Stain estimation method and module for air conditioner, estimation model, and stain presentation method and presentation module for air conditioner

Abstract

The invention provides a smudginess estimation method and module for an air conditioner, an estimation model, and a smudginess presentation method and module for the air conditioner, which can efficiently and accurately estimate the smudginess of other parts of the air conditioner or the whole air conditioner based on the smudginess of one part of the air conditioner. The dirt estimation method includes: a first step of acquiring an image of a visible portion of the air conditioner; 2, acquiring smudginess information which is associated with the image and represents the smudginess condition of the visible part; and a third step of estimating, on the basis of the image and the contamination information, the contamination situation of a portion other than the visible portion of the air conditioner.

Description

Technical Field

[0001] This application relates to both the air conditioning field and the image processing field. More specifically, it relates to a method for estimating the dirt of an air conditioning device, a dirt estimation module, an estimation model, a dirt indication method and module for an air conditioning device, a computer system, a storage medium, and a program product. Background Technology

[0002] It is known that during long-term operation, air conditioning units can accumulate dust, oil, microorganisms, and other pollutants in their internal components (such as heat exchangers, fans, fan casings, and condensate trays). This can lead to decreased overall performance, increased energy consumption, deterioration of air quality, and even equipment malfunction. Therefore, conducting a comprehensive and accurate assessment of the dirt and grime status of air conditioning units and carrying out targeted maintenance accordingly is crucial for ensuring their efficient and stable operation.

[0003] Currently, common methods for detecting and assessing air conditioner dirt mainly include manual visual inspection, monitoring based on fixed-position sensors, and localized dirt analysis based on image processing. For example, some existing technologies use fixed cameras installed at specific locations on the air conditioning unit (such as near the air outlet, air inlet, or drip tray) to collect images and combine them with equipment operating parameters (such as operating time and air volume) and environmental data (such as air quality and humidity) to perform multi-dimensional data fusion, thereby improving the accuracy of judging dirt in specific local areas (e.g., CN201980093123, JP2020139673A). Other technologies attempt to use images of the air outlet or air inlet grille taken by the user to achieve simple classification or model identification of the degree of dirt on the visible component (e.g., JP2022142018A). Still other technologies analyze images of the drip tray taken from a fixed perspective to assess the risk of microbial growth and issue early warnings (e.g., JP2022139587A).

[0004] However, most existing methods can only assess the dirt status of specific local components (single visible parts) directly photographed or monitored in isolation. They cannot effectively infer the overall dirt status of other components (whether visible or invisible) of the air conditioning unit based on the dirt status of one component. For example, it is impossible to accurately infer the dirt level of the air-facing side or fan casing based on the dirt on the heat exchanger's air outlet surface, resulting in a lack of a holistic perspective in maintenance decisions. Consequently, because a comprehensive dirt status assessment report covering multiple key components cannot be provided, it is difficult to generate scientific and accurate cleaning or maintenance plans. Users may therefore perform unnecessary disassembly inspections or take incomplete cleaning measures, leading to inefficiency.

[0005] In addition, some existing technologies rely heavily on pre-installed fixed cameras or dedicated sensors, which not only increases the initial cost and installation complexity of the equipment, but also limits its application flexibility in existing equipment or specific installation environments (such as confined spaces).

[0006] On the other hand, in existing technologies, even if images of a portion or the entire air conditioning unit are obtained through photography or other means, the determination of whether a portion or the entire unit needs cleaning is often done manually. However, due to varying skill levels among staff, there may be instances where a portion or the entire air conditioning unit is severely dirty but is deemed unnecessary to clean, leading to unnecessary malfunctions. Summary of the Invention

[0007] This application was developed in consideration of the above circumstances, and its purpose is to provide a method for estimating the dirt level of an air conditioning unit, which can efficiently and accurately infer the dirt level of other parts of the air conditioning unit or the entire air conditioning unit based on the dirt level of a part of the air conditioning unit. Based on this, this application also provides a dirt level estimation module for an air conditioning unit, an estimation model for dirt level estimation, a computer system capable of implementing the above-described dirt level estimation method, a computer-readable storage medium, and a computer program product. On the other hand, this application also provides a dirt level indication method for an air conditioning unit, a dirt level indication module, a computer system capable of implementing the above-described dirt level estimation method, a computer-readable storage medium, and a computer program product.

[0008] The first technical solution of the present invention provides a method for estimating the dirt status of an air conditioning device, comprising: a first step of acquiring an image of a visible portion of the air conditioning device; a second step of acquiring dirt information associated with the image and characterizing the dirt status of the visible portion; and a third step of estimating the dirt status of portions of the air conditioning device other than the visible portion based on the image and the dirt information.

[0009] According to the dirt estimation method for air conditioning units described in the first technical solution, dirt estimation from local to overall or from one local to another is achieved by establishing a dirt correlation between the visible parts of the air conditioning unit and other parts (whether visible or invisible) of the unit. In particular, for assessing the dirt condition of invisible parts, the method can efficiently and non-disassemble components that are difficult to directly observe or touch, based on easily obtainable images of the visible parts. This significantly reduces the workload of dirt detection, improves detection efficiency, and provides crucial decision-making basis for equipment maintenance.

[0010] Optionally, in the second technical solution of the present invention, in the first step, multiple local images showing local areas of the visible portion are acquired. The second step includes: acquiring local dirt information associated with each local image, characterizing the dirt condition of the local area; and estimating the dirt condition of the portion outside the visible portion based on the multiple local images and the local dirt information associated with each local image.

[0011] According to the second technical solution, the method for estimating the dirt level of an air conditioning unit allows users or the system to submit multiple partial images representing local areas of the visible portion, rather than requiring images representing the entire visible portion. This achieves the estimation of dirt levels in areas outside the visible portion while reducing the requirements on the performance of the shooting equipment and the user's shooting skills. Users can use ordinary mobile devices to take multiple shots from multiple angles, greatly improving operability and flexibility in complex or confined environments.

[0012] Optionally, in the third technical solution of the present invention, a positioning step is further included. In this positioning step, based on a pre-stored template image corresponding to the visible portion, the position information of a local area shown in each partial image on the visible portion is determined. The template image includes position labels for representing different areas. In the third step, based on the partial image, the position information corresponding to the partial image, and the local dirt information associated with the partial image, the dirt condition of the portion outside the visible portion is estimated.

[0013] According to the dirt estimation method for air conditioning devices described in the third technical solution, by introducing a positioning step, the location information of each dirt point can be explicitly obtained. This provides key locational features for subsequent dirt estimation, enabling the model to not only know "how dirty a certain place is" during training and inference, but also "which specific part of the visible area this dirt point is located in," thereby further improving the accuracy and reliability of the estimation results. Simultaneously, it avoids errors that may arise during the stitching of multiple local images due to misalignment, fusion defects, etc.

[0014] Optionally, in the fourth technical solution of this application, obtaining the local dirt information includes:

[0015] Each of the local images is input into a pre-trained first association model; and the first association model outputs local dirt information associated with each input local image. The first association model is trained on a first training dataset, which includes multiple sets of first training data. Each set of first training data includes first sample data and first label data. The first sample data includes a local image showing a local region of the visible portion, and the first label data includes local dirt information associated with the local image, characterizing the dirt status of the local region of the visible portion.

[0016] According to the method for estimating the dirt level of an air conditioning unit as described in the fourth technical solution, local dirt information is automatically acquired by using a first association model trained on a large amount of sample data. This achieves intelligent, efficient, and objective judgment of the dirt level in local images, overcoming the subjectivity and inefficiency of manual judgment, ensuring the accuracy and consistency of local dirt information acquisition, and laying a reliable data foundation for subsequent overall estimation.

