Construction machine cleaning control method

Abstract

The present application relates to the technical field of construction machinery cleaning equipment, and discloses a construction machinery cleaning control method. The method solves the problem that the cleaning equipment is difficult to automatically adapt to different types and sizes of construction machinery and cannot automatically determine the key cleaning positions and the priority order. A template library is provided to store type templates of construction machinery types, record the names of a plurality of predetermined positions and the dirt adhesion tendency grades, obtain an appearance image of a construction machinery to be cleaned, determine the construction machinery type and divide the surface area into a plurality of position areas, retrieve the corresponding type template from the template library according to the determined type, establish a corresponding relationship between the predetermined positions in the template and the divided position areas, sort the position areas according to the dirt adhesion tendency grades, and generate a cleaning position list of the cleaning priority order. The method can realize automatic identification and priority sorting of the cleaning positions of construction machinery such as excavators, rotary drilling rigs and loaders, and improve the cleaning efficiency and adaptability.

Description

Technical Field

[0001] This invention relates to the field of engineering machinery cleaning equipment technology, and specifically to a method for controlling the cleaning of construction machinery. Background Technology

[0002] Construction machinery cleaning encompasses a wide range of engineering equipment, including excavators, bulldozers, loaders, rotary drilling rigs, cranes, concrete pump trucks, and road rollers. After operating at construction sites such as tunnels, foundation pits, mines, and roads, these machines accumulate a large amount of dirt, slurry, concrete residue, oil, and other grime on their surfaces, requiring regular cleaning. However, construction machinery varies greatly in type and size, ranging from mini excavators weighing less than two tons to large rotary drilling rigs weighing tens or even hundreds of tons. Their external structures are completely different, and they generally possess complex surface features such as uneven surfaces, articulated structures, gaps, and pipelines. This makes it difficult for cleaning equipment to automatically determine which parts should be prioritized for cleaning and the cleaning order for each part when faced with different construction machinery.

[0003] Currently, construction machinery cleaning mainly relies on manual high-pressure water guns, with operators visually identifying areas of concentrated dirt and determining the cleaning sequence based on experience. This method is not only labor-intensive and time-consuming, but the cleaning quality is also heavily influenced by the operator's experience, making cleaning high places and concealed areas particularly difficult. Some construction sites are equipped with fixed cleaning platforms or channels that use pre-installed spray units to wash the chassis and tires of construction vehicles. However, these devices are only suitable for vehicles with relatively regular shapes and cannot handle large, irregularly shaped equipment such as excavators and rotary drilling rigs. Furthermore, they cannot automatically identify and mark key cleaning areas based on the specific structural characteristics of the machinery. Other mobile cleaning devices have simple spraying functions, using a U-shaped frame to rotate or multi-directional nozzles for cleaning. However, their cleaning paths and control parameters are usually preset, lacking adaptability to different types of construction machinery and specific parts. When faced with various construction machinery with significantly different shapes, operators still need to manually adjust the cleaning position and direction.

[0004] Chinese patent document CN1213887C discloses a cleaning and recycling device for on-site cleaning of heavy machinery. This device includes a support frame, a detergent storage tank, and a washing hose. It can be moved to the site of heavy machinery to spray and clean it, and a lifting device supports the operator to work around the heavy machinery. While this device can achieve on-site cleaning of heavy machinery, the determination of cleaning areas and the cleaning operation are all performed by the operator. The device itself does not have the ability to automatically identify the type of heavy machinery and its functional parts, nor does it address how to automatically determine the key cleaning areas and their priority. In the prior art, there are also technical solutions based on visual detection for cleaning trajectory planning. For example, images and point cloud data of parts are acquired using 2D and 3D cameras, the location of dirt on the part surface is identified and mapped into a 3D point cloud, and then the robot's cleaning trajectory is planned. This type of technical solution is aimed at small curved parts, and its technical approach is based on real-time visual detection of dirt to drive the generation of the cleaning path; that is, cleaning only where dirt is detected. However, for large and complex objects like construction machinery, the aforementioned technical solutions do not address the identification of the overall machine type or the semantic segmentation of individual functional parts. Construction machinery surfaces are generally covered in mud, and the visual differences between different functional parts may be obscured by the mud. Relying solely on real-time dirt detection is insufficient to accurately identify the structural affiliation and cleaning needs of each part under mud-covered conditions.

[0005] Furthermore, existing cleaning equipment lacks a comprehensive technical solution that integrates the identification of construction machinery types, the division of functional parts, the marking of key cleaning areas, and the prioritization of cleaning. Current sensing technologies (such as LiDAR and cameras) are primarily used for environmental perception and autonomous navigation in the field of construction machinery cleaning, rather than for a structured understanding of the functional parts of the machinery being cleaned. This means that existing mobile cleaning equipment still requires manual intervention to judge and set cleaning parameters when dealing with different types of construction machinery on-site, preventing truly automated and intelligent cleaning operations. Summary of the Invention

[0006] One objective of this invention is to address the following problems: Construction machinery comes in a wide variety of types, with vastly different dimensions and irregular surface structures. Existing cleaning equipment cannot automatically identify the type and functional parts of different construction machines, nor can it automatically determine which parts should be prioritized for cleaning or the cleaning order for each part. Manually determining cleaning areas is labor-intensive, inefficient, and relies heavily on operational experience. Fixed cleaning stations are only suitable for vehicles with regular shapes, and mobile cleaning devices lack adaptability to different construction machines. Existing trajectory planning schemes based on real-time dirt detection cannot predict which areas should be prioritized for cleaning when covered by mud or before dirt is visible.

[0007] To achieve the above objectives, the present invention provides a method for controlling the cleaning of construction machinery, comprising the following steps: A template library is provided, which stores type templates for at least two types of construction machinery. Each type template records the names of multiple predetermined parts corresponding to the type of construction machinery and the dirt adhesion tendency level of each predetermined part. An image of the construction machinery to be cleaned is acquired by an image acquisition device. Through image data processing, the type of construction machinery to be cleaned is determined, and the surface area of ​​the construction machinery to be cleaned in the image is divided into multiple parts. Based on the determined type of construction machinery, the corresponding type of template is retrieved from the template library, and a corresponding relationship is established between the predetermined parts in the type of template and the multiple part areas. Based on the dirt adhesion tendency level recorded in the type template, the corresponding regions are sorted to generate a list of key cleaning areas and their cleaning priority order for the construction machinery to be cleaned.

