether the electrical component area is located adjacent to the electrical control box or wiring cavity. When both the object shape condition and the positional relationship condition are met, the candidate object region is retained as the state object region; the output is the state object region set, and the state object region set is written into the object region result table for S2 to read; if a candidate object region only meets the object shape condition but not the positional relationship condition, it is judged as a structurally similar interference region and removed; if a candidate object region meets the positional relationship condition but the object shape is incomplete, it is combined with its adjacent candidate regions and the region is merged and re-judged to reduce false removals caused by local occlusion. In this embodiment, the boundary of the target HVAC equipment can be stably determined in the on-site image first. Then, based on the internal structural rules of the equipment, candidate object regions related to state recognition can be decomposed and further filtered to obtain a set of state object regions. This provides input with clear boundaries, reliable location, and clear structural relationship for subsequent object type recognition and surrounding evidence region positioning, reducing the impact of background interference, local occlusion, and structural similarity misidentification on subsequent state determination. In practical applications: After staff take on-site photos of a modular air conditioning unit in the air conditioning room of a shopping mall, the system first identifies the outer contour and internal partition boundaries of the air conditioning unit from the images. Then, based on the installation patterns of the filter screen located in the return air side frame area, the valve adjacent pipeline interface area, and the electrical components located in the electrical control box area, the equipment boundary results are decomposed into multiple candidate object areas. Subsequently, each candidate object area is checked for shape and position, and areas that are similar to the filter screen frame but are actually shell reinforcement ribs are eliminated, as are invalid areas that are adjacent to the valve connection position but lack valve body shape. Finally, the filter screen area, valve area, and electrical component area are retained as a set of state object areas, and the set of state object areas is directly read by subsequent steps to continue to complete object type identification, surrounding evidence area location, and equipment status determination.