[0017] Optionally, in the fifth technical solution of this application, the visible portion is the portion that can be observed from at least one of the air inlet and air outlet of the air conditioning device. The portion other than the visible portion is the invisible portion that cannot be observed from either the air inlet or the air outlet.

[0018] Optionally, in the sixth technical solution of this application, the air conditioning device is a duct-type air conditioning device including a heat exchanger and a fan volute. The visible portion is at least a part of the heat exchanger as seen from the air outlet or at least a part of the fan volute as seen from the air inlet. The invisible portion is another part of the fan volute that cannot be seen from the air inlet and the air outlet.

[0019] According to the fifth and sixth technical solutions, the method for estimating the dirt condition of an air conditioning unit can estimate the dirt condition of invisible parts that are difficult or impossible to image based on the dirt distribution of visible parts that are easily imaged. Therefore, it is possible to estimate the dirt condition of internal components of an air conditioning unit without disassembling the unit.

[0020] Optionally, in the seventh technical solution of this application, in the second step, at least one of the type information, usage duration, and usage environment of the air conditioning device is also obtained. In the third step, based on the image, the dirt information, and at least one of the type information, usage duration, and usage environment, the dirt status of the parts other than the visible parts is estimated.

[0021] According to the method for inferring dirt in air conditioning devices described in the seventh technical solution, multi-dimensional auxiliary information such as equipment type, usage time, and usage environment is introduced for fusion inference, realizing multi-modal data fusion. This enables the inference model to learn more complex and comprehensive dirt formation and evolution patterns, thereby significantly improving the accuracy, adaptability, and generalization ability of the inference results under different models, different service years, and different usage scenarios.

[0022] Optionally, in the eighth technical solution of this application, the third step includes: inputting the image and the dirt information into a pre-trained second association model; and causing the second association model to output an inference result representing the dirt condition of the parts other than the visible parts based on the input image and the dirt information. The second association model is trained based on a second training dataset, which includes multiple sets of second training data. Each set of second training data includes second sample data and second label data. The second sample data includes an image showing the visible parts and dirt information associated with the image that characterizes the dirt condition of the visible parts. The second label data includes data associated with the second sample data that characterizes the dirt condition of the parts other than the visible parts.

[0023] According to the method for estimating the dirt level of an air conditioning unit as described in the eighth technical solution, dirt estimation is achieved by employing a second association model trained on a large amount of paired sample data. Its technical advantage lies in its ability to automatically and efficiently learn and establish complex, nonlinear dirt mapping relationships between visible parts and components outside those visible parts, ensuring the objectivity of the estimation process and the accuracy and reliability of the estimation results.

[0024] Optionally, in the ninth technical solution of this application, an output step is also included, in which an estimated result representing the dirt condition of the invisible part is output together with the image and the dirt information associated with the image.

[0025] According to the method for presuming dirt in an air conditioning unit as described in the ninth technical solution, a complete and traceable data chain is constructed by associating the final presumed result with the original image and its intermediate dirt information as the basis for reasoning. This enhances the transparency of the entire presuming process and the credibility of the results, providing convenience for users to verify conclusions, conduct audits, and perform in-depth analysis, thus forming a closed-loop feedback.

[0026] On the other hand, the tenth technical solution of this application provides a method for indicating the dirt status of an air conditioning device, including: a first step of acquiring an image of a visible part of the air conditioning device; a second step of acquiring dirt information associated with the image that characterizes the dirt status of the visible part; and a fourth step of indicating whether the air conditioning device or the visible part needs cleaning based on the image and the dirt information.

[0027] According to the dirt indication method for air conditioning units described in the tenth technical solution, by analyzing the acquired image and the dirt information associated with the image indicating the dirt condition, it is possible to objectively indicate whether the air conditioning unit or its visible parts need cleaning. This avoids misjudgments caused by varying skill levels among staff, thereby preventing unnecessary malfunctions of the air conditioning unit. Attached Figure Description

[0028] Figure 1 This is a cross-sectional view of an indoor unit of an air conditioning device that uses the dirt estimation method and dirt indication method according to an embodiment and a variation thereof of this application.

[0029] Figure 2 This is a flowchart illustrating a method for estimating dirt according to an embodiment of this application.

[0030] Figure 3 This means that it is implemented based on a pre-trained second association model. Figure 2 A schematic diagram of step ST3 in the dirt estimation method shown.

[0031] Figure 4 This is a flowchart illustrating a first variation of the dirt estimation method according to an embodiment of this application.

[0032] Figure 5 This is a schematic diagram illustrating how local dirt information corresponding to each local image is obtained based on a pre-trained first association model.

[0033] Figure 6 This is a flowchart illustrating a second variation of the dirt estimation method according to an embodiment of this application.

[0034] Figure 7 This is a flowchart illustrating a third variation of the dirt estimation method according to an embodiment of this application.

[0035] Figure 8 This is a schematic diagram representing an example of a template image with visible portions labeled with location tags representing different predefined regions.

[0036] Figure 9This is a flowchart illustrating a fourth variation of the dirt estimation method according to an embodiment of this application.

[0037] Figure 10 It means based on Figure 9 This is a schematic diagram illustrating an example of the output results of the dirt estimation method.

[0038] Figure 11 This is a flowchart illustrating a fifth variation of the dirt estimation method according to an embodiment of this application.

[0039] Figure 12 This is a block diagram illustrating an example of the configuration of a dirt estimation module used in a dirt estimation method according to an embodiment and its variations of this application.

[0040] Figure 13 It means and Figure 12 This is a schematic diagram of an example of the user interface of an application that interacts with the dirt estimation module.

[0041] Figure 14 This is a flowchart illustrating a dirt notification method according to an embodiment of this application. Detailed Implementation

[0042] Figure 1 A cross-sectional view of an indoor unit of an air conditioning device using an embodiment of the present application and its modifications thereof, and a dirt estimation method and a dirt indication method of an embodiment of the present application and its modifications thereof, is shown.

[0043] Figure 1 A cross-sectional view of the indoor unit S of a ducted air conditioning unit, as an example of an air conditioning system, is shown. For ease of description, it is assumed that... Figure 1 The indoor unit S of the ducted air conditioning unit shown has been installed in the designated location of the object space (e.g., the ceiling), and will Figure 1 The left and right directions shown are called the first direction X. Figure 1 The up-down direction shown is called the second direction Y, and the direction orthogonal to the first direction X and the second direction Y is called the third direction Z. Therefore, Figure 1 A cross-sectional view of the indoor unit S of the aforementioned duct-type air conditioning unit, taken along the second direction Y, is shown.

[0044] like Figure 1 As shown, the indoor unit S of the ducted air conditioning unit includes a casing 1, a heat exchanger 2, a drain pan 3, a fan 4, and a fan volute 5.

[0045] The outer casing 1 houses a heat exchanger 2, a water collection tray 3, a fan 4, and a fan volute 5. Furthermore, an air outlet 6 is formed at one end of the outer casing 1 in the first direction X for regulating airflow within the outer casing 1, and on the other side in the first direction X and on one side in the second direction Y (… Figure 1 An air inlet S2 is formed at the lower part (as shown) to allow external air to enter the indoor unit S.