[0008] Preferably, in the type template, each predetermined part also records the surface structure type of the predetermined part; when generating the list of cleaning parts, for at least one part area in the list, the cleaning intensity parameter of the part area is jointly determined according to the surface structure type of the predetermined part corresponding to the part area and the dirt adhesion tendency level of the predetermined part, and the cleaning intensity parameter is associated with the corresponding part area in the list of cleaning parts; wherein, the cleaning intensity parameter includes at least one of the working pressure of the high-pressure water jet, the nozzle moving speed, and the number of cleaning passes.

[0009] Preferably, for the at least one area, a baseline cleaning intensity parameter is determined based on the dirt adhesion tendency level of the area, and an intensity limit threshold is determined based on the surface structure type of the area; the baseline cleaning intensity parameter is compared with the intensity limit threshold: if the baseline cleaning intensity parameter does not exceed the intensity limit threshold, the baseline cleaning intensity parameter is used as the actual cleaning intensity parameter for the area; if the baseline cleaning intensity parameter exceeds the intensity limit threshold, the intensity limit threshold is used as the actual cleaning intensity parameter for the area, and the cleaning operation time for the area is extended compared to the case where it does not exceed the threshold; wherein, the extension of the cleaning operation time is achieved by at least one of the following methods: reducing the moving speed of the high-pressure water jet nozzle and increasing the number of cleaning passes.

[0010] Preferably, after establishing a correspondence between the predetermined parts in the type template and the plurality of part regions, the method further includes: if there is a part region among the plurality of part regions that is not corresponding to any predetermined part, marking it as an unmatched region; obtaining the visual feature information of the unmatched region, and obtaining the visual feature information of at least one part region that has established a correspondence with the predetermined part; the visual feature information includes at least one of the following: color features, texture features, and material reflection characteristics; comparing the visual feature information of the unmatched region with the visual feature information of the at least one part region, and determining whether the unmatched region should be included in the cleaning part list based on the comparison result; when the unmatched region is marked as an unmatched region and the visual feature information is consistent in multiple consecutive frames of appearance images, attempting to merge the unmatched region with a spatially adjacent already corresponding part region; if the merged region can establish a correspondence with a predetermined part in the type template, then it is included in the cleaning part list; if the merged region still cannot establish a correspondence with any predetermined part, then it is excluded from the automatic cleaning range, and a prompt message is generated for the operator to confirm.

[0011] Preferably, the at least one part region is two or more part regions that have established a correspondence with a predetermined part, and when determining whether the unmatched region should be included in the cleaning part list based on the comparison result: the spatial positional relationship between the unmatched region and the two or more part regions is obtained; for each of the two or more part regions, the corresponding part region referenced by the unmatched region is determined from the two or more part regions by combining its visual feature similarity and spatial positional relationship with the unmatched region; and the unmatched region is determined to be included in the cleaning part list based on the information of the referenced corresponding part region in the cleaning part list.

[0012] Preferably, the spatial positional relationship includes at least one of the following: the spatial distance, adjacency relationship, and relative orientation between the unmatched region and the two or more part regions respectively; and, the step of determining the corresponding part region referenced by the unmatched region from the two or more part regions by combining its visual feature similarity with the unmatched region and its spatial positional relationship includes: when there is a part region among the two or more part regions, the visual feature similarity between the part region and the unmatched region meets a preset condition, and the part region and the unmatched region meet a preset positional relationship in space, the part region is determined as the corresponding part region referenced by the unmatched region.

[0013] Preferably, the image acquisition device includes a first type of imaging device and a second type of imaging device, which are based on different physical imaging principles. The first type of imaging device acquires the three-dimensional geometric information of the construction machinery to be cleaned; the second type of imaging device acquires the two-dimensional texture information or depth information of the construction machinery to be cleaned. Determining the type of construction machinery to be cleaned and dividing the surface area of ​​the construction machinery to be cleaned in the appearance image into multiple part regions includes: generating a preliminary type identification result and a preliminary part region division result of the construction machinery to be cleaned based on the three-dimensional geometric information; and correcting the preliminary type identification result and the preliminary part region division result by combining the two-dimensional texture information or depth information to obtain the type of construction machinery to be cleaned and the multiple part regions.

[0014] Preferably, the type templates in the template library are organized in a hierarchical structure, which includes at least a type layer and a specification layer. The type layer records the names of predetermined parts common to the corresponding construction machinery type, as well as the dirt adhesion tendency level of each predetermined part. The specification layer records adjustment information for the dirt adhesion tendency level of at least one predetermined part in the type layer under a specific specification variant of the corresponding construction machinery type, and / or supplementary predetermined parts and their dirt adhesion tendency levels. When retrieving the corresponding type template from the template library based on the determined construction machinery type, the information of the type layer and the specification layer are merged according to the specification variant of the construction machinery to be cleaned to generate an applicable type template instance.

[0015] The present invention has at least the following beneficial effects: The construction machinery cleaning control method provided by this invention matches pre-stored type templates in a template library with image recognition results, enabling the cleaning system to automatically identify various functional parts and determine key cleaning areas and their priority order based on the specific type and specifications of the construction machinery. This method reduces the reliance on operator experience in cleaning operations, eliminating the need for operators to manually judge and set cleaning area parameters each time they encounter different construction machinery, thus reducing the impact of manual operation intensity and human judgment differences on cleaning quality.

[0016] By incorporating surface structure type into the determination of cleaning intensity parameters, the cleaning system can adopt differentiated cleaning strategies for areas with different structural characteristics. Standard cleaning methods can be used for continuous curved surfaces, while water pressure can be appropriately reduced and cleaning time increased for hinged gaps. For grid-like areas, the spray angle and pressure can be adjusted to avoid damaging subsequent components. This method of adapting cleaning intensity based on the structural characteristics of the area helps to remove dirt while minimizing the risk of damage to sensitive structures.

[0017] When the need for decontamination conflicts with structural protection, a mechanism compares the baseline cleaning intensity parameters with the intensity limit threshold, and compensates for the cleaning energy loss caused by the intensity limitation through time compensation. This ensures that sensitive parts such as hinge gaps, guide rail surfaces, and radiator grilles receive sufficient cleaning time within a safe pressure range. This mechanism provides a certain degree of protection for the structural safety of different functional parts of construction machinery during the cleaning process.

[0018] By identifying unmatched regions and comparing them using visual feature similarity, the system can automatically distinguish areas in the image segmentation results that cannot correspond to the template location. Areas that are incompletely segmented due to occlusion or soil cover and require cleaning can be included in the cleaning range; non-original accessories and other non-cleaning targets can be excluded from the automatic cleaning range, reducing ineffective rinsing. Introducing spatial relationships as an auxiliary judgment dimension, even when the visual features of multiple locations are homogenized due to soil cover, the system can utilize the relatively fixed spatial structure patterns between the functional parts of the construction machinery to differentiate them, improving the accuracy of identifying unmatched regions.