[0015] S2. Perform object type identification processing on each state object region in the state object region set, determine the state object type corresponding to each state object region, and locate the surrounding evidence region carrying residual traces within the outer adjacent range according to the state formation mechanism corresponding to each state object type, and output the surrounding evidence region set. In this specific embodiment, S2 is used to further determine the object type corresponding to the state object based on the already obtained state object area, and according to the mechanism of the abnormal state formed by the object type in actual operation, reversely search for the surrounding evidence area that can carry the residual traces, thereby providing input with clear source and structural constraints for subsequent identification of surrounding changes; Its working mechanism is as follows: First, by identifying the object type, a correspondence is established between the state object area and the specific object type in the filter, valve, dial, or electrical component. Then, the state formation mechanism mapping relationship corresponding to the object type is invoked to solve the problem of what kind of residual traces the object type usually leaves in the adjacent area when it is in a state of dirt blockage, jamming, aging, etc., as well as the spatial location, structural connection method, and visibility conditions of these residual traces. Subsequently, candidate surrounding areas are generated and screened within the outer perimeter of the target state object area, and finally, a set of surrounding evidence areas is output to ensure that the subsequent identification is not an arbitrary neighboring area, but an evidence-bearing area consistent with the state formation mechanism. The implementation process includes the following steps: The purpose of S2-1 is to establish a clear object type attribution for the target state object region, thereby providing a type index for subsequent calls to the state formation mechanism mapping relationship; the input is the target state object region in the state object region set, as well as the region coordinates, relative device orientation, and adjacent structure identifiers written by S1; the processing actions include first performing region normalization on the target state object region, rotating the region to the same orientation as the device's main direction and unifying the resolution, and then extracting the morphological features, texture features, and structural features of the target state object region, where morphological features are used to characterize the shape of the region's outer contour and the closure of its boundaries, texture features are used to characterize the mesh texture, dial scale texture, or shell surface texture, and structural features are used to characterize the distribution relationship of interfaces, connectors, borders, and auxiliary components; The features are then input into the object type recognition model. The output of the object type recognition model is the model output, which includes the corresponding object type identifier for filters, valves, dials, and electrical components. If the model outputs multiple candidate object types, the installation position relationship and structural adjacency relationship written by S1 are combined to perform a consistency check on the multiple candidate object types, retaining the object type identifier that is consistent with the installation position relationship and structural adjacency relationship. The output is the object type result, which is written into the type index table corresponding to the target state object region for S2-2 and S2-21 to read. If the target state object region is occluded, resulting in incomplete features, the supplementary image features of the same region in adjacent frames are called to perform splicing recognition. If adjacent frames are missing, the object types with the highest model output ranking are retained and written with low confidence markers for further verification during subsequent screening. The purpose of S2-2 is to establish a mechanistic association between the target state object area and its surrounding areas that may produce residual traces based on the object type result, thereby generating a candidate surrounding area set. The input quantities are the object type result, the target state object area and its surrounding adjacent range. The processing actions include reading the state formation mechanism mapping relationship that corresponds one-to-one with the object type identifier. The state formation mechanism mapping relationship is used to describe the adjacency direction, connection position and manifestation conditions of residual traces when the object is in an abnormal state. For example, when the filter screen is clogged for a long time, residual traces usually appear on the filter screen edge attachment strip, the air inlet side adjacent strip and the air duct inlet adjacent strip. When the valve is stuck, residual traces usually appear at the valve body connection, limit contact and actuator adjacent strip. When electrical components are aged, residual traces usually appear at the shell edge, terminal adjacent strip and fastener adjacent strip. Then, based on the mapping relationship, the set of residual trace bearing elements is solved, and with each residual trace bearing element as a constraint, candidate surrounding areas corresponding to the bearing elements are generated within the outer perimeter of the target state object area. The value of the outer perimeter is determined by rule constraints based on the structural scale and installation distance relationship corresponding to the object type. The output is a set of candidate surrounding areas, and the bearing element identifier, area coordinates and object type corresponding to each candidate surrounding area are written into the candidate surrounding area table for S2-3 and S2-22 to read. If a bearing element does not have a corresponding area detected within the current outer perimeter, the missing bearing element mark is retained and other candidate surrounding areas are generated to avoid the failure of the entire candidate surrounding area set generation due to the missing bearing element. The purpose of S2-3 is to remove regions from the candidate surrounding region set that lack actual adjacency and structural correspondence with the target state object region, thereby obtaining a set of surrounding evidence regions that can be used for subsequent change identification. The inputs are the candidate surrounding region set, the target state object region, and the object type result. The processing actions include calculating the regional distance, relative orientation, and boundary connection relationship between each candidate surrounding region in the candidate surrounding region set and the target state object region. The regional distance is calculated by the distance between the center point of the candidate surrounding region and the nearest point of the boundary of the target state object region. The relative orientation is calculated by the direction angle of the center point of the candidate surrounding region relative to the center point of the target state object region. The boundary connection relationship is determined by whether there is a continuous connection structure or a shared adjacency structure between the boundary of the candidate surrounding region and the boundary of the target state object region. Then, the distance between regions is compared with the adjacency distance corresponding to the object type, the relative orientation is compared with the orientation relationship corresponding to the object type, and the boundary connection relationship is compared with the structural connection rule corresponding to the object type. Candidate surrounding regions that simultaneously meet all three criteria are retained as surrounding evidence regions. The output is a set of surrounding evidence regions, which is written into the surrounding evidence region index table for S3 to read. If a candidate surrounding region meets the requirements in terms of distance between regions but lacks structural connection relationship, it is determined to be a background adjacent interference region and is removed. If a candidate surrounding region has a valid structural connection relationship but a large relative orientation deviation, the relative orientation is recalculated based on the equipment orientation correction result before the filtering is performed to reduce false rejections caused by oblique shooting. The purpose of S2-21 is to further solve the state formation mechanism mapping relationship into a set of bearing element constraints that can be directly used for searching and filtering, thereby transforming the mechanism description into executable region generation conditions. The input is the object type result and the state formation mechanism mapping relationship that corresponds one-to-one with the object type identifier. The processing actions include performing element solving on the state formation mechanism mapping relationship, decomposing the adjacency direction, connection position, display mode and visibility conditions in the mapping relationship into a set of residual trace bearing elements, and then generating spatial orientation constraints, structural connection constraints and visibility constraints for each bearing element. The spatial orientation constraints are used to limit the direction and angle range that the bearing element should appear relative to the target state object area. The structural connection constraints are used to limit the type of adjacency structure that should exist between the area where the bearing element is located and the target state object area. The visibility constraints are used to limit whether the area should have identifiable boundaries, separable textures or observable surfaces under the current shooting conditions. The spatial orientation constraints and structural connection constraints are derived from rule constraints, while the visibility constraints are derived from the model output and rule constraints. The output is the set of load-bearing element constraints, which is written into the candidate region generation cache for S2-22 and S2-23 to read. If a load-bearing element cannot meet the visibility constraints in the current equipment structure due to occlusion or installation method, the invisible marker corresponding to the load-bearing element is retained, and its expansion priority is reduced in subsequent searches instead of being directly deleted, so as to retain the possibility of compensation recognition. The purpose of S2-22 is to perform a constrained search within the peripheral adjacency range of the target state object region based on the constraint set of carrying elements, and generate a sorted initial candidate surrounding region set; the inputs are the target state object region, the peripheral adjacency range, and the constraint set of carrying elements; the processing actions include first performing boundary extraction and structural primitive extraction on the image region within the peripheral adjacency range to obtain edge segments, closed contour fragments, interface turning points, and adjacent texture blocks, and then constructing a candidate region map using edge segments, closed contour fragments, interface turning points, and adjacent texture blocks; Subsequently, for each candidate region in the candidate region map, the orientation deviation, connectivity inconsistency, and visibility loss are calculated relative to the target state object region. The orientation deviation characterizes the degree of deviation between the candidate region and the spatial orientation constraint, the connectivity inconsistency characterizes the degree of inconsistency between the candidate region and the structural connectivity constraint, and the visibility loss characterizes the degree of absence between the candidate region and the visibility constraint. Then, the three are combined into a constraint violation cost, and the search expansion order is determined according to the constraint violation cost from low to high. Candidate regions that satisfy the spatial orientation constraint and structural connectivity constraint are expanded one by one to generate candidate surrounding regions. The output is an initial set of candidate surrounding regions sorted by constraint violation cost, which is written into the initial candidate region table for S2-23 to read. If the number of regions in the candidate region map is too large, resulting in excessive search overhead, pre-pruning is performed according to the region area ratio, boundary clarity, and adjacency with the target state object region to control computational resource consumption. If a region has a low constraint violation cost but significant visibility loss, it is retained as a candidate region to be corrected and is not directly used as a high-priority expansion region. The purpose of S2-23 is to iteratively update the confidence of the initial candidate surrounding region set and perform convergence screening to output a stable candidate surrounding region set. The inputs are the initial candidate surrounding region set, the load-bearing element constraint set, and the candidate region map. The processing actions include first calculating the degree of satisfaction between each initial candidate surrounding region and the load-bearing element constraint set. The degree of satisfaction is obtained by combining the degree of satisfaction of spatial orientation constraints, structural connection constraints, and visibility constraints. Then, the degree of satisfaction is converted into the confidence of the candidate surrounding region. Then, in the next iteration, the confidence level from the previous round is reused to prune the candidate region map. Priority is given to retaining candidate surrounding regions with high confidence rankings and stable matching relationships with the constraint set of carrying elements, while deleting candidate regions that have been ranked low for two consecutive rounds and lack matching relationships with the main carrying elements. After each iteration, the confidence ranking of the candidate surrounding regions is recalculated, and it is determined whether the confidence ranking remains unchanged for two consecutive iterations and whether the candidate set is no longer being added to or deleted from. When both conditions are met, the iteration stops and the candidate surrounding region set is output. The output is the candidate surrounding region set, which is written into the candidate surrounding region result table for use in S2-3 filtering and S3 identification. If the confidence ranking of all candidate surrounding regions fluctuates drastically during the iteration process, the process is backtracked to S2-22 to re-check whether there are boundary extraction errors in the candidate region map construction. If only some candidate regions exhibit ranking fluctuations, stable candidate regions are retained for subsequent steps, and fluctuating candidate regions are marked as regions to be reviewed. In this embodiment, object type identification can be performed on the state object region first, and then the state formation mechanism mapping relationship can be transformed into an executable set of carrying element constraints. Candidate surrounding regions are generated, sorted, and iteratively filtered within the outer adjacent range. Finally, a set of surrounding evidence regions that satisfy both spatial adjacency and structural connection relationships is output, thereby ensuring that subsequent surrounding change identification is based on evidence carrying regions with mechanistic basis, rather than on arbitrary neighboring regions, thereby improving the accuracy and verifiability of abnormal state determination. In practical applications: When the system identifies a filter as a certain state object area in an air conditioning unit image, it first calls the state formation mechanism mapping relationship corresponding to the filter to solve the bearing elements that usually form residual traces in the filter frame attachment zone, the air inlet side adjacent zone, and the air duct inlet adjacent zone when the filter is clogged for a long time. Then, it generates corresponding spatial orientation constraints, structural connection constraints, and visibility constraints for each bearing element. Subsequently, it extracts boundary and structural primitives in the adjacent area of ​​the filter area and constructs a candidate region map. It calculates the orientation deviation, connection inconsistency, and visibility loss for each candidate region and generates initial candidate surrounding regions item by item according to the constraint violation cost from low to high. Then, it retains candidate regions with stable matching relationships with the filter bearing elements through iterative confidence updates, and removes background regions that are close to the filter but lack border connection relationships in subsequent screening. Finally, it outputs a set of evidence regions around the filter for subsequent steps to continue to identify attachment changes and deposition distribution changes to determine whether the filter is clogged.