[0046] A side heat exchanger 2 is disposed within the housing 1 on the side near the air outlet S1. For example, it is a cross-finned finned tube heat exchanger. The air flowing through this heat exchanger is heated or cooled by exchanging heat between the refrigerant flowing through it and the air flowing through the side heat exchanger 2. It should be noted that, although in Figure 1 The heat exchanger 2 shown is a straight plate shape, but its shape is not limited to this; it can also be a bent shape (e.g., a "<" shape). Furthermore, it should be noted that although in Figure 1 The heat exchanger 2 shown is a single layer, but the number of layers is not limited to this; it can also be double-layered. That is, it is also possible that two or more heat exchangers 2 are arranged at certain intervals along the first direction X in the part of the housing 1 near the air outlet S1.

[0047] The water collection tray 3 is disposed inside the outer casing 1 below the utilization-side heat exchanger 2, and is used to collect condensate formed by the refrigerant condensing during heat exchange with air and sliding down from the utilization-side heat exchanger 2. Although not shown, a drainage mechanism (e.g., a drain pipe) is connected to a portion of the water collection tray 3. This drainage mechanism allows wastewater such as condensate stored in the water collection tray 3 to be discharged.

[0048] The fan 4 is located inside the housing 1 on the side near the air inlet S2. By rotating, the fan 4 draws air from outside the indoor unit S, i.e., the target space, into the housing 1. The fan 4 can be of various types, such as a propeller fan, an axial fan, or a Sirocco fan.

[0049] The fan volute 5 is configured inside the outer casing 1 to house the fan 4, and its structural shape matches the blade morphology of the fan 4, for example, it is a spirally expanding structure. The fan volute 5 is used to converge, compress and accelerate the air drawn in by the fan 4 as it rotates, and to guide the converged airflow to the heat exchanger 2.

[0050] The above description uses the indoor unit S of a ducted air conditioning unit as an example. However, this is only one of the applicable objects of the dirt estimation method of one embodiment and its modifications of this application. That is, the dirt estimation method and the dirt indication method of one embodiment and its modifications of this application are also applicable to other types of air conditioning units.

[0051] Next, the terms "visible portion" and other terms described in the dirt estimation method and dirt indication method of an embodiment and a modification thereof of this application will be explained in detail.

[0052] In this application, "visible portion" refers to a portion that can be observed from at least one of the air inlet or air outlet of an air conditioning unit (indoor or outdoor unit). Furthermore, the "visible portion" described herein can be part or all of a constituent element (or component) of the air conditioning unit. Figure 1 The indoor unit S shown is used as an example for explanation. For example, in the direction of... Figure 1 When observing the interior of the casing 1 through the air inlet S2 of the indoor unit S, the "visible part" can be a portion of the fan 4 or a portion of the fan volute 5 that can be observed through the air inlet S2. For example, when facing... Figure 1 When observing the interior of the casing 1 from the air outlet S1 of the indoor unit S, the "visible portion" can be part or all of the heat exchanger 2 on the utilization side, which can be observed through the air outlet S1. For example, when facing... Figure 1 When observing the interior of the housing 1 from the air inlet S2 and / or air outlet S1 of the indoor unit S shown, the "visible part" may be part or all of the grille (not shown) installed on the air inlet S2 and / or air outlet S1, or part or all of the grille on the air outlet S1 side and part of the side heat exchanger 2, or part or all of the grille on the air inlet S2 side and part of the fan 4 and / or fan volute 5.

[0053] In this application, "dirt information" refers to information characterizing the degree of dirtiness of an evaluation object. This information can be expressed numerically (e.g., a dirtiness score), in natural language (e.g., "moderate dirtiness"), or as a two-dimensional color contour map (also known as a heat map), a three-dimensional rendering based on the visible portion, statistical charts, etc. Furthermore, "dirt information" can also be expressed as a set of vectors, where each vector includes at least a location information component and an information component characterizing the degree of dirtiness at that location.

[0054] In this application, "partial area" refers to the area covered by a portion of the visible portion. For example, assuming the visible portion is a part of the utilizing side heat exchanger 2, then "partial area" refers to the area covered by a portion of that utilizing side heat exchanger 2. As another example, assuming the visible portion is the entire grille at the air inlet S2, then "partial area" refers to the area covered by a portion of the grille at the air inlet S2.

[0055] In this application, "training dataset" is a term in the fields of machine learning or deep learning, referring to the data set used by the model to be learned to train and learn in order to build a fully learned model, including multiple sets of training data. In unsupervised learning, each set of training data includes only sample data, also referred to as sample data used for training. On the other hand, in supervised learning, each set of training data includes not only sample data but also labeled data as teacher data, and is therefore also referred to as example data used for training.

[0056] Next, refer to Figure 2 The main steps of the method for estimating dirt levels in an air conditioning unit according to one embodiment of this application are described below. It should be noted that... Figure 2 The steps in the flowchart shown are merely an example and are not limiting. That is, as long as the dirt estimation method described in this embodiment can be implemented, other steps may be included, and the order of some steps may be changed.

[0057] First, in step ST1, an image of the visible portion of the air conditioning unit, which is the presumed object, is acquired. Hereinafter, [the following is an example]. Figure 1 The indoor unit S of the ducted air conditioning unit shown is used as an example for explanation. Specifically, at least one of the air outlet S1 and air inlet S2 of the indoor unit S is photographed using a photographing device (e.g., a digital camera) (here, for ease of explanation, it is assumed that only the air outlet S1 is photographed) to obtain a digital image containing feature information of the visible portion of the indoor unit S that can be observed through the air outlet S1 (for ease of explanation, this is illustrated by using part or all of the side heat exchanger 2 as an example). However, the above image may not be acquired in real time; for example, it may be an image previously photographed and pre-stored in a storage device. That is, "acquisition" as described here includes not only images acquired in real time by the photographing device, but also images pre-stored in a storage device.

[0058] Next, in step ST2, dirt information characterizing the dirt condition of the visible portion, associated with the image representing the visible portion of the air conditioning unit obtained in step ST1, is acquired. In other words, dirt information characterizing the dirt condition of the visible portion is acquired based on the feature information representing the visible portion contained in the image obtained in step ST1. Here, two specific implementations of acquiring dirt information based on feature information are described. However, it should be understood that the specific implementations of acquiring dirt information based on feature information are not limited to these two, and any suitable calculation method can be used.

[0059] In one specific implementation, the acquired image is input into a pre-trained learning-complete model. The learning-complete model processes the image as input and outputs dirt information representing the dirt status of the visible parts of the image. The "learning-complete model" mentioned here is a pre-trained image recognition model (such as a convolutional neural network CNN), formed by training on a pre-constructed fourth training dataset.

[0060] Regarding the training phase, a supervised learning model will be used as an example. In supervised learning, the fourth training dataset includes multiple sets of fourth training data, each set comprising fourth sample data and fourth label data. The fourth sample data includes historical image data showing various visible parts, and the fourth label data includes historical dirt information associated with each historical image, characterizing the dirt status of the visible parts contained in that historical image. By using the aforementioned fourth training dataset to train the convolutional neural network to be trained under supervised learning, a learning-completed model is constructed that can extract features related to dirt in visible parts from images used as input data.

[0061] Regarding the inference phase, after the training and construction of the learning completion model are completed as described above, the image containing the feature information of the visible part obtained in step ST1 is input into the learning completion model. The learning completion model performs forward propagation processing on the image, thereby outputting dirt information characterizing the dirt status of the visible part.

[0062] As a second specific implementation method, traditional digital image processing technology can be used to obtain dirt information representing the dirt status of the visible parts by performing image preprocessing, feature and rule judgment on the image containing feature information of the visible parts obtained in step ST1.

[0063] Specifically, depending on the quality of the input image containing feature information of the visible parts, preprocessing operations such as denoising, grayscale conversion, and contrast enhancement can be performed on the image to highlight dirty areas.