[0019] Two types of imaging devices based on different physical principles work together to generate preliminary identification and segmentation results with three-dimensional geometric information as a guide, and then correct them with two-dimensional texture or depth information. This allows for relatively stable identification of construction machinery types and regional division of functional parts even under common adverse conditions in construction sites, such as insufficient light inside the tunnel, strong outdoor light and shadow, and surface soil coverage.

[0020] The type template library is organized in a hierarchical structure of type and specification layers. The type layer records information on common parts of machines of the same type, while the specification layer records adjustment and supplementary information for specific specification variations. These two layers are merged upon call to generate an applicable template instance. This organization allows the template library to cover different specification variations of the same type of machine. When new models need to be added, this can be achieved by adding specification layer records or inheriting part information from existing type templates, thus reducing the maintenance workload of the template library.

[0021] The method of this invention forms a relatively complete construction machinery cleaning control process, from template library construction, image acquisition and part recognition, part correspondence, to cleaning part list generation and cleaning parameter determination, providing a highly automated control scheme for on-site cleaning of construction machinery at construction sites.

[0022] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0023] Figure 1This is a flowchart of the construction machinery cleaning control method of the present invention; Figure 2 This is a schematic diagram of the process for determining the cleaning intensity parameters of the present invention; Figure 3 This is a schematic diagram of the unmatched region processing logic of the present invention; Figure 4 This is a schematic diagram showing the division of the construction machinery parts in this invention; Figure 5 This is a schematic diagram of the template library hierarchy of the present invention. Detailed Implementation

[0024] The present invention will now be described in further detail with reference to examples, so that those skilled in the art can implement it based on the description.

[0025] (1) such as Figure 1 As shown, in one embodiment of the construction machinery cleaning control method provided by the present invention, a template library is first established. This template library can be deployed in the vehicle-mounted control unit of the mobile cleaning system, or it can be deployed on a remote server and communicate with the on-site cleaning equipment via a wireless network. The template library stores type templates for at least two types of construction machinery, such as type templates for common construction machinery like excavators, loaders, rotary drilling rigs, and bulldozers. Each type template records the names of multiple predetermined parts corresponding to the construction machinery type, as well as the dirt adhesion tendency level corresponding to each predetermined part. Taking the excavator type template as an example, the predetermined parts recorded may include the bucket, boom, stick, cab, slewing platform, and tracked walking device. The dirt adhesion tendency level corresponding to the bucket can be set to high, the boom joint and stick joint can be set to high, the top of the cab and the slewing platform can be set to medium, and the tracked walking device can be set to high. The dirt adhesion tendency level can be divided into three levels: high, medium, and low, or it can be represented by a numerical level of 1 to 5. The data in the template library can be pre-compiled and stored in the form of a data table or a structured file, and entered before the cleaning operation.

[0026] During cleaning operations, an image acquisition device captures images of the construction machinery to be cleaned. This device can be a LiDAR (Light Detection and Ranging) unit, mounted on the front or top support of the mobile cleaning vehicle. The scanning distance can be set between 10 and 50 meters to obtain 3D point cloud data of the machinery. Alternatively, a binocular vision camera or depth camera can be used as a supplement or replacement. The binocular vision camera can be mounted at a height of approximately 2 to 3 meters above the ground at the front of the vehicle. Image data processing determines the type of construction machinery to be cleaned and divides its surface area into multiple regions. Image data processing can employ point cloud semantic segmentation algorithms to identify the type of construction machinery based on 3D point cloud data and divide the point cloud regions corresponding to each functional part; alternatively, it can combine texture and color information from 2D images to assist in the judgment. The determination of the construction machinery type and the division of each region are completed in the onboard control unit, with a processing time typically within 1 to 3 minutes. Compared to this implementation method, existing cleaning equipment typically does not identify the type of construction machinery or segment the parts. Instead, the operator visually determines the cleaning area or simply rinses based on preset fixed nozzle positions. This makes it impossible to automatically distinguish the functional parts for the specific structure of different construction machinery, resulting in unclear cleaning priorities.

[0027] After determining the type of construction machinery and dividing the areas into sections, the corresponding template is retrieved from the template library based on the determined machinery type. A correspondence is established between the predetermined sections in the template and the divided sections. For example, the bucket in the template corresponds to the bucket area identified in the image, the boom in the template corresponds to the boom area identified in the image, and so on. This correspondence can be achieved through section name matching or by comparing spatial location and structural features. After establishing the correspondence, the sections are sorted according to the dirt adhesion tendency level recorded in the type template, generating a cleaning section list. During sorting, sections corresponding to higher-level predetermined sections are listed first, followed by medium-level sections, and then lower-level sections. The cleaning section list can include the section name, dirt adhesion tendency level, and cleaning priority order, serving as the basis for subsequent boom movement control and high-pressure water jet start / stop. Through these steps, this method can automatically determine the key cleaning sections and their cleaning priority order based on the type of construction machinery, eliminating the need for manual parameter resetting when dealing with different types of construction machinery, thus improving the efficiency and targeting of cleaning operations.

[0028] (2) In another embodiment, each predetermined part in the type template also records the surface structure type of that predetermined part. The surface structure type can be classified according to the structural characteristics of the construction machinery parts, such as continuous curved surfaces, hinge gaps, grids or holes, and planar shells. Taking the excavator type template as an example, the surface structure type of the inner and outer walls of the bucket can be recorded as a continuous curved surface, the joint hinge between the boom and the stick can be recorded as a hinge gap, the radiator guard in the engine compartment can be recorded as a grid, and the top of the cab can be recorded as a planar shell. The surface structure type data can be stored in the template library along with the dirt adhesion tendency level as part of the part attribute information.

[0029] When generating the list of cleaning areas, for at least one area in the list, the cleaning intensity parameter for that area is determined based on the surface structure type of the corresponding predetermined area and the dirt adhesion tendency level of that predetermined area. Specifically, the determination method can be as follows: the dirt adhesion tendency level is used as a baseline factor, reflecting the likelihood and extent of dirt adhesion to that area under typical operating conditions; a higher level indicates a stronger cleaning requirement. Simultaneously, the surface structure type is used as an adjustment factor, reflecting the surface response characteristics and cleaning difficulty of that area to the cleaning water jet. For example, for areas with a high degree of dirt adhesion and a continuous curved surface structure, the high-pressure water jet working pressure in the cleaning intensity parameters can be set to the range of 200 bar to 350 bar, and the nozzle movement speed can be set to a medium speed. For areas with a high degree of dirt adhesion but a hinged gap surface structure, the working pressure can be set to the range of 150 bar to 250 bar, the nozzle movement speed can be set to a low speed, and the number of cleaning passes can be set to two or more to allow sufficient time for the water flow to penetrate the gaps and wash away the dirt. For areas with a grid surface structure, the nozzle movement speed can be set to a high speed, and the working pressure can be set to the range of 100 bar to 200 bar to avoid high-pressure impact damaging the heat sink behind the grid. The cleaning intensity parameters include at least one of the high-pressure water jet working pressure, nozzle movement speed, and number of cleaning passes. Once these parameters are determined, they are associated with the corresponding areas in the cleaning area list for use by the subsequent work arm motion controller and cleaning control module.