[0016] S3. Perform image recognition processing on each set of surrounding evidence regions to identify changes in attachment, wear, corrosion, connection or obstruction in the surrounding evidence regions, and merge them into the corresponding state object regions according to the adjacency relationship, and output the set of evidence of surrounding changes. In this embodiment, S3 is used to perform image recognition of residual traces on the already located surrounding evidence region set, transforming local visible changes in the surrounding evidence region into a set of surrounding change evidence that can be directly invoked for state determination. Its mechanism involves first performing image standardization processing on the surrounding evidence region to eliminate interference from differences in lighting, reflection, and blurring during image capture; then calculating the corresponding change amount for different types of residual traces to separate adhesion, wear, corrosion, connection anomalies, and obstruction anomalies from the image level; finally, merging is completed based on the spatial and structural adjacency relationships between the surrounding change candidates and the state object region, ensuring that the output retains both change type information and consistency with the corresponding state object region. This implementation process includes the following steps: The purpose of S3-1 is to eliminate the image performance deviation caused by the difference in shooting conditions between the surrounding evidence areas, so as to provide a unified input for subsequent change type identification. The input is the set of surrounding evidence areas and the object type result, area coordinates and structural adjacency identifier written by S2. The processing actions include performing brightness normalization, reflection highlight suppression and separation and sharpness compensation on each surrounding evidence area. Among them, brightness normalization is achieved by statistically analyzing the pixel brightness distribution in the current surrounding evidence area and performing linear stretching on the pixel brightness with the mean brightness and brightness dispersion of the area, so that the overall brightness of the area falls into a unified brightness range. The value of the unified brightness range is a statistical estimate. Highlight suppression and separation of reflective patches involves identifying clusters of bright pixels with sudden increases in brightness and abnormal decreases in saturation, separating them from the original image, and compensating them with the interpolation results of surrounding non-bright pixels to prevent reflective patches from being misidentified as wear or adhesion changes. Sharpness compensation calculates the gradient intensity of region edges and the degree of local blurring, performing deconvolution recovery or sharpening enhancement on edges in the motion blur direction to restore connection boundaries and texture details. The output is a standardized set of surrounding evidence regions, and the standardized results are written to a standardized region cache for S3-2 to read. If a surrounding evidence region has insufficient effective visible area due to occlusion, the region is marked as low-visibility and the existing visible portion is retained for subsequent identification. If the reflective region is too large and effective texture cannot be recovered after interpolation compensation, the region is marked as a region to be reviewed to avoid direct deletion and loss of evidence. The purpose of S3-2 is to extract changes that characterize residual traces from standardized peripheral evidence regions and generate corresponding peripheral change candidates. The inputs are the standardized peripheral evidence region set, object type results, and the bearing element identifier written by S2. The processing includes calculating, for each standardized peripheral evidence region, the texture density increment of attachment changes, the edge continuity breakage of wear changes, the color diffusion connectivity of corrosion changes, the structural alignment offset of connection changes, and the morphological passage gap of obstructed changes. The texture density increment is obtained by comparing the difference between the current region's texture unit distribution density and the reference texture density of a similar clean region. The reference texture density is determined jointly by the model output and historical sample statistical estimation. The edge continuity breakage is obtained by calculating the proportion of interrupted lengths of the main edge segments in the current region in the continuous direction. The color diffusion connectivity is obtained by performing a combined calculation on the area of ​​connected regions within the corrosion color gamut, the degree of edge diffusion, and the number of connected segments. The structural alignment offset is obtained by calculating the translational and angular deviations between the structural baselines on both sides of the connector. The morphological passage gap is obtained by calculating the proportion of the length of the obstructed portion and the number of continuous interrupted segments in the expected opening profile. Subsequently, each change quantity is compared with the corresponding change type discrimination condition. The source of the change type discrimination condition is jointly determined by rule constraints and model output. When a change quantity meets the corresponding change type discrimination condition, a corresponding surrounding change candidate is generated. The output quantity is the surrounding change candidate set, and the change type, change quantity value, carrying element identifier and regional coordinates of each surrounding change candidate are written into the change candidate index table for S3-3 to read. If the same region meets multiple change type discrimination conditions at the same time, multi-label change candidates are retained and the order of each change quantity is recorded for further differentiation in the subsequent adjacency merging stage. If the change quantity of a certain region is close to the discrimination boundary and cannot be stably classified, an uncertain mark is written and retained for subsequent structural adjacency verification. The purpose of S3-3 is to establish a stable attribution relationship between the surrounding change candidates and their corresponding state object regions, thereby forming a set of surrounding change evidence that can be directly used for state determination. The inputs are the surrounding change candidate set, the state object region set, the object type result, and the surrounding evidence region index table written by S2. The processing actions include calculating the regional distance and relative orientation between each surrounding change candidate and its corresponding state object region, and verifying its structural adjacency relationship. The regional distance is calculated by the distance from the center point of the surrounding change candidate to the nearest point of the boundary of the state object region. The relative orientation is calculated by the direction angle of the center point of the surrounding change candidate relative to the center point of the state object region. The structural adjacency relationship is determined by judging whether the surrounding change candidate is located on the bearing element position corresponding to the object type and whether there is a continuous boundary connection or shared adjacency structure with the target state object region. Then, the spatial distance is compared with the adjacency distance corresponding to the object type, the relative orientation is compared with the orientation corresponding to the object type, and the validity of the structural adjacency relationship is judged. The surrounding change candidates that simultaneously meet the conditions of spatial distance being within the adjacency range, relative orientation conforming to the orientation rules of the object type, and structural adjacency relationship being valid are retained and merged into the corresponding state object region. The output is the surrounding change evidence set, which is written into the evidence result table corresponding to the state object region for S4 to read. If a surrounding change candidate meets the adjacency condition with multiple state object regions, it is preferentially merged into the state object region with stronger structural adjacency relationship and closer spatial distance. If a surrounding change candidate only meets the spatial adjacency condition but not the structural adjacency relationship, it is determined to be a neighboring interference change and is removed to avoid mistakenly merging background dirt or irrelevant corrosion into the target state object region. In this embodiment, the surrounding evidence area can be standardized first, then the residual traces in the standardized area can be calculated and identified in terms of change amount and type. Finally, the effective surrounding change candidates can be stably assigned to the corresponding state object area through adjacency merging, thereby forming a set of surrounding change evidence that has both the ability to distinguish change types and the consistency of object attribution, providing direct input for subsequent object state determination. In practical applications: When the system processes the surrounding evidence area set corresponding to a certain valve, it first performs brightness normalization, reflection suppression, and sharpness compensation on the areas of valve body connection, limit contact, and adjacent actuator to eliminate interference from flashlight supplementary lighting and oblique shooting reflection; then it calculates the structural alignment offset of connection changes and the morphological passage gap of obstructed changes, and calculates the color diffusion connectivity of adjacent areas with rust marks to generate multiple surrounding change candidates; then it merges these change candidates based on the area spacing, relative orientation, and structural adjacency relationship between them and the valve state object area, removes background rust areas that are adjacent to the valve but do not belong to the valve connection structure, and finally outputs the surrounding change evidence set corresponding to the valve for subsequent steps to determine whether the valve is in a stuck state.