[0064] Next, feature extraction is performed on the image. For example, edge detection algorithms (such as Canny and Sobel) are first used to identify structural edges in the image. Then, color analysis (such as saturation (S) and lightness (V) in the HSV color space) is used to identify anomalous color patches. Then, texture analysis (such as Local Binary Mode (LBP) and Gray-Level Co-occurrence Matrix (GLCM)) is applied to detect surface inhomogeneities. Finally, morphological operations (such as erosion and dilation) can be used to further highlight dirty areas.

[0065] Then, based on the extracted features characterizing the visible dirt condition (such as the area ratio of the dirty region, the degree of color deviation, and the texture complexity), and combined with preset thresholds or rules, the degree of dirtiness is calculated. For example, if the color difference between a certain area and its cleanliness exceeds the threshold, it is judged as slightly dirty. Or, if the edge detection results show that the structure is blurred or severely occluded, it is judged as having accumulated dirt, classified as moderately dirty. In conclusion, the above judgment results output dirt information characterizing the dirt condition (degree of dirtiness) of the visible parts.

[0066] After completing step ST2, proceed to step ST3.

[0067] In step ST3, based on the image containing feature information characterizing the visible portion obtained in step ST1 and the dirt information characterizing the dirt condition of the visible portion obtained in step ST2, the dirt condition of the parts of the air conditioning unit other than the visible portion is estimated. For ease of understanding, ... Figure 1 An example is given of the ducted air conditioning unit shown, in which part or all of the heat exchanger 2 on the side at the air outlet S1 of the indoor unit S is shown as a visible part.

[0068] In this example, the "part other than the visible part of the air conditioning unit" refers to a portion of the indoor unit S of the air conditioning unit that is covered in the acquired image of the utilization-side heat exchanger 2. When the visible part is part of the utilization-side heat exchanger 2, the "part other than the visible part" can be the remaining portion of the utilization-side heat exchanger 2 not covered in the aforementioned image, or it can be other components of the indoor unit S besides the utilization-side heat exchanger 2 (such as the water tray 3, fan 4, and fan volute 5). Here, we will use the surface of the utilization-side heat exchanger 2 near the air outlet S1 as an example, and the "part other than the visible part of the air conditioning unit" as the inner surface of the fan volute 5.

[0069] Figure 3A schematic diagram illustrating step ST3 based on the second correlation model M2 is shown. Figure 3 As shown, the image obtained in step ST1, which includes feature information representing the visible portion of the surface of the heat exchanger 2 near the air outlet S1, and the dirt information obtained in step ST2 representing the dirt condition of the visible portion, are input as input data to a pre-trained second association model M2. After receiving the input data, the second association model M2 performs intermediate processing based on the input data and outputs an estimated result representing the dirt condition of the inner surface of the fan casing 5 (i.e., the portion other than the visible portion of the indoor unit S). The "second association model M2" mentioned here is a learned model (e.g., a machine learning model or deep learning model for image recognition) trained on a second training dataset.

[0070] like Figure 3 As shown, as an imaginary model of one embodiment of this application, the second association model M2 includes an input module M201, a processing module M202, and an output module M203. The input module M201 is configured to input an image showing the visible portion of the air conditioning unit (the surface of the heat exchanger 2 near the air outlet S1) and associated dirt information characterizing the dirt condition of the visible portion. The processing module M202 is configured to generate an imaginary result representing the dirt condition of portions of the air conditioning unit other than the visible portion based on the input image and the dirt information. The output module M203 is configured to output the imaginary result.

[0071] Regarding the training phase, we will continue to use supervised learning as an example. In supervised learning, the second training dataset includes multiple sets of second training data, each set including second sample data and second label data.

[0072] The second sample data includes historical images showing the visible portion (i.e., the surface near the air outlet S1 side of the side heat exchanger 2) and historical dirt information characterizing the dirt condition of that visible portion. Specifically, for different types and models of air conditioning units, a large amount of image data and corresponding dirt information are collected in advance for each type and model of visible portion. For example, for the indoor unit of each type and model of ducted air conditioning unit, a large number of images containing feature information of that type of visible portion are collected for each type of visible portion (e.g., the surface near the air outlet S1 side of the side heat exchanger 2). At the same time, for each collected image, dirt information characterizing the dirt condition of the visible portion shown in the image is recorded and collected. As an example of this dirt information, a dirt thermal distribution map of the surface near the air outlet S1 side of the side heat exchanger 2 is provided. Thus, the aforementioned second sample data is formed. However, it should be noted that the form of the second sample data is not limited to the above form. For example, taking the side heat exchanger 2 as an example, it is also possible to collect a large number of historical images of multiple surfaces of the side heat exchanger 2 (the surface near the air outlet S1 side, i.e., the air outlet surface and the surface away from the air outlet S1 side, i.e., the windward surface) and the corresponding historical dirt information.

[0073] The second label data includes data associated with the aforementioned sample data that characterizes the dirt condition of the parts other than the visible parts. Specifically, when the part other than the visible parts of the air conditioning unit is the inner surface of the fan volute 5, for each set of second sample data, the dirt condition of the inner surface of the fan volute 5 is measured using specialized testing instruments (such as an endoscope, ultrasonic detector, etc.) to obtain and form dirt information characterizing the dirt condition of the inner surface of the fan volute 5. As an example of this dirt information, it is, for example, a dirt thermal distribution map of the inner surface of the fan volute 5.

[0074] After collecting the second sample data and second label data that are related to each other, they are integrated to complete the construction of the second training dataset.

[0075] Feature extraction is performed before inputting the second training dataset into the initial model to be trained. Key features (such as the area and location of the dirty area) that are highly correlated with the dirt status of the visible parts are extracted from the historical images. For the initial model to be trained, a suitable regression or classification model (such as linear regression, support vector machine, random forest, etc.) can be selected based on the characteristics of the input data. Then, the initial model is trained using the second training dataset in a manner that minimizes the loss function, establishing the relationship between the dirt information of the visible part (i.e., the surface of the side heat exchanger 2 near the air outlet S1) and the dirt information of the parts outside the visible part (i.e., the inner surface of the fan volute 5), thereby completing the construction of the second correlation model M2.

[0076] Regarding the inference phase, the image containing characteristic information of the visible portion (the surface of the side heat exchanger 2 near the air outlet S1) and the corresponding dirt information, acquired in real time, are input as input data to the pre-trained second association model M2 as described above. The second association model M2 processes the above-mentioned input data in real time and outputs an inference result representing the dirt status of the portion other than the visible portion (the inner surface of the fan volute 5) corresponding to the visible portion.

[0077] According to the dirt estimation method for the air conditioning unit described in this embodiment, by establishing a dirt correlation relationship between the visible part of the air conditioning unit and other parts (whether invisible or other visible parts), dirt estimation from local to overall or from one local to another is realized, and the dirt status of the entire air conditioning unit and other components of the air conditioning unit can be evaluated based on the local dirt status of the air conditioning unit.

[0078] In particular, according to the dirt estimation method for the air conditioning unit described in this embodiment, when the parts other than the visible parts of the air conditioning unit are invisible parts (e.g., components inside the housing 1), since a correlation is established between the dirt condition of the visible parts and the dirt condition of the invisible parts, the dirt condition of the invisible parts can be assessed without disassembling the air conditioning unit. This reduces the workload of dirt detection and improves dirt detection efficiency.

[0079] Next, refer to Figure 4 The flow of a first variation of the method for estimating the dirt level of an air conditioning unit according to the above-described embodiment will be described. However, to avoid repetition, only the differences between the first variation and the above-described embodiment will be described here. Furthermore, for ease of understanding, in the first variation, the method will still be referred to as... Figure 1 The indoor unit S of the ducted air conditioning unit shown is used as an example for explanation.