[0030] Existing cleaning equipment typically does not differentiate between structural types of different parts of the construction machinery, applying the same or similar cleaning pressure and water flow time to all soiled areas. This can easily lead to over-rinsing of simple areas and under-rinsing of complex areas. This embodiment, by pre-setting surface structure types in a template and using both surface structure type and dirt adhesion tendency level as the basis for determining cleaning intensity parameters when generating a list of cleaning areas, automatically matches differentiated cleaning methods to areas with different structural characteristics, improving the utilization efficiency of cleaning resources and the uniformity of cleaning results.

[0031] (3) In another implementation, such as Figure 2 As shown, this further defines how to handle potential conflicts between decontamination requirements and structural protection during the determination of cleaning intensity parameters. When determining the cleaning intensity parameter for at least one area in the cleaning area list, a baseline cleaning intensity parameter is first determined based on the dirt adhesion tendency level of that area, and then an intensity limit threshold is determined based on the surface structure type of that area. The baseline cleaning intensity parameter reflects the cleaning force requirement calculated based on the dirt adhesion tendency of that area. The higher the dirt adhesion tendency level, the higher the working pressure can be set in the baseline cleaning intensity parameter, or the lower the nozzle movement speed can be set. The relationship between the dirt adhesion tendency level and the baseline cleaning intensity parameter can be set as follows: When the dirt adhesion tendency level is high, the baseline working pressure can be set in the range of 250 bar to 350 bar, and the nozzle movement speed can be set in the range of 0.1 m / s to 0.3 m / s; when the dirt adhesion tendency level is medium, the baseline working pressure can be set in the range of 150 bar to 250 bar, and the nozzle movement speed can be set in the range of 0.3 m / s to 0.5 m / s; when the dirt adhesion tendency level is low, the baseline working pressure can be set in the range of 50 bar to 150 bar, and the nozzle movement speed can be set in the range of 0.5 m / s to 0.8 m / s. These numerical ranges can be adjusted appropriately according to the specific type of machinery being cleaned, the type of surface dirt, and the actual site conditions. The intensity limit threshold reflects the upper limit of cleaning intensity that the surface structure of that area can withstand, and is determined by the surface structure type. For example, for areas with a continuous curved surface, the upper limit of the working pressure in the strength limit threshold can be set to 500 bar; for areas with a hinged gap, considering that there may be seals or bushings at the gap, the upper limit of the working pressure can be set to 250 bar; for areas with a grille, the upper limit of the working pressure can be set to 200 bar to avoid the heat sink from deforming or collapsing due to high pressure water flow.

[0032] After determining the baseline cleaning intensity parameter and the intensity limit threshold, the two are compared. If the baseline cleaning intensity parameter does not exceed the intensity limit threshold, it indicates that there is no conflict between the decontamination requirements and the structural bearing capacity of that area. In this case, the baseline cleaning intensity parameter is used as the actual cleaning intensity parameter for that area. For example, for a continuous curved surface, the baseline working pressure is 300 bar, which does not exceed the 500 bar upper limit corresponding to this surface type. Therefore, the actual working pressure is taken as 300 bar. If the baseline cleaning intensity parameter exceeds the intensity limit threshold, it indicates that there is a conflict between the decontamination requirements and the structural bearing capacity of that area. In this case, the intensity limit threshold is used as the actual cleaning intensity parameter for that area. Compared to the case where it does not exceed the threshold, the cleaning operation time for that area is extended to compensate for the reduced cleaning energy input due to the intensity limitation. For example, for a hinge gap, the baseline working pressure is 300 bar, but the upper limit corresponding to this surface type is 250 bar. Therefore, the actual working pressure is taken as 250 bar, and the cleaning operation time is extended. The duration of cleaning operations can be extended by at least one of the following methods: reducing the moving speed of the high-pressure water jet nozzle to increase the time the water jet acts on a unit area; or increasing the number of cleaning passes by repeatedly rinsing the same area.

[0033] Cleaning equipment typically determines cleaning intensity based solely on the severity of the dirt, applying higher pressure to heavier dirt without considering the structural tolerance of the surface being cleaned. This approach can easily cause seal damage or component deformation at sensitive structures such as hinge gaps and radiator grilles. This embodiment introduces an intensity limit threshold corresponding to the surface structure type and automatically employs time compensation when the baseline intensity exceeds the limit. This allows the cleaning system to ensure cleaning effectiveness while avoiding damage to the surface being cleaned, balancing decontamination efficiency and structural safety.

[0034] (4) In another implementation, such as Figure 3As shown, this paper addresses potential mismatches that may occur after establishing a correspondence between predetermined parts in the type template and the area regions divided in the image. After establishing the correspondence between the predetermined parts in the type template and each area region, if any area cannot be matched with any predetermined part in the type template, that area is marked as an unmatched area. Reasons for unmatched areas may include: non-original components such as toolboxes, guardrails, or temporary ropes attached to the surface of the construction machinery; limited visibility of a functional part in the image due to shooting angle limitations, preventing image data processing from fully identifying it as an independent part; or excessively thick soil coverage blurring the boundaries of the area, causing image data processing to merge two adjacent areas into one. After marking an area as unmatched, the visual feature information of the unmatched area is obtained, along with the visual feature information of at least one area region that has already been matched with a predetermined part. The visual feature information may include at least one of color features, texture features, and material reflectivity. Color features can be extracted from RGB images captured by a 2D camera, texture features can be calculated from images using algorithms such as gray-level co-occurrence matrix or local binary mode, and material reflection characteristics can be obtained by analyzing the intensity of lidar echoes or the differences in surface reflection under different illumination angles.