[0017] S4. Perform state determination processing on each set of surrounding change evidence according to the corresponding state object type. Based on the set of surrounding change evidence, determine the state of filter clogging, valve jamming, or electrical component aging, and output the object state result. In this embodiment, S4 is used to further transform the surrounding change evidence that has been merged into each state object region into object state results that can directly characterize the abnormal state of the object, thereby completing the determination from image change to device state. Its working mechanism is to first perform quantitative extraction on different change types in the surrounding change evidence set to form a change determination feature set, then call the corresponding state determination rule according to the object type to map the change determination feature set into state candidate results, and finally perform conflict verification on the state candidate results to eliminate candidate states that are locally valid but contradict other change evidence in the same area, and retain state results that satisfy the co-occurrence relationship and do not violate the exclusion relationship. To improve the stability of the state candidate results, the state determination rules are first decomposed into a set of rule constraints during the state candidate result generation process. Then, the support, conflict degree, and missing value are jointly solved for each state candidate category. Iterative correction is then used to eliminate the influence of single change evidence bias on the state determination. This implementation process includes the following steps: The purpose of S4-1 is to unify the different change evidence in the surrounding change evidence set into a change judgment feature set that can participate in state determination, thereby providing consistent input for subsequent rule matching. The input includes the surrounding change evidence set corresponding to each state object area, the object type result, and the change type identifier, change amount value, area coordinates, and bearing element identifier written by S3. The processing actions include calculating the change intensity, change distribution range, and change continuity for attachment change, wear change, corrosion change, connection change, and obstruction change in the surrounding change evidence set, respectively. The change intensity is used to characterize the significance of the corresponding change amount in the surrounding area of ​​the current state object area, and can be calculated by the deviation of each change amount value from the change amount of the same type of reference area. The source of the change amount of the reference area is jointly determined by statistical estimation and model output. The change distribution range is used to characterize the area proportion and the number of distribution segments of this type of change in the surrounding area of ​​the current state object area. The change continuity is used to characterize whether this type of change forms a continuous occurrence relationship between adjacent bearing elements, and can be calculated by the proportion of discontinuity length between adjacent change areas and the number of continuous segments. Subsequently, the intensity, distribution range, and continuity of change corresponding to each change type within the same state object area are summarized to form a change judgment feature set corresponding to that state object area. The output is the change judgment feature set, which is written into the state judgment feature table for S4-2 and S4-21 to read. If a certain change type is missing within the current state object area, the intensity, distribution range, and continuity of change corresponding to that change type are written into the missing flag instead of being directly set to zero, so as to avoid misjudging unobserved changes as non-existent. If there are multiple separate areas of the same change type, they are first merged according to the location of the carrying element before calculating the distribution range and continuity of change. The purpose of S4-2 is to map the change judgment feature set to the corresponding state candidate result based on the object type result, thereby establishing a direct correspondence between change evidence and abnormal state; the input is the object type result and the change judgment feature set; the processing action includes calling the state judgment rule corresponding to the object type identifier according to the object type result, where the filter corresponds to the dirt blockage state judgment rule, the valve corresponds to the jamming state judgment rule, and the electrical component corresponds to the aging state judgment rule. Subsequently, corresponding matching processing is performed on the change judgment feature sets of each state object area. For filters, the focus is on matching whether the adhesion changes and the distribution range of the changes form a continuous distribution in the filter frame adhesion strip, the air inlet side adjacent strip, and the air duct inlet adjacent strip to determine the candidate of the dirt blockage state. For valves, the focus is on matching whether the connection changes and obstruction changes appear together at the valve body connection, limit contact, and actuator adjacent strip to determine the candidate of the jamming state. For electrical components, the focus is on matching whether the rust changes, wear changes, and adhesion changes form a time-accumulated feature at the shell edge, terminal adjacent strip, and fastener adjacent strip to determine the candidate of the aging state. The output is the state candidate result for each state object area, and the state candidate result is written into the state candidate result table for S4-3 and S4-23 to read. If the change judgment feature set of a certain state object area only meets some of the state judgment rules, the intermediate matching result of the state object area is retained and marked as incomplete evidence for further judgment by subsequent conflict verification and iterative correction. The purpose of S4-3 is to remove candidate states from the candidate state results that conflict with other change evidence within the same state object area, thereby outputting a stable object state result. The inputs are each candidate state result, the change judgment feature set, and the rule constraint set written by S4-21. The processing actions include performing conflict checks on each change judgment feature within the same state object area to determine whether they simultaneously satisfy the co-occurrence relationship and the exclusion relationship in the state judgment rules. The co-occurrence relationship is used to limit the combination and distribution relationship of change types that should co-occur when a certain state is established. The exclusion relationship is used to limit the opposite change manifestations or change manifestations that should not co-occur with other states when a certain state is established. Then, each candidate state result is compared with its corresponding change judgment feature to see if it satisfies the co-occurrence relationship and does not violate the exclusion relationship. The candidate state results that satisfy the co-occurrence relationship and do not violate the exclusion relationship are retained as the object state results. The output is the object state results, and the object state results are written into the object state result table for S5 to read. If multiple candidate state results in the same object state region simultaneously satisfy part of the co-occurrence relationship, the candidate state results with higher overall consistency with the change judgment feature set are retained first. If all candidate state results do not satisfy the co-occurrence relationship, the object state region is marked as the object region to be reviewed and the evidence chain is retained for manual review or subsequent re-shooting. The purpose of S4-21 is to decompose the state determination rule into a set of rule constraints that can directly participate in the candidate solution, thereby transforming the object-oriented type determination logic into computable matching conditions. The inputs are the object type result and the change determination feature set. The processing actions include calling the corresponding state determination rule according to the object type identifier, and decomposing the state determination rule into feature satisfaction conditions, feature exclusion conditions, and feature co-occurrence conditions. The feature satisfaction conditions are used to limit the minimum combination of change features required for a certain state candidate category to be valid. The feature exclusion conditions are used to limit the combination of change features that would weaken or deny the validity of the state candidate category. The feature co-occurrence conditions are used to limit the collaborative appearance of multiple change features in terms of spatial location, distribution range, and continuity relationship. The decomposition process is executed according to rule constraints, which are derived from the state formation mechanism and historical sample statistical summarization corresponding to the object type. The output is the rule constraint set corresponding to each state object region, and the rule constraint set is written to the rule constraint cache for S4-22 and S4-3 to read. If the object type result has a low confidence mark, the auxiliary rule constraints of adjacent object types are retained when decomposing the state determination rules, but they are marked as secondary constraints to avoid the object type error directly causing the state determination distortion. The purpose of S4-22 is to perform candidate solving on the change judgment feature set of each state object region to obtain an initial state candidate result set sorted by priority relationship; the input is the change judgment feature set and the rule constraint set; the processing actions include calculating the support, conflict degree and missing degree of each change judgment feature to each state candidate category in the rule constraint set respectively, wherein the support is calculated by statistically analyzing the number of change judgment features that satisfy the feature satisfaction condition, the consistency of the distribution range and the consistency of the continuity relationship, the conflict degree is calculated by statistically analyzing the number of change judgment features that violate the feature exclusion condition and the conflict intensity, and the missing degree is calculated by statistically analyzing the number of feature items that are required to appear in the rule constraint set but are not detected in the current change judgment feature set and the importance of the missing position; Subsequently, according to the order of support from high to low, conflict from low to high, and missing value from low to high, a joint sorting is performed on each state candidate category. The state candidate categories with higher support ranking and simultaneously satisfying that their conflict and missing values ​​are no higher than those of the other state candidate categories in the corresponding ranking relationship are retained as the initial state candidate results. The output is the initial state candidate result set, which is written into the initial candidate result table for S4-23 to read. If the support, conflict, and missing value rankings of two state candidate categories are close, both state candidate categories are retained and marked as parallel candidates, and handed over to subsequent iterations for further differentiation. If a certain state candidate category has high support but also high missing value, the state candidate category is retained but marked as incomplete evidence to prevent the omission of the true state caused by occlusion. The purpose of S4-23 is to eliminate fluctuations in candidate ranking caused by missing local evidence, strong single changes, or interference from adjacent regions through iterative correction, thereby outputting stable state candidate results. The inputs are the initial state candidate result set, the change judgment feature set, and the surrounding change evidence adjacent to the current state object region. The processing actions include using the support, conflict, and missing values ​​in each initial state candidate result as confidence update inputs, and performing confidence correction on each initial state candidate result in combination with the co-occurrence consistency among the surrounding change evidence adjacent to the current state object region. The co-occurrence consistency is used to determine whether the current initial state candidate result is consistent with the change distribution pattern, continuity relationship, and structural position in the adjacent bearing elements. In each iteration, if the co-occurrence consistency between an initial state candidate result and adjacent change evidence increases, its ranking priority is increased; if a new conflict arises between it and adjacent change evidence, its ranking priority is decreased. After each iteration, the state candidate category ranking is regenerated, and it is determined whether the confidence ranking remains unchanged for two consecutive rounds and whether the state candidate category no longer changes. When both conditions are met, the iteration stops, and the corresponding initial state candidate result is determined as the state candidate result. The output is the state candidate result for each state object region, which is written back to the state candidate result table for S4-3 to read. If the ranking fluctuates continuously during the iteration process, the change judgment feature set is backtracked to check whether there are too many missing markers or conflict markers appearing in the set, and the corresponding state object region is marked as a region to be supplemented. If only a few candidate categories experience local fluctuations, the stable candidate categories are retained for subsequent conflict verification. In this embodiment, the surrounding change evidence set can be transformed into a change judgment feature set first, and then the corresponding state judgment rule can be called according to the object type to complete the state candidate solution. Through rule decomposition, joint sorting, iterative correction and conflict verification, the result is gradually converged to the object state result, so that the final output result is not only based on a single change evidence, but also on the co-occurrence relationship, exclusion relationship and adjacency consistency between multiple change features, thereby improving the stability and interpretability of the judgment of filter clogging, valve jamming and electrical component aging. In practical applications: When the system performs state determination on the surrounding change evidence set corresponding to a certain electrical component, it first calculates the intensity, distribution range, and continuity of changes in the corrosion changes at the shell edge, the adhesion changes at the terminal connections, and the wear changes at the fasteners, forming a change determination feature set for the electrical component. Then, it calls the aging state determination rule corresponding to the electrical component, decomposes the rule into feature satisfaction conditions, feature exclusion conditions, and feature co-occurrence conditions, further calculates the support, conflict, and missing values ​​of the aging state candidates for the above change features, and generates initial state candidate results. Subsequently, iterative correction is performed on the initial state candidate results based on the co-occurrence consistency among the adjacent surrounding change evidence. After two consecutive rounds of sorting without change, the aging state candidate results are output. Then, the candidate results that contradict the features of locally newly replaced components are eliminated through conflict checking. Finally, the object state result of the electrical component is output for subsequent equipment collection and processing.