[0080] Figure 4 A flowchart illustrating the main steps of a first variation of the method for estimating the dirt level of an air conditioning unit according to the above embodiments is shown.

[0081] First, unlike step ST1 in the above embodiment, in step ST1A, multiple partial images are acquired, each showing a local area of ​​the visible portion (the surface of the side heat exchanger 2 near the air outlet S1). In other words, multiple partial images are acquired, each containing feature information characterizing a specific local area of ​​the visible portion. Furthermore, the local areas of the visible portion shown in the multiple partial images may partially overlap or not overlap at all. Regarding the number of partial images, preferably, multiple partial images are acquired such that the combination of the local areas of the visible portion shown in the multiple partial images can cover the entire area of ​​the visible portion. However, in this first variation, it is not mandatory that the combination of the local areas of the visible portion shown in the multiple partial images is sufficient to cover the entire area of ​​the visible portion. Specifically, after completing step ST1A, step ST1B is performed.

[0082] In step ST1B, it is determined whether the combination of each local region shown in the acquired multiple local images covers the entire visible portion and whether this coverage reaches a preset threshold (e.g., 80%). If the threshold is not reached, the process proceeds to step ST1C. If the threshold is reached, the process proceeds to step ST2A. Here, an example of a specific implementation of step ST1B will be described.

[0083] For example, after acquiring multiple local images of the visible portion, depending on the quality of these local images, some or all of them can be preprocessed (e.g., distortion correction, brightness normalization, specific point enhancement, etc.). Next, computer vision techniques (such as SIFT, ORB, SURF, and other feature point extraction and matching algorithms) can be used to match and align these multiple local images. Simultaneously, the type of the visible portion (or, in other words, the type and model of the air conditioning unit and the type of the visible portion) is inferred based on the information from the multiple local images, and a template image corresponding to that visible portion is retrieved from a pre-stored set of template images based on the inference result. For example, if, based on the aforementioned multiple local images, it is inferred that the visible portion is the surface near the air outlet S1 of the side heat exchanger 2, more specifically, if, based on the aforementioned multiple local images, it is inferred that the type of air conditioning unit is a ducted air conditioning unit, the model is a dual-fan turbine ducted air conditioning unit, and the visible portion is the surface near the air outlet S1 of the indoor unit S utilizing the side heat exchanger 2, then, based on this inference result, a template image of the entire visible portion of the indoor unit S of the aforementioned type of ducted air conditioning unit—that is, the entire surface near the air outlet S1 of the side heat exchanger 2—is found from a pre-stored set of templates. Then, feature matching is performed between the aforementioned multiple local images and the template image to determine the area range of the local region shown in each local image on the visible portion. Next, based on the matching result between the aforementioned multiple local images and the template image, the combined coverage range of the areas covered by all the aforementioned local images on the entire visible portion is determined. In other words, based on the matching result, it is determined whether the combination of the various local regions of the visible portion shown in the aforementioned multiple local images reaches a preset threshold for the coverage of the entire visible portion. More specifically, the coverage rate of the combined coverage area of ​​the aforementioned local regions relative to the area of ​​the entire visible portion is calculated to determine whether it reaches a preset threshold. If the calculation result indicates that the threshold is reached or exceeded, it is determined that the aforementioned multiple local images are sufficient to generate an image of the entire visible portion, and the process proceeds to step ST2A. On the other hand, if the calculation result indicates that the threshold is not reached, it is determined that the aforementioned multiple local images alone are insufficient to generate an image of the entire visible portion, and the process proceeds to step ST1C.

[0084] In step ST1C, a prompt message is generated and sent to the user. This prompt message informs the user that the previously input multiple partial images are insufficient to generate an image covering the entire visible area. As an example of the prompt message, a diagram showing the coverage area of ​​the template image and the matched partial images is simultaneously displayed on the display device. Based on this, covered and uncovered areas can be marked with different colors on the template image. Optionally, a text prompt such as "The image you captured did not completely cover the detection area. Please capture more images from different angles based on the display results" is also displayed.

[0085] On the other hand, in step ST2A, local dirt information is acquired, which is associated with each of the aforementioned local images and characterizes the dirt condition of the local region shown in that local image. There are various specific implementation methods for acquiring local dirt information. Here, we will illustrate the acquisition of local dirt information based on a machine learning model or a deep learning model.

[0086] Figure 5 This diagram illustrates how a first association model M1, based on a machine learning or deep learning model, is used to obtain local contamination information for each local image. Figure 5 As shown, each of the multiple local images obtained in step ST1A, each including feature information of a local region of the visible portion (the surface of the side heat exchanger 2 near the air outlet S1), is input as input data to a pre-trained first association model M1. After receiving the input data, the first association model M1 performs intermediate processing on each local image and outputs local dirt information characterizing the dirt status of the local region of the visible portion shown in that local image. The "first association model M1" mentioned here is a learned model (e.g., a machine learning model or deep learning model for image recognition) trained on a first training dataset. Furthermore, as... Figure 5 As shown, in the first variant, the first association model M1 adopts a deep learning model with convolutional neural networks and fully connected layers as the main architecture, but the type of the first association model M1 is not limited to this.

[0087] Regarding the training phase, we will continue to use supervised learning as an example. In supervised learning, the first training dataset includes multiple sets of first training data, each set including first sample data and first label data.

[0088] The first sample data includes historical local images showing various local areas of the visible portion (i.e., the surface near the air outlet S1 side of the side heat exchanger 2). The first label data includes historical local dirt information associated with each historical local image, characterizing the dirt status of the local area of ​​the visible portion shown in that historical local image. Specifically, for different types and models of air conditioning units, for each type and model of air conditioning unit, a large amount of local image data of that type of visible portion and the corresponding local dirt information are collected in advance. Thus, a first training dataset is formed.

[0089] Feature extraction is performed before inputting the first training dataset into the initial model to be trained. Key features (such as the area and location of dirt in local areas) that are highly correlated with the dirt status of the visible parts are extracted from the historical local images that serve as the first sample data. For the initial model to be trained, a suitable regression model or classification model can be selected based on the characteristics of the input data. Then, the initial model is trained using the first training dataset in a manner that minimizes the loss function, establishing the association between each local image of the visible part (i.e., the surface of the side heat exchanger 2 near the air outlet S1) and the local dirt information characterizing the dirt status of each local area shown in each local image, thereby completing the construction of the first association model M1.

[0090] Regarding the inference phase, multiple local images, which are acquired in real time and contain feature information of the local area (using the surface of the side heat exchanger 2 near the air outlet S1), are input as input data to the first association model M1, which is pre-trained as described above. The first association model M1 processes the input data input in real time and outputs local dirt information associated with each local image, characterizing the dirt status of the local area shown in each local image.

[0091] Then, in step ST3A, based on the multiple local images acquired in real time and the local dirt information associated with each local image and characterizing the dirt status of the local area of ​​the visible part shown in each local image, acquired in step ST2A, the dirt status of the parts of the air conditioning unit other than the visible part is estimated.

[0092] Specifically, the aforementioned multiple local images and corresponding local dirt information are input as input data to a pre-trained learning-complete model for processing. As output, the learning-complete model outputs a representation of the dirt status of the portion of the air conditioning unit outside the visible portion. Compared to the first embodiment which uses an image of the entire visible portion, in this first variation, multiple local images representing local areas of the visible portion are used. Therefore, the training dataset used to construct the learning-complete model differs in composition from the second training dataset in the first embodiment. In the first embodiment, the second sample data of the second training dataset includes an image representing the entire visible portion and dirt information characterizing the dirt status of that visible portion. In the first variation, the sample data of the training dataset includes multiple local images representing each local area of ​​the visible portion and local dirt information characterizing the dirt status of the local area shown in each local image. The training method and inference process are the same as in the first embodiment, so repeated explanations are omitted.