[0035] The visual feature information of unmatched areas is compared with that of corresponding areas to determine whether the unmatched area should be included in the cleaning list. During the comparison, the distance value in color features, the similarity score in texture features, or the difference in material reflectivity between the unmatched area and each corresponding area can be calculated. When the visual feature similarity between an unmatched area and a corresponding area meets a preset condition (e.g., the distance value in color features is below a preset threshold, or the texture similarity score is above a preset threshold), the unmatched area is considered similar to the corresponding area in material and appearance, and is included in the cleaning list and cleaned according to the cleaning parameters for that area. When the visual features of an unmatched area do not meet the preset similarity conditions with any corresponding area, the unmatched area may be a non-original component or not a cleaning target, and is not included in the cleaning list. Furthermore, when the same unmatched area is marked as an unmatched area in multiple consecutive frames of appearance images and the visual feature information remains consistent across frames, the unmatched area is first attempted to be merged with spatially adjacent corresponding areas. Spatially adjacent regions are determined based on the spatial distance between two areas in a 3D point cloud. If this distance is less than a preset adjacent threshold (e.g., 100 mm to 300 mm), the two regions are considered spatially adjacent. During the merging attempt, the point cloud data of the unmatched region is merged into the adjacent already-matched region, and the merged region is re-evaluated to see if it can establish a correspondence with a predetermined region in the type template. If a correspondence is established after merging, it indicates that the unmatched region may have been separated due to local occlusion, soil coverage, or segmentation boundary errors, resulting in a region belonging to the already-matched region being isolated. This region is considered a cleaning area and is included in the cleaning list, with cleaning parameters applied according to the merged already-matched region. If a correspondence cannot be established with any predetermined region after merging, it indicates that the region does not match the functional region in the template library in terms of spatial structure and visual features. It may be a non-original accessory or a non-cleaning target. In this case, it is excluded from the automatic cleaning range, and a prompt message is generated for the operator to confirm whether manual cleaning of the region is necessary. The number of consecutive frames can be set to three or five frames, and the vehicle control unit performs cumulative judgment.

[0036] Existing cleaning equipment typically lacks a part template matching and non-matching identification mechanism. Operators must manually observe and determine whether all visible areas belong to the parts that need cleaning, which can easily lead to overlooking obscured or structurally complex blind spots, and may also result in ineffective rinsing of non-cleaning targets such as toolboxes and decorative parts. This embodiment identifies non-matching areas, acquires visual features, and performs similarity comparison, enabling the system to automatically distinguish between parts that need cleaning due to obstruction or segmentation errors and non-cleaning targets such as non-original accessories. This reduces the need for manual intervention and improves the accuracy of the cleaning part list and the targeting of cleaning operations.

[0037] (5) In another embodiment, the attribution determination when an unmatched area and multiple corresponding part areas are visually similar is further defined. In actual construction machinery cleaning scenarios, due to surface soil coverage, different functional parts may exhibit similar soil color and texture, resulting in a high degree of visual similarity between an unmatched area and two or more corresponding part areas. In this case, the at least one part area is two or more part areas that have established a correspondence with a predetermined part. When determining whether an unmatched area should be included in the cleaning part list based on the comparison results, the spatial positional relationship between the unmatched area and these two or more part areas is also obtained. The spatial positional relationship can be extracted from the three-dimensional point cloud or surface model corresponding to the unmatched area, including the relative position of the unmatched area and each corresponding part area in three-dimensional space.

[0038] For each of these two or more areas, the corresponding area to which the unmatched area is referenced is determined by combining its visual feature similarity and spatial relationship with the unmatched area. The comprehensive judgment method can be as follows: first, candidate areas whose visual feature similarity meets preset conditions are selected; then, the spatial relationship between these candidate areas and the unmatched area is compared, and the one that best conforms to the common sense of construction machinery structure is selected as the reference object. For example, if an unmatched area on the excavator surface is visually similar to both the boom and stick areas, but is spatially located at the end of the boom and closer to the stick area, it is preferentially identified as the reference stick area. As another example, if an unmatched area is spatially located in the gap between the slewing platform and the traveling body, it is preferentially identified as the reference platform or the traveling body with higher visual similarity. After determining the referenced corresponding area, based on the information of the corresponding area in the cleaning area list, including its cleaning priority and cleaning intensity parameters, it is determined whether the unmatched area should be included in the cleaning area list and the cleaning parameters after inclusion.

[0039] Existing cleaning equipment typically lacks mechanisms for matching parts to templates and determining the attribution of mismatched areas. Faced with construction machinery surfaces obscured by mud, operators struggle to accurately determine the functional attribution of different areas, easily leading to confusion between them. This implementation introduces spatial relationships as a supplementary dimension to visual feature similarity. When multiple candidate objects have similar visual features, spatial structural principles are used for differentiation, improving the accuracy of attribution determination for mismatched areas and reducing misattribution and cleaning omissions caused by mud coverage.

[0040] (6) In another embodiment, the specific types of spatial positional relationships and the comprehensive judgment logic are further defined. The spatial positional relationships include at least one of the following: spatial distance, adjacency relationship, and relative orientation between the unmatched region and two or more corresponding part regions. The spatial distance can be obtained by calculating the Euclidean distance between the point cloud center coordinates of the unmatched region and the point cloud center coordinates of each corresponding part region, which can be directly extracted from the three-dimensional point cloud data. The adjacency relationship can be determined by judging whether two regions share a boundary or have point cloud contact in three-dimensional space. When the point clouds of two regions are within a preset distance threshold and there are no other regions between them, they can be considered to have an adjacency relationship. The preset distance threshold can be set to 50 mm to 200 mm. The relative orientation can be determined by calculating the direction angle of the center of the unmatched region relative to the center of the corresponding part region in the spatial coordinate system. For example, it can be determined whether the unmatched region is located above, below, in front, behind, to the left, to the right, or at the end of the corresponding part region.

[0041] When making a comprehensive judgment to determine the corresponding part area referenced by the unmatched area, the judgment logic is as follows: when there is a part area among two or more part areas that meets the preset conditions for visual feature similarity with the unmatched area, and the part area and the unmatched area meet the preset positional relationship in space, the part area is determined as the corresponding part area referenced by the unmatched area. The preset conditions can be set as follows: the color feature distance is less than a preset color threshold, the texture similarity score is higher than a preset texture threshold, or the material reflectance difference value is less than a preset reflectance threshold. These thresholds can be preset according to the actual coating type of the construction machinery and common soil coverage conditions. The preset positional relationship can include: the spatial distance is less than a preset distance threshold, which can be set to 500 mm to 1000 mm; or the two are adjacent, that is, they are directly adjacent in three-dimensional space; or the unmatched area is located at a preset position of the corresponding part area. For example, in the structural common sense of construction machinery, the bucket is always located at the end of the boom, and the slewing platform is always located above the traveling body. When a certain region simultaneously meets both the conditions of visual feature similarity and spatial position relationship, it is identified as the reference object. Subsequently, the inclusion and cleaning method of the unmatched region are determined according to the cleaning priority and cleaning intensity parameters of the reference object.