[0018] S5. Perform equipment aggregation processing on the status results of each object, summarize the status results of objects belonging to the same target HVAC equipment, form the survey status results corresponding to the target HVAC equipment, and output the operating status identification results of the target HVAC equipment based on the survey status results. In this embodiment, S5 is used to aggregate the object status results output from each state object area to the device level, forming an investigation status result that can characterize the overall investigation conclusion of the target HVAC equipment, and further mapping the investigation status result into an operational status identification result that can be directly used for subsequent business processing; its working mechanism is as follows: first, based on the location information and object type attribution of the state object area, the object status results are merged into device status units within the device; then, co-occurrence relationships, conflict relationships, and supplementary relationships are calculated between device status units; through combination and reconstruction, the instability caused by local misjudgment and lack of evidence is eliminated; finally, based on the combination of reconstructed device status units, the operational status category of the target HVAC equipment is determined, so that the output result is not dependent on a single object status but is based on the structural consistency of multiple object states within the device; this implementation process includes the following steps: The purpose of S5-1 is to organize the scattered object status results into a set of composable and comparable device status units within the device, thereby providing a unified structure for subsequent association and aggregation. The input quantities are the status results of each object and the location information of the corresponding status object area. The object status results are written into the object status result table by S4, and the location information is written into the object area result table by S1, including at least the area coordinates of the status object area within the device area, the relative device orientation, and the adjacent structure identifier. The processing actions include classifying each object status result according to the location attribution relationship within the device area. The location attribution relationship is determined based on whether the center point of the status object area falls into the device area and whether its relative device orientation satisfies the corresponding installation location relationship. The source of the installation location relationship is rule constraints. Next, the status results of each object are categorized according to the object type attribution relationship. The status results of objects of the same object type and located in the same installation location partition are merged into one device status unit. The device status unit records the object type identifier, installation location partition identifier, object status category, and the surrounding change evidence summary information corresponding to the object status result. The output is a set of device status units, which is written into the device status unit table for S5-2 to read. If there is a missing object status result but a corresponding status object area exists in the installation location partition, a missing status unit is generated and a missing mark is written. The partition is retained for subsequent supplementary relationship calculation. If multiple object status results of the same type appear in the same installation location partition and the status categories are inconsistent, each object status result is retained and a multi-source conflict mark is written. The result is then handed over to S5-2 for processing in the conflict relationship calculation. The purpose of S5-2 is to calculate and perform combined reconstruction of the relationships between equipment state units at the equipment level, thereby converging local results into a verifiable set of survey state results. The inputs are the set of equipment state units and the object type results and surrounding change evidence summary information written by S4. The processing actions include calculating the co-occurrence relationships, conflict relationships, and complementary relationships between each equipment state unit. The co-occurrence relationship is used to describe the combination of two or more state units that appear simultaneously in the same equipment and are not structurally contradictory. For example, the co-occurrence of filter clogging and changes in air duct inlet adhesion, valve jamming and connection changes, and aging of electrical components and corrosion of wiring terminals. The source of co-occurrence relationships is determined jointly by rule constraints and historical sample statistical estimation; conflict relationships are used to describe the phenomenon that two state units will mutually negate each other when they are simultaneously established at the same installation location or on the same structural link. For example, the same valve zone may simultaneously exhibit a stuck state and a normally aligned state. The source of conflict relationships is rule constraints; supplementary relationships are used to describe the relationship that can be corroborated by state units on adjacent installation location zones or adjacent structural links when a certain state unit is missing. For example, when a filter area is missing, supplementary evidence can be provided by changes in the adhesion of the adjacent strip on the air inlet side. The source of supplementary relationships is determined jointly by rule constraints and model output. Subsequently, based on co-occurrence, conflict, and supplementary relationships, the equipment state units are combined and reconstructed. This includes merging state units satisfying co-occurrence relationships into the same exploration state branch, splitting state units satisfying conflict relationships into mutually exclusive branches and retaining the branch with more complete evidence, and completing missing state units satisfying supplementary relationships using the lateral evidence information of the supplementary state units and writing it into a completion marker. The output is an exploration state result set, which is written into the exploration state result table for S5-3 to read. If a conflict relationship prevents multiple exploration state branches from being deleted, the results of multiple branches are retained and marked as pending review for manual review or subsequent re-shooting. If the supplementary relationship is insufficient to complete the completion, a missing marker is retained, and the missing object type and missing installation location partition are recorded in the exploration state results. The purpose of S5-3 is to map the survey status result set to the operation status identification result, thereby outputting equipment-level status categories that can be directly used for subsequent operation analysis and strategy generation. The input is the survey status result set and the operation status mapping rule corresponding to the operation status category. The source of the operation status mapping rule is jointly determined by rule constraints and historical sample statistical estimation. The processing actions include extracting the equipment status unit combination features contained in each survey status branch in the survey status result set. The equipment status unit combination features include at least the object type coverage, status category combination, co-occurrence relationship status, conflict relationship handling results, and completion mark distribution. Subsequently, based on the operational status mapping rules, the combined features are mapped to the operational status categories corresponding to the target HVAC equipment. When there are multiple survey status branches, the operational status category with higher evidence completeness and fewer conflict markers is output first. The output is the operational status identification result of the target HVAC equipment, and the operational status identification result is written into the equipment operational status result table for subsequent steps to read. If all survey status branches have a pending review marker, the operational status identification result is output along with a review prompt message, indicating the object type or installation location partition that needs to be re-photographed. If the operational status mapping rules cannot cover the current combined features, the uncovered markers are output while retaining the complete evidence chain of the survey status branches. In this embodiment, the object status results can be formed into a set of equipment status units according to the location and object type attribution relationships. Then, through co-occurrence, conflict and supplementary relationships, combined reconstruction is performed to converge the scattered, possibly incomplete or conflicting object status results into a set of equipment-level survey status results. Furthermore, the set of survey status results is mapped to the operation status identification results, thereby improving the stability, verifiability and directness of the equipment-level output conclusions for subsequent business processing. In practical applications: After the system determines the status of an air conditioning unit based on its filter clogging, valve jamming, and electrical component aging, it first categorizes each status object area into return air side filter zone, pipeline valve zone, and electrical control box zone according to their relative positions within the equipment area, forming corresponding equipment status unit sets. Then, it merges evidence chains based on the co-occurrence relationships of filter clogging and inlet side adhesion changes, and valve jamming and connection offset. Simultaneously, it splits parallel candidates of jamming and normal operation within the same valve zone through conflict relationships, retaining the more complete evidence branch. For missing panel zones, it retains the missing marker through supplementary relationships. Finally, it maps the combined equipment status unit features to the target equipment's operating status category, such as high resistance operating status or abnormal control response operating status, and outputs the operating status identification result to the subsequent energy-saving strategy generation process.