[0093] According to the method for estimating the dirt level of an air conditioning unit as described in the first variation, the requirements for the performance of the shooting equipment and the user's shooting skills are reduced because it allows users to submit multiple partial images instead of requiring a single image representing the entire visible area. Therefore, users can use ordinary mobile devices (e.g., smartphone cameras) to take multiple shots from multiple angles, improving operability in complex or confined environments.

[0094] Furthermore, the method for estimating the dirtiness of the air conditioning unit according to the first modification introduces a coverage judgment mechanism, which can effectively prevent estimation errors caused by image omissions or insufficient coverage, thereby improving the accuracy of subsequent estimations from the source.

[0095] Furthermore, according to the dirt estimation method for the air conditioning device described in the first modification, when the combined coverage area of ​​the initially submitted local image by the user is less than a preset threshold relative to the area of ​​the entire visible portion, the user can be guided to accurately locate and supplement the missing portion based on the matching results. This improves fault tolerance and user experience.

[0096] Next, refer to Figure 6 The flow of a second variation of the method for estimating the dirt level of an air conditioning unit according to the above-described embodiment will be described. However, to avoid repetition, only the differences between the second variation and the above-described embodiment and the first variation will be described here. Furthermore, for ease of understanding, in the second variation, the method will still be referred to as... Figure 1 The indoor unit S of the ducted air conditioning unit shown is used as an example for explanation.

[0097] like Figure 6As shown, compared with the method for estimating the dirt level of the air conditioning unit described in the first modification, the second modification further includes steps ST2B and ST2C. Furthermore, as an alternative to step ST3, step ST3B is also included.

[0098] In step ST2B, each piece of local dirt information obtained in step ST2A is attached as annotation data to a corresponding local image, so that each of the multiple local images obtained in step ST1A becomes an annotated local image.

[0099] As a specific implementation method, for example, local dirt information (such as a thermal image segment representing the degree of dirt) can be superimposed on the corresponding local image as a semi-transparent layer (Alpha Blending), so that the local image contains both the original visual features and the dirt distribution features.

[0100] As another specific implementation method, local dirt information (such as a set of vectors containing location and dirt values) can be stored as metadata of the corresponding local image and associated with the image file of that local image itself.

[0101] After completing step ST2B, proceed to step ST2C.

[0102] In step ST2C, an image stitching algorithm is used to stitch together all the labeled local images to generate dirt distribution information for the entire visible area. This stitching not only involves stitching together the local images that form the original visual image, but also includes simultaneously stitching and fusing the attached local dirt information layers or metadata.

[0103] Specifically, based on the feature point matching results of each local image in multiple local images, an adaptive Homography matrix is ​​calculated, thereby mapping each local image to a unified global coordinate system (e.g., the coordinate system of the template image). In particular, for overlapping regions, the values ​​representing the degree of dirtiness in these regions are smoothed using specific algorithms (such as weighted averaging, multi-band mixing, etc.). This generates dirt distribution information for the entire visible area. As a specific representation, this could be, for example, a color heatmap.

[0104] Then, in step ST3B, the dirt distribution information of the entire visible part generated in step ST2C is input as input data to the second association model M2 to estimate the dirt status of the parts of the air conditioning unit other than the visible part.

[0105] According to the method for estimating the dirt level of an air conditioning unit as described in the second variation, by fusing each local image with its corresponding local dirt information to generate a global dirt distribution information, refined and semantic input features can be provided for estimating the dirt level of parts outside the visible area. This reduces the amount and complexity of the original data that the second correlation model needs to process, allowing the model to focus more on learning the deep, non-linear correlation between global dirt patterns (such as the distribution rules and central tendency of dirt) and local dirt levels, thereby further improving the accuracy and computational efficiency of the estimation results.

[0106] Next, refer to Figure 7 The flow of a third variation of the method for estimating the dirt level of an air conditioning unit according to the above-described embodiment will be described. However, to avoid repetition, only the differences between the third variation and the above-described embodiment, as well as the first and second variations, will be described here. Furthermore, for ease of understanding, in the third variation, the method will still be referred to as... Figure 1 The indoor unit S of the ducted air conditioning unit shown is used as an example for explanation.

[0107] like Figure 7 As shown, the difference from the above-described embodiments and the first and second modifications is that it further includes steps ST1D (i.e., the positioning step) and ST2D. Furthermore, the third modification also includes step ST3C to replace step ST3 or step ST3A.

[0108] In step ST1D, based on a pre-stored template image representing the entire visible portion as indicated by the multiple local images obtained in step ST1A, the positional information of the local region shown in each local image on the visible portion is determined. In other words, based on the pre-stored template image, the positional information of the local region shown in each local image on the template image is determined.

[0109] The “template image” described herein is a reference image pre-prepared for various visible parts of air conditioning units of various types and models (such as the surface near the air outlet S1 side of the side heat exchanger 2), representing the entire visible part. A key feature of this template image is that it includes location labels indicating different predefined areas. Figure 8 An example of a template image is shown, with visible portions labeled with location tags representing different predefined regions. For example... Figure 8 As shown, the template image is pre-divided into multiple (six in the figure) blocks, and each block is assigned a number as a location label. However, it should be noted that the location labels can be as follows: Figure 8 As shown, it can be explicit or implicit (e.g., a location coding system bound to image coordinates).

[0110] Regarding the specific implementation of localization, one approach is to perform feature matching between each of the multiple local images and the aforementioned template image (e.g., using algorithms such as SIFT or ORB). After successful matching, based on the corresponding positions of the feature points in the local image on the template image, the area covered by that local image is calculated, and finally, the location information corresponding to that area is parsed out. For example, through the above parsing, it can be seen that a certain local image covers the location ranges of "number 1" and "number 2" in the template image.

[0111] Furthermore, considering that the coverage of a local image may only include a portion of the predefined regions corresponding to multiple location labels, rather than all of them, various further processing methods can be adopted.

[0112] For example, the precise coordinates of each pixel in the local region shown in the local image on the template image can be calculated. Based on the calculation results of the precise coordinates, the intersection of the coverage area of ​​the local image (more precisely, the smallest bounding rectangle or convex hull it covers on the template image) and the predefined region boundaries corresponding to multiple location labels can be determined, thereby outputting the final location information. For example, the location information can be {"coverage area": ​​["region 1", "region 2"],"coverage ratio in region 1": 0.6, "coverage ratio in region 2": 0.4}.

[0113] For example, it can determine which predefined region a certain local image mainly falls into and classify it into that predefined region.

[0114] Then, in step ST2D, the position information determined for each local image in step ST1D is associated and bound with that local image and the corresponding local contamination information obtained in step ST2A to form a data structure. This data structure includes at least data representing the local image, data representing the corresponding local contamination information, and data representing the position information of the local image.

[0115] In step ST3C, the dirt condition of the parts of the air conditioning unit other than the visible parts is estimated based on the multiple local images obtained in step ST1A, the local dirt information associated with each local image and characterizing the dirt condition of the local area shown in the local image obtained in step ST2A, and the location information obtained in step ST2D.

[0116] It should be noted that, unlike the above-described implementation method and the first and second variations, the second sample data of the second training dataset also includes historical location information corresponding to each historical local image.