[0042] Existing cleaning equipment typically lacks the ability to determine the attribution of unmatched areas. Even when image segmentation technology is used, it relies solely on nearest neighbor matching based on a single visual feature distance, which can easily lead to misclassification when soil cover causes visual homogeneity across multiple areas. This embodiment refines spatial relationships into spatial distance, adjacency, and relative orientation, and employs a judgment logic constrained by both visual similarity and spatial relationship. This allows for attribution differentiation based on the inherent structural spatial patterns of construction machinery when multiple candidate areas have similar visual features, improving the reliability of attribution judgment and reducing misclassification and cleaning omissions caused by soil cover or insufficient visual information.

[0043] (7) In another implementation, such as Figure 4 As shown, the composition of the image acquisition device and the specific methods for type identification and part division are further defined. The image acquisition device includes a first type of imaging device and a second type of imaging device, based on different physical imaging principles. The first type of imaging device can be a lidar, which acquires three-dimensional point cloud data of the target surface by emitting a laser beam and receiving the echo signal. This device is not sensitive to the lighting conditions of the construction environment and can operate normally even in low light conditions inside the tunnel or under strong shadows caused by direct sunlight outdoors, making it suitable for acquiring the contour and spatial structure information of construction machinery. The lidar can be installed on the top bracket or front gimbal of the mobile cleaning vehicle, and the scanning range can cover an area of ​​10 to 50 meters in front of the vehicle. The second type of imaging device can be a binocular vision camera or a depth camera. A binocular vision camera can capture the same scene from different angles through two lenses, acquiring depth information using the parallax principle, while simultaneously recording the color and texture features of the construction machinery surface. The binocular vision camera can be installed at a height of approximately 2 to 3 meters above the ground at the front of the vehicle. The depth camera can be a time-of-flight camera or a structured light camera, acquiring a depth map of the target surface in real time by measuring the time of flight of light or projecting encoded light spots.

[0044] When performing type identification and part region segmentation, preliminary results are first generated based on the 3D geometric information acquired by the first type of imaging equipment. Specifically, the 3D point cloud data acquired by the lidar is filtered and denoised to remove ground point clouds and surrounding debris point clouds, resulting in the initial 3D point cloud of the construction machinery. Then, a point cloud semantic segmentation algorithm can be used to classify each point in the point cloud based on a pre-trained deep learning model, identifying the type of construction machinery, such as different models like excavators, loaders, and rotary drilling rigs. Simultaneously, the point cloud is divided into point cloud regions corresponding to different functional parts, such as the bucket region, boom region, cab region, and track region of an excavator, serving as the preliminary type identification and preliminary part region segmentation results. Since the 3D geometric information reflects the shape and structural features of the construction machinery, even when the surface is covered with soil or the on-site lighting conditions are poor, the basic geometric shape of the construction machinery remains stable, thus the reliability of the preliminary results is high.

[0045] After obtaining preliminary results, the results are corrected by combining two-dimensional texture information or depth information acquired by the second type of imaging equipment. For example, when the soil on the surface of construction machinery is thin in a certain local area, the color and texture information acquired by the binocular vision camera can help distinguish whether the area belongs to the metal bucket surface or the rubber track surface, thereby correcting potential boundary errors in the initial part area division. Another example is when different functional parts of construction machinery are similar in three-dimensional geometry but different in material, such as the metal bottom plate of a rotating platform and the metal top plate of a walking body; differences in color or rust marks in the two-dimensional texture information can help confirm the accurate location of the part boundaries. The corrected results constitute the final determination of the construction machinery type and multiple part areas. Existing cleaning equipment typically relies solely on operator visual inspection or a single type of sensor for judgment. Under conditions of poor lighting or soil coverage, the accuracy and stability of type identification and part division are difficult to guarantee. This implementation method utilizes two types of imaging devices based on different physical principles to work together, and employs a phased processing approach that prioritizes geometric information and supplements it with texture and depth information. This enables the stable and reliable identification of construction machinery types and the regional division of functional parts even under complex construction site conditions, providing an accurate input basis for subsequent part matching and cleaning planning.

[0046] (8) In another implementation, such as Figure 5As shown, the organizational structure of the type templates in the template library is further defined. The type templates in the template library are organized in a hierarchical structure, which includes at least a type layer and a specification layer. The type layer records the names of predetermined parts common to the corresponding construction machinery type, as well as the dirt adhesion tendency level of each predetermined part. Taking the excavator type as an example, the predetermined parts recorded in its type layer may include the bucket, boom, stick, cab, slewing platform, and tracked walking device. The dirt adhesion tendency level of the bucket can be set to high, the boom joint and stick joint can be set to high, the top of the cab and the slewing platform can be set to medium, and the tracked walking device can be set to high. This information reflects the common patterns of dirt adhesion in various functional parts of this type of excavator under typical working conditions and is applicable to most specifications of excavators under this type. The specification layer records the adjustment information of the dirt adhesion tendency level of at least one predetermined part in the type layer for specific specification variants of this construction machinery type, as well as supplementary predetermined parts and their dirt adhesion tendency levels specific to the specification variant. For example, for the mini excavator specification variant, the specification layer can lower the bucket dirt adhesion tendency level from high to medium, because mini excavators have small bucket capacity and relatively limited dirt adhesion. For the large mining excavator specification variant, the specification layer can supplement a predetermined part not recorded in the type layer, such as the rock bucket wear plate, and set its dirt adhesion tendency level to high.

[0047] In practical applications, the specifications of construction machinery can be identified using features from appearance images acquired by image acquisition devices. For example, geometric features such as the overall size range of the construction machinery, the ratio of track width to body width, and the ratio of boom length to body height can be used to distinguish different specifications of the same type. Specification variation identification can also be done manually by the operator through a human-machine interface. When retrieving the corresponding type template from the template library based on the determined construction machinery type, the information from the type layer and specification layer is merged according to the specifications of the construction machinery to be cleaned, generating an applicable type template instance. During merging, the common information recorded in the type layer is used as a basis, and then the adjustment information in the specification layer is used to overwrite or supplement it. For example, when the machinery to be cleaned is identified as a mini excavator, the bucket, boom, stick, cab, slewing platform, and tracked walking device, along with their respective dirt adhesion tendency levels, are first loaded from the excavator type layer. Then, the adjustment information for the bucket level in the mini excavator specification layer is loaded, adjusting the bucket level from high to medium, while simultaneously loading the supplementary part information from the specification layer. The type template instance generated after the merging is completed can be used for subsequent part area correspondence and cleaning part list generation.