[0019] The working principle of this solution is as follows: First, locate the target HVAC equipment from the on-site images. Then, locate the status objects such as filters, valves, dials, and electrical components inside the equipment. Instead of directly judging whether these components themselves are abnormal, it further looks for related evidence areas in the surrounding area, identifying changes such as dust accumulation, wear, corrosion, connection misalignment, and obstruction marks, and categorizing these changes to the corresponding components. Finally, based on these surrounding changes, it determines whether the filter is clogged, the valve is stuck, and the electrical components are aged. The results of each component are then summarized into the overall inspection and operating status of the entire equipment. The core idea is that many anomalies may not be fully manifested on the component itself, but will leave stable traces in the surrounding area. This solution uses these traces to improve the accuracy of identification. For example, when staff take a picture of an air conditioning unit in a shopping mall's air conditioning room, the system first identifies the unit's boundaries, then locates the filter, valves, and electrical control box. Next, the system determines whether the filter is clogged due to long-term dirt accumulation by examining the filter frame and the dust accumulation near the air inlet; it determines whether the valve is stuck by observing changes in valve connections and limit positions; and it determines whether electrical components are aging by observing rust and wear around the wiring terminals and fasteners. After all these results are obtained, the system comprehensively judges whether the equipment is currently operating with high resistance, experiencing control abnormalities, or aging, and directly provides the results for subsequent energy-saving analysis.