[0117] According to the dirt estimation method for the air conditioning unit described in the third variation, the precise location of each dirt point is explicitly obtained through independent positioning steps. This provides important one-dimensional input features for model training, enabling the model to not only know "how dirty a certain place is," but also further know "which specific part of the entire visible area this dirt point is located in." Therefore, the accuracy and reliability of the estimation results can be further improved.

[0118] Furthermore, according to the method for estimating the dirt of the air conditioning unit as described in the third modification, by introducing position information, errors caused by misalignment, fusion defects, etc., during the image stitching process can be avoided.

[0119] Next, refer to Figure 9 The flow of a fourth variation of the method for estimating the dirt level of an air conditioning unit according to the above-described embodiment will be described. However, to avoid repetition, only the differences between the fourth variation and the above-described embodiment, as well as the first to third variations, will be described here. Furthermore, for ease of understanding, in the fourth variation, the method will still be referred to as... Figure 1 The indoor unit S of the ducted air conditioning unit shown is used as an example for explanation.

[0120] like Figure 9 As shown, unlike the above-described embodiments and the first to third modifications, the fourth modification further includes step ST4 (i.e., the output step). In step ST4, the estimation result representing the dirt condition of the parts of the air conditioning unit other than the visible parts, generated in step ST3, ST3A, ST3B, or ST3C, is output in conjunction with the corresponding image or partial image and the dirt information or partial dirt information associated with the image or partial image. Optionally, the estimation result may include not only assessment information for evaluating the dirt condition of the parts of the air conditioning unit other than the visible parts, but also solutions proposed based on the assessment information.

[0121] Figure 10 A schematic diagram is shown as an example of the output result. Specifically, as an example of the output result, Figure 10 A report containing text descriptions and icons is displayed. For example... Figure 10 As shown, the report includes "Duct type (i.e., type of air conditioning unit)" information, "(visible part and part outside the visible part) area" information, "dirt score" information corresponding to the "area" information, "remarks" information to describe the degree of dirtiness, "(visible part and part outside the visible part) image" information, and "solution" information.

[0122] According to the method for presuming dirt in an air conditioning unit as described in the fourth variation, a complete, closed-loop data chain can be constructed by associating the final presumed result with the original image (or original partial image) and the corresponding dirt information (or partial dirt information) used as the basis for reasoning. This helps users or the system to retrospectively verify and check at any time, providing a solid foundation for result verification and auditing.

[0123] Next, refer to Figure 11 The flow of the fifth variation of the method for estimating the dirt level of the air conditioning device described in the above embodiments will be described. However, to avoid repetition, only the differences between the fourth variation and the above embodiments, as well as the first to fourth variations, will be described here. Furthermore, for ease of understanding, in the fourth variation, the method will still be referred to as... Figure 1 The indoor unit S of the ducted air conditioning unit shown is used as an example for explanation.

[0124] like Figure 11 As shown, unlike the above-described embodiments and the first to fourth variations, steps ST2AA and ST3D are included as alternatives to steps ST2 or ST2A and ST3 or ST3A or ST3B or ST3C.

[0125] In step ST2AA, based on the image or partial image of the visible portion obtained in step ST1 or step ST1A, in addition to obtaining dirt information or partial dirt information, at least one of the following is also obtained: the type information of the air conditioning device including the visible portion, the usage duration information of the air conditioning device, and the usage environment information of the air conditioning device.

[0126] Subsequently, in step ST3D, the dirt condition of the parts of the air conditioning unit other than the visible parts is estimated by taking at least one of the images or partial images obtained in step ST1, dirt information or partial dirt information obtained in step ST2AA, air conditioning unit type information, usage duration information, and usage environment information.

[0127] It should be noted that, unlike the first embodiment, the first variant, and the second variant described above, the second sample data of the second training dataset also includes at least one of the historical type information, historical usage duration information, and historical usage environment information of the air conditioning device.

[0128] According to the fifth variation of the air conditioning device dirt estimation method, the accuracy and reliability of the estimation can be further improved by introducing multi-dimensional auxiliary information. Specifically, the input data used as the basis for estimation, in addition to image data and corresponding dirt information, further incorporates at least one of the following: equipment static attribute information (type), time dimension information (usage duration), and spatial environment dimension information (usage environment). This achieves multi-modal data fusion. Through multi-modal data fusion, the second association model M2 used for estimation can learn more complex and comprehensive dirt formation and evolution patterns during training. Therefore, the accuracy and adaptability (generalization ability) of the estimation results under different models and usage scenarios can be improved.

[0129] The following is for reference Figure 12 This application describes a dirt estimation module for an air conditioning unit according to one embodiment. The dirt estimation module described in this embodiment is used to implement the dirt estimation method described in the above embodiments and their variations. It should be understood that the dirt estimation module described in this embodiment can be integrated into the control system of the air conditioning unit (e.g., a remote control or control panel), or it can be deployed on a cloud server or a standalone computing device, interacting with the user through an application installed on a mobile terminal (e.g., a smartphone) or personal computer.

[0130] Figure 12 A block diagram illustrating the configuration of the dirt estimation module 100 of this embodiment is shown. Figure 12 As shown, the dirt estimation module 100 mainly includes an input unit 110, an acquisition unit 120, an estimation unit 130, and an output unit 140 that transmit data and signals to each other via a data bus or communication interface.

[0131] Figure 13 A schematic diagram showing an example of the user interface (UI) of an application that interacts with the dirt estimation module 100 of this embodiment is shown.

[0132] The input unit 110 is configured to allow users or external systems to input images or partial images of the visible portion of the air conditioning unit to the dirt estimation module 100.

[0133] Specifically, the input unit 110 has the same software features as applications installed on mobile terminals or personal computers. Figure 12 The user interface shown includes an image upload control A1 (e.g., a button labeled "Upload Image" or "Take a Photo"). By operating this image upload control A1, users can choose to upload a pre-taken image or a partial image from local storage, or directly access the mobile device's camera for live shooting.

[0134] Optionally, images or partial images uploaded or captured by the user or system will be displayed instantly in the image preview area A2 of the user interface. In particular, for the upload of multiple partial images, the user interface will display all uploaded partial images in the form of a thumbnail list, and provide add and delete management functions.

[0135] Furthermore, the function of the input unit 110 is not limited to receiving images of the visible portion or partial images. Optionally, according to the dirt estimation method described in the fifth modification above, the input unit 110 can also receive information such as the type of air conditioning unit, usage duration, and usage environment information manually input by the user or automatically acquired by the system as auxiliary input for the estimation process. However, it should be understood that if the acquisition unit 120 has the function of acquiring the aforementioned auxiliary information, the input unit 110 may not have this function.

[0136] The acquisition unit 120 is configured to acquire dirt information or local dirt information associated with an image or local image input through the input unit 110, which characterizes the dirt condition of the visible portion or local area of ​​the visible portion shown in the image or local image.

[0137] Specifically, the acquisition unit 120 encapsulates the core logic of step ST2, ST2A, or ST2AA described in the above embodiments and their variations, and integrates multiple processing engines. These multiple processing engines include, but are not limited to, a first association model processing engine, a conventional image processing engine, and a metadata parser. The first association model processing engine has a built-in or invokes a pre-trained first association model M1. When an image or a local image is input through the input unit 110, the first association model processing engine automatically performs model inference, thereby outputting the corresponding dirt information or local dirt information. Alternatively, the conventional image processing engine has built-in image processing and feature extraction algorithms, and can output the corresponding dirt information or local dirt information according to pre-designed rules. On the other hand, when the acquired dirt information or local dirt information is stored as metadata associated with the image or local image file, the metadata parser is responsible for reading this associated stored information.