[0048] Existing cleaning equipment typically lacks a hierarchical template library. Creating completely independent templates for different specifications of construction machinery leads to data redundancy and is inconvenient for maintenance. Conversely, creating a single template by type fails to reflect the differences between specification variations. This implementation uses a hierarchical organization method where the type layer records common information and the specification layer records individual differences. This allows the template library to reduce data redundancy by utilizing the structural commonalities of similar machinery, while ensuring adaptability to specific specification variations through adjustments and additions at the specification layer. Furthermore, when encountering new specification variations, the template library's coverage can be expanded simply by adding the corresponding specification layer record. Example

[0049] At a tunnel construction site, the tunnel floor is an unpaved surface composed of a mixture of gravel and soil, approximately 8 meters wide, relying primarily on construction lighting and the lighting provided by the mobile equipment itself. A 20-ton medium-sized tracked excavator, having completed earthmoving operations, is parked inside the tunnel. A thick layer of soil adheres to the inner and outer walls of its bucket, and mud and sand are embedded in the gaps between the boom and stick hinges. Dust is scattered on the slewing platform, and mud clumps have accumulated in the track tracks and support rollers. A 3-ton mini excavator is also parked in the same tunnel, with relatively less soil adhering to its bucket and tracks.

[0050] The mobile cleaning system uses a mobile chassis, with a water tank, high-pressure water pump set, and control unit mounted on the vehicle body. The image acquisition device includes a LiDAR and a set of binocular vision cameras. The LiDAR is mounted on a bracket on the top of the vehicle body, and the binocular vision cameras are mounted at the front of the vehicle body approximately 2.5 meters above the ground. The template library contains pre-stored excavator type templates. The type layer records six predetermined parts: bucket, boom, stick, cab, slewing platform, and tracked travel system. The bucket, boom joint, stick joint, and tracked travel system are designated as having a high dirt adhesion tendency level, while the top of the cab and the slewing platform are designated as having a medium level. The specification layer contains adjustment information for mini excavators, lowering the bucket's dirt adhesion tendency level from high to medium. Each predetermined part also records its surface structure type: the bucket is a continuous curved surface, the boom and stick joints have articulated clearances, the tracked travel system has articulated clearances, and the slewing platform is a flat shell.

[0051] The mobile cleaning system stops approximately 6 meters to the side of the medium-sized excavator and activates its image acquisition device. The LiDAR scans the excavator within about 30 seconds, acquiring 3D point cloud data. Simultaneously, a binocular vision camera acquires 2D texture images. The control unit first filters and denoises the 3D point cloud data, then uses a point cloud semantic segmentation algorithm to generate preliminary type identification and preliminary part region division results. This identifies the machine as an excavator and divides the point cloud regions for the bucket, boom, stick, cab, slewing platform, and tracked travel device. The 2D texture information is then used to correct the part boundaries, confirming the extent of each part region. Based on the excavator type, a type template is retrieved from the template library, establishing a correspondence between the six predetermined parts in the template and the six part regions divided from the image. According to the dirt adhesion tendency level of each predetermined part, the bucket, boom joint, stick joint, and tracked travel device, corresponding to high-level dirt, are listed first, while the slewing platform and the top of the cab, corresponding to medium-level dirt, are listed later, generating a list of cleaning parts.

[0052] For each area in the cleaning list, the cleaning intensity parameters are determined based on its surface structure type and dirt adhesion tendency level. For the bucket, which is a continuous curved surface with a high dirt adhesion tendency, the baseline working pressure is set to 300 bar, the nozzle movement speed is set to medium, and the intensity limit threshold corresponding to the continuous curved surface structure is 500 bar. Since the baseline pressure does not exceed the threshold, the actual working pressure is 300 bar. For the boom joint, which is an articulated gap with a high dirt adhesion tendency, the baseline working pressure is set to 300 bar, and the intensity limit threshold corresponding to the articulated gap structure is 250 bar. Since the baseline pressure exceeds the threshold, the actual working pressure is 250 bar, and the cleaning operation time is extended by reducing the nozzle movement speed and increasing the number of cleaning passes. The treatment for the tracked travel device is similar to that of the boom joint. For the cab top, which is a flat shell with a medium dirt adhesion tendency, the baseline working pressure is set to 150 bar, and the intensity limit threshold corresponding to the flat shell is 400 bar. Since the baseline pressure does not exceed the threshold, the actual working pressure is 150 bar. The above parameters are associated with each area in the cleaning list.

[0053] The boom, following the checklist, sequentially drives the cleaning units to clean each part of the medium-sized excavator. The actual working pressure and cleaning time are executed according to the parameters associated in the checklist. After the medium-sized excavator is fully cleaned, the mobile cleaning system moves to the side of the mini excavator. The image acquisition device captures an image of the mini excavator's exterior, processes the data to identify the excavator type, and determines it as a mini excavator specification variant based on its external dimensions. When retrieving the template, the excavator type layer and the mini excavator specification layer information are merged, and the bucket's dirt adhesion tendency level is adjusted to medium. The subsequent steps for matching, sorting, and determining cleaning parameters are the same as for the medium-sized excavator, completing the cleaning of the mini excavator. No manual parameter reset is required between the two cleaning operations.

[0054] Comparative Example In contrast, at the same tunnel construction site, operators used a lifting device to spray and wash the same 20-ton medium-sized excavator. During cleaning, operators visually determined the location of dirt on the surface of the construction machinery and rinsed the visible dirt areas one by one. Standing on the lifting device, the operators visually inspected the excavator's surface, first rinsing the thick layer of visible mud on the inner and outer walls of the bucket, then moving to the boom and stick areas to wash away visible mud and sand, and finally rinsing the slewing platform and tracks. During the cleaning process, the operators relied entirely on personal experience to judge the cleaning time and water gun distance for each part, and some of the mud and sand embedded in the gaps at the boom and stick joints were missed due to the limited observation angle. The same water gun pressure and rinsing method were used for the inner and outer walls of the bucket and the track area, without any differentiation. After cleaning, when cleaning the mini excavator, the operators still had to manually judge and operate according to the same procedure.

[0055] After cleaning the medium-sized excavators, cleanliness checks were performed on six functional areas of both the medium-sized excavators cleaned using the method of this invention and those cleaned using the comparative method. The inspection method involved selecting a certain area in each area and visually observing for any clumps of soil residue or embedded mud. The results showed that the medium-sized excavator cleaned using the method of this invention had no obvious soil residue on the inner and outer walls of the bucket, mud had been flushed out from the gaps in the boom and stick joints, mud clumps had been removed from the track track gaps, and the top of the cab and the surface of the slewing platform were clean. The medium-sized excavator cleaned using the comparative method had a large area of ​​soil removed from the inner and outer walls of the bucket, but mud residue was visible in the gaps of the boom joint, a small amount of mud clumps were still embedded in some track track gaps, and some areas of the slewing platform had unwashed dust.