[0020] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An image processing method for surveying HVAC equipment, characterized in that, include: S1. Acquire on-site images of the target HVAC equipment, perform equipment positioning processing on the on-site images, determine the equipment area corresponding to the target HVAC equipment, extract the status object area within the equipment area, and output the status object area set. The state object region represents the process of performing object validity determination on the candidate object region set, retaining candidate object regions that meet the object shape conditions and positional relationship conditions; S2. Perform object type identification processing on each state object region in the state object region set, determine the state object type corresponding to each state object region, and locate the surrounding evidence region carrying residual traces within the outer adjacent range according to the state formation mechanism corresponding to each state object type, and output the surrounding evidence region set. S3. Perform image recognition processing on each set of surrounding evidence regions to identify changes in attachment, wear, corrosion, connection, or obstruction in the surrounding evidence regions, and merge them into the corresponding state object regions according to their adjacency relationship, and output the set of evidence of surrounding changes. S4. Perform state determination processing on each set of surrounding change evidence according to the corresponding state object type. Based on the set of surrounding change evidence, determine the state of filter clogging, valve jamming, or electrical component aging, and output the object state result. S4 includes: S4-1. Obtain the surrounding change evidence set and object type results corresponding to each state object area. Calculate the change intensity, change distribution range and change continuity for each surrounding change evidence set, including attachment change, wear change, corrosion change, connection change or obstruction change. Output the change judgment feature set corresponding to each state object area. S4-2. Based on the object type result, call the state determination rule corresponding to the object type identifier, perform dirt and blockage matching processing on the change determination feature set corresponding to the filter, perform jamming matching processing on the change determination feature set corresponding to the valve, perform aging matching processing on the change determination feature set corresponding to the electrical components, and output the state candidate results corresponding to each state object area. S4-3. Perform conflict verification processing on each state candidate result, determine whether the change judgment features in the same state object area meet the preset co-occurrence condition and exclusion condition, retain the state candidate result that meets the preset co-occurrence condition and does not meet the exclusion condition as the object state result, and output the object state result. S5. Perform equipment aggregation processing on the status results of each object, summarize the status results of objects belonging to the same target HVAC equipment, form the survey status results corresponding to the target HVAC equipment, and output the operating status identification results of the target HVAC equipment based on the survey status results.

2. The image processing method for HVAC equipment surveying according to claim 1, characterized in that: S1 includes: S1-1. Perform equipment contour recognition and structural boundary recognition processing on the on-site image to determine the outer contour range and internal structure distribution range of the target HVAC equipment in the on-site image, and output the equipment boundary results. S1-2. Perform region constraint decomposition processing on the equipment boundary results. According to the preset state objects in the target HVAC equipment, decompose the equipment boundary results into multiple candidate object regions and output the candidate object region set. S1-3. Perform object validity determination processing on the candidate object region set, retain the candidate object regions that meet the object shape conditions and positional relationship conditions as state object regions, and output the state object region set.