[0056] The above comparison shows that the present invention automatically identifies the type of construction machinery and key cleaning areas through template-driven methods, and applies differentiated cleaning parameters to areas with different structural characteristics. This ensures that hidden areas such as complex hinge gaps receive cleaning treatment that matches their structural characteristics, reducing missed cleaning and uneven cleaning caused by differences in human experience and limitations in observation conditions. Simultaneously, the image acquisition device uses LiDAR and a binocular vision camera working in tandem, enabling area identification and segmentation even under insufficient lighting conditions inside the tunnel. In contrast, relying solely on manual visual inspection or ordinary cameras is significantly less effective in dim lighting conditions.

[0057] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. It can be applied to various fields suitable for the present invention. Further modifications can be readily implemented by those skilled in the art.

Claims

1. A method for controlling the cleaning of construction machinery, characterized in that, Includes the following steps: A template library is provided, which stores templates for at least two types of construction machinery. Each template type records the names of multiple predetermined parts corresponding to the construction machinery type and the dirt adhesion tendency level of each predetermined part. The image acquisition device acquires the appearance image of the construction machinery to be cleaned. Through image data processing, the type of construction machinery to be cleaned is determined, and the surface area of ​​the construction machinery to be cleaned in the appearance image is divided into multiple parts. Based on the determined type of construction machinery, retrieve the corresponding type of template from the template library, and establish a correspondence between the predetermined parts in the type of template and multiple part areas; Based on the dirt adhesion tendency level recorded in the type template, the corresponding areas are sorted to generate a list of key cleaning areas and their cleaning priority order for the construction machinery to be cleaned.

2. The construction machinery cleaning control method according to claim 1, characterized in that: In the type template, each predetermined part also records the surface structure type of that predetermined part; When generating a list of cleaning areas, for at least one area in the list, the cleaning intensity parameter for that area is determined based on the surface structure type of the predetermined area corresponding to that area and the dirt adhesion tendency level of the predetermined area, and the cleaning intensity parameter is associated with the corresponding area in the list of cleaning areas. Among them, the cleaning intensity parameters include at least one of the following: the working pressure of the high-pressure water jet, the nozzle moving speed, and the number of cleaning passes.

3. The construction machinery cleaning control method according to claim 2, characterized in that: For at least one area, a baseline cleaning intensity parameter is determined based on the dirt adhesion tendency level of that area, and an intensity limit threshold is determined based on the surface structure type of that area. Compare the baseline cleaning intensity parameter with the intensity limit threshold: If the baseline cleaning intensity parameter does not exceed the intensity limit threshold, then the baseline cleaning intensity parameter shall be used as the actual cleaning intensity parameter for that area. If the baseline cleaning intensity parameter exceeds the intensity limit threshold, the intensity limit threshold shall be used as the actual cleaning intensity parameter for that area, and the cleaning operation time for that area shall be extended compared to the case where it does not exceed the threshold. The extension of cleaning operation time is achieved by at least one of the following methods: reducing the moving speed of the high-pressure water jet nozzle and increasing the number of cleaning passes.

4. The construction machinery cleaning control method according to claim 1, characterized in that, After establishing the correspondence between the predetermined parts in the type template and multiple part regions, it also includes: If there are multiple regions that do not correspond to any of the predetermined regions, they are marked as unmatched regions. Obtain visual feature information of unmatched areas, and obtain visual feature information of at least one area that has been associated with a predetermined area; the visual feature information includes at least one of the following: color features, texture features, and material reflectivity. The visual feature information of the unmatched area is compared with the visual feature information of the at least one part area, and the unmatched area is determined to be included in the list of cleaning parts based on the comparison result. When an unmatched area is marked as an unmatched area in multiple consecutive frames of appearance images and has consistent visual feature information, the unmatched area is merged with the spatially adjacent corresponding area. If the merged area can be matched with a predetermined part in the type template, it is included in the cleaning part list. If the merged area still cannot be matched with any predetermined part, it is excluded from the automatic cleaning range, and a prompt message is generated for the operator to confirm.

5. The construction machinery cleaning control method according to claim 4, characterized in that, At least one area is two or more area regions that have been established as corresponding to predetermined areas, and when determining whether to include unmatched areas in the list of areas to be cleaned based on comparison results: Obtain the spatial relationship between the unmatched region and the two or more part regions; For each of two or more part regions, based on the visual feature similarity and spatial relationship with the unmatched region, determine the corresponding part region referenced by the unmatched region from the two or more part regions; Based on the information of the corresponding areas in the cleaning area list, determine whether the unmatched areas should be included in the cleaning area list.

6. The construction machinery cleaning control method according to claim 5, characterized in that, Spatial positional relationships include at least one of the following: spatial distance, adjacency relationship, and relative orientation between the unmatched region and the two or more part regions, respectively; Furthermore, by combining the visual feature similarity and spatial relationship with the unmatched region, the corresponding region referenced by the unmatched region is determined from two or more region regions, including: When there is a region among two or more regions, and the visual feature similarity between the region and the unmatched region meets a preset condition, and the region and the unmatched region satisfy a preset positional relationship in space, the region is determined as the corresponding region referenced by the unmatched region.

7. The construction machinery cleaning control method according to claim 1, characterized in that, Image acquisition devices include a first type of imaging device and a second type of imaging device. The first type of imaging device and the second type of imaging device are based on different physical imaging principles. The three-dimensional geometric information of the construction machinery to be cleaned is obtained through the first type of imaging equipment; Two-dimensional texture or depth information of the construction machinery to be cleaned is obtained through a second type of imaging device; Determine the type of construction machinery to be cleaned, and divide the surface area of ​​the machinery in the appearance image into multiple regions, including: Based on the three-dimensional geometric information, preliminary type identification results and preliminary part area division results of the construction machinery to be cleaned are generated; By combining two-dimensional texture information or depth information, the preliminary type identification results and preliminary part area division results are corrected to obtain the type of construction machinery to be cleaned and multiple part areas.

8. The construction machinery cleaning control method according to claim 1, characterized in that, The type templates in the template library are organized in a hierarchical structure, which includes at least a type layer and a specification layer. The type layer records the names of the predetermined parts common to the corresponding construction machinery type, as well as the dirt adhesion tendency level of each predetermined part; The specification layer records adjustment information for the dirt adhesion tendency level of at least one predetermined part in the type layer under a specific specification variant of the corresponding construction machinery type, and / or supplementary predetermined parts and their dirt adhesion tendency levels; When retrieving the corresponding type template from the template library based on the determined type of construction machinery, the information of the type layer and the specification layer are merged according to the specification variant of the construction machinery to be cleaned, and an applicable type template instance is generated.

Images