3. The image processing method for HVAC equipment surveying according to claim 2, characterized in that: S2 includes: S2-1. Obtain the target state object region from the state object region set, perform object type recognition processing on the target state object region, generate an object type identifier corresponding to the target state object region, and output the object type result. S2-2. Based on the object type result, call the state formation mechanism mapping relationship that corresponds one-to-one with the object type identifier, solve the set of residual trace carrying elements corresponding to the object type identifier, and generate candidate surrounding areas within the outer adjacent range of the target state object area with each residual trace carrying element as a constraint, and output the set of candidate surrounding areas. S2-3. Perform region filtering processing on each candidate surrounding region in the candidate surrounding region set, calculate the region spacing, relative orientation and boundary connection relationship between each candidate surrounding region and the target state object region, retain the candidate surrounding regions whose region spacing falls within the preset adjacency distance range, whose relative orientation conforms to the preset orientation rule corresponding to the object type and whose boundary connection relationship conforms to the structural connection rule corresponding to the object type as the surrounding evidence region, and output the surrounding evidence region set.

4. The image processing method for HVAC equipment surveying according to claim 3, characterized in that: The process of outputting the candidate surrounding region set in S2-2 also includes: S2-21. Obtain the object type result and read the state formation mechanism mapping relationship that corresponds one-to-one with the object type identifier. Perform element calculation on the state formation mechanism mapping relationship to obtain the set of residual trace bearing elements. Generate a constraint set containing spatial orientation constraints, structural connection constraints and visibility constraints for each bearing element in the set of residual trace bearing elements. Output the bearing element constraint set. S2-22. Obtain the target state object region and limit its peripheral adjacent range. Perform boundary and structural primitive extraction on the image regions within the peripheral adjacent range to form a candidate region map. Calculate the orientation deviation, connectivity inconsistency, and visibility loss of each candidate region in the candidate region map relative to the target state object region to form the constraint violation cost. When performing constrained search on the candidate region map, determine the search expansion order from low to high constraint violation cost and generate candidate surrounding regions item by item under the premise of satisfying spatial orientation constraints and structural connectivity constraints. Output the initial candidate surrounding region set sorted by constraint violation cost. S2-23. Perform iterative confidence update processing on the initial candidate surrounding area set, convert the degree of satisfaction between each initial candidate surrounding area and the constraint set of carrying elements into confidence, and reuse the confidence of the previous round to prune the candidate area map in the iteration to control the consumption of computing resources. When the confidence ranking remains unchanged for two consecutive rounds of iteration and the candidate set is no longer added or deleted, output the candidate surrounding area set.

5. The image processing method for HVAC equipment surveying according to claim 4, characterized in that: S3 includes: S3-1. Obtain the set of surrounding evidence regions and perform image standardization processing on each of the surrounding evidence regions, including brightness normalization for uneven illumination, suppression and separation for reflective highlights, and sharpness compensation for motion blur, and output the standardized set of surrounding evidence regions. S3-2. Perform change type identification processing on each peripheral evidence region in the standardized peripheral evidence region set, including calculating the texture density increment of attachment change, the edge continuity break of wear change, the color diffusion connectivity of corrosion change, the structural alignment offset of connection change, and the morphological passage gap of obstruction change, and output the peripheral change candidate set according to the relationship between each change quantity and the corresponding change type discrimination condition.

6. The image processing method for HVAC equipment surveying according to claim 5, characterized in that: S3 also includes: S3-3. Perform adjacency merging processing on the candidate set of surrounding changes, including calculating the regional distance and relative orientation between each candidate of surrounding changes and its corresponding state object region and verifying its structural adjacency relationship. Retain the candidates of surrounding changes that simultaneously meet the requirements of regional distance within the adjacency range, relative orientation conforming to the orientation rules of object type and structural adjacency relationship and merge them into the corresponding state object region, and output the evidence set of surrounding changes.

7. The image processing method for HVAC equipment surveying according to claim 6, characterized in that: The process of outputting the state candidate results for each state object region in S4-2 also includes: S4-21. Obtain the object type result and change judgment feature set, call the corresponding state judgment rule according to the object type identifier, and decompose the state judgment rule into feature satisfaction condition, feature exclusion condition and feature co-occurrence condition, and output the rule constraint set corresponding to each state object region; S4-22. Perform candidate solution processing on the change judgment feature set of each state object region. Calculate the support, conflict degree and missing degree of each change judgment feature for each state candidate category in the rule constraint set. Sort each state candidate category in the order of support from high to low, conflict degree from low to high and missing degree from low to high. Retain the state candidate categories with the highest support ranking and which simultaneously satisfy the conditions that the conflict degree and missing degree are not higher than the other state candidate categories in the corresponding ranking relationship as the initial state candidate results. Output the initial state candidate result set.

8. The image processing method for HVAC equipment surveying according to claim 7, characterized in that: The process of outputting the state candidate results for each state object region in S4-2 also includes: S4-23. Perform iterative correction processing on the initial state candidate result set, and use the support, conflict degree and missing degree in each initial state candidate result as confidence update input; In addition, confidence correction is performed on each initial state candidate result based on the co-occurrence consistency of the surrounding change evidence adjacent to the current state object region. The initial state candidate results that remain unchanged in confidence ranking for two consecutive rounds and whose state candidate categories no longer change are retained as state candidate results, and the corresponding state candidate results for each state object region are output.

9. The image processing method for HVAC equipment surveying according to claim 8, characterized in that: S5 includes: S5-1. Obtain the status results of each object and its corresponding status object area location information. Perform intra-device classification processing on the status results of each object according to the location belonging relationship and object type belonging relationship within the device area, and output the device status unit set. S5-2. Perform association and aggregation processing on the equipment status unit set, calculate the co-occurrence relationship, conflict relationship and supplementary relationship between each equipment status unit, and perform combination reconstruction on each equipment status unit according to the co-occurrence relationship, conflict relationship and supplementary relationship, and output the exploration status result set. S5-3. Perform operational status identification processing on the survey status result set, determine the operational status category corresponding to the target HVAC equipment based on the combination results of each equipment status unit in the survey status result set, and output the operational status identification result of the target HVAC equipment.

Images