[0138] Optionally, after completing the above processing, the acquisition unit 120 can send the processing result to the user interface. Data representing dirt information or local dirt information (such as numerical scores) will be displayed in the dirt information display area A3. In particular, if the dirt information or local dirt information is represented in the form of a heatmap, it can be overlaid on the corresponding image or local image in the image preview area A2. In this way, the user can intuitively observe which parts of the image are dirty and the degree of dirt.

[0139] The estimation unit 130 is configured to estimate the dirt condition of parts other than the visible parts of the air conditioning unit (e.g., the surface of the side heat exchanger 2 near the air outlet S1) based on the image or partial image input by the input unit 110 and the dirt information or partial dirt information obtained by the acquisition unit 120, and generate an estimation result.

[0140] Specifically, the estimation unit 130 encapsulates the core logic of step ST3, ST3A, ST3B, or ST3C in the above-described embodiments and their variations, and integrates the corresponding second correlation model M2 internally. Depending on the different embodiments or variations, the combination of input data for the estimation unit 130 can be:

[0141] A single image representing the entire visible area + dirt information representing the dirt condition of that visible area (+ at least one of the following: air conditioning unit type information, usage duration information, and usage environment information);

[0142] Show multiple local images of each local area of ​​the visible part, plus corresponding local dirt information (plus at least one of the following: air conditioning unit type information, usage duration information, and usage environment information);

[0143] The system displays multiple local images of each visible local area, along with corresponding local dirt information, location information (and at least one of the following: air conditioning unit type information, usage duration information, and usage environment information).

[0144] The information on the distribution of dirt in the entire visible area is formed by stitching together multiple labeled local images (plus at least one of the following: air conditioning unit type information, usage duration information, and usage environment information).

[0145] The estimation unit 130 calls the second correlation model M2 to process the above input data, performs inference calculations, and outputs an estimation result of the dirt status of the parts other than the visible parts.

[0146] Output unit 140 is configured to output the estimation result generated by estimation unit 130. Specifically, output unit 140 is responsible for formatting and presenting the estimation result. More specifically, output unit 140 sends the above estimation result to the estimation result display area A4 of the user interface for display. The display format can be simple text (such as "Fan volute dirt level: moderate") or a visual chart (such as a dashboard or progress bar). In particular, according to the dirt estimation method described in the fourth modification, output unit 140 has an associated output function, which can associate and bind the above estimation result with the original image (or original partial image) and the corresponding dirt information (or partial dirt information) used as the basis for reasoning and output it. For example, in the user interface, this output result can be Figure 10 The analysis is presented in the form of a report. Based on this, the user can trigger the output unit 140 to generate and download / preview a comprehensive report containing inputs, intermediate results, and final estimated results by clicking a button such as "Generate Report" on the user interface.

[0147] Next, refer to Figure 14 The main steps of the dirt detection method for an air conditioning device according to one embodiment of this application will be described. It should be noted that... Figure 14 The steps in the flowchart shown are merely an example and are not limiting. That is, as long as the dirt estimation method described in this embodiment can be implemented, other steps may be included, and the order of some steps may be changed.

[0148] The difference between the method for estimating the dirt of an air conditioning unit described in the first embodiment above is that, instead of step ST3 (i.e., the third step), the method for indicating the dirt of an air conditioning unit in this embodiment includes step ST5 (i.e., the fourth step).

[0149] In step ST5, based on the image of the visible part of the air conditioning unit (such as the surface of the side heat exchanger 2 near the air outlet S1) obtained in step ST1 and the dirt information associated with the image that characterizes the dirt condition of the visible part obtained in step ST2, a prompt is given as to whether the air conditioning unit or the visible part needs to be cleaned.

[0150] Specifically, as a concrete implementation method, estimation can be made based on preset rules or thresholds. For example, dirt information may include a numerical score (such as a score of 0-100) characterizing the degree of dirt on visible parts or the percentage of the area of ​​the dirty region relative to the total visible area. For different types of visible parts (e.g., the surface of the side heat exchanger 2 near the air outlet S1), one or more thresholds (such as a dirt score below 60 or an area percentage exceeding 40%) are preset, and the acquired dirt information is compared with these thresholds. If the degree of dirt represented in the dirt information exceeds the aforementioned thresholds, it is estimated that cleaning is required, and optionally, a prompt message indicating that the air conditioning unit or the aforementioned visible part needs cleaning can be generated.

[0151] As another specific implementation, prompts can be provided using an artificial intelligence model. Specifically, as input data, an image of the visible part of the air conditioning unit (such as the surface near the air outlet S1 side of the side heat exchanger 2) obtained in step ST1, along with dirt information associated with the image and characterizing the dirt condition of the visible part obtained in step ST2, are input into a pre-trained third association model. This third association model infers whether the air conditioning unit or the visible part needs cleaning based on the input image and dirt information, and optionally outputs prompts indicating whether cleaning is needed (such as "Immediate cleaning recommended" or "Current condition is good"). The "third association model" mentioned here is a learned model (e.g., a classification model or a regression model) trained on a pre-prepared third training dataset. Here, we will still use supervised learning as an example for explanation.

[0152] Specifically, during the training phase, a certain number of third training datasets are pre-configured. These third training datasets include multiple sets of third training data, each set comprising third sample data and third label data. The third sample data includes historical images showing the visible portions and associated historical dirt information characterizing the historical dirt levels of those portions. The third label data includes historical prompts based on actual cleaning needs (such as labels like "Needs cleaning" or "Current condition is good, no cleaning required," marked by maintenance personnel according to the actual condition). Through supervised learning training, the third association model can learn the complex nonlinear relationship between the image features of the visible portions, dirt information, and cleaning needs.

[0153] It should be noted that in this embodiment, step ST5 is used as an alternative to step ST3, but the method is not limited to this. That is, the dirt indication method described in this embodiment may also include both step ST3 and step ST5, and they may be executed simultaneously or sequentially.

[0154] Furthermore, in this embodiment, the image of the visible portion can be multiple local images showing a local area of ​​the visible portion, and the dirt information can be local dirt information associated with each local image. Based on this, the processing related to local images and local dirt information described in the first to third modifications of the dirt estimation method of the above embodiments, the processing related to the output step described in the fourth modification, and the processing described in the fifth modification are all applicable to the dirt indication method described in this embodiment.

[0155] Specifically, based on this embodiment, a further modification of the first variation of the dirt estimation method of the above embodiment includes step ST5A instead of step ST5. In step ST5A, based on multiple local images acquired in real time and local dirt information associated with each local image and characterizing the dirt status of local areas of the visible portion shown in each local image, acquired in step ST2A, a prompt is given as to whether the air conditioning unit or the visible portion needs cleaning. Furthermore, unlike the second embodiment, in this modification, the third sample data in the third training dataset includes multiple historical local images representing each local area of ​​the visible portion and historical local dirt information characterizing the historical dirt status of the local areas shown in each local image.

[0156] Based on this embodiment, a further variation of the second modification of the dirt estimation method of the above embodiment includes step ST5B instead of step ST5. In step ST5B, the dirt distribution information of the entire visible portion generated in step ST2C is input as input data to the third association model to indicate whether the air conditioning unit or the visible portion needs cleaning.

[0157] Based on this embodiment, as a further variation of the third variation of the first embodiment described above, step ST5C is included instead of step ST5. In step ST5C, based on the multiple local images obtained in step ST1A, the local dirt information associated with each local image and characterizing the dirt status of the local area shown in the local image obtained in step ST2A, and the location information obtained in step ST2D, a prompt is given as to whether the air conditioning unit or visible parts need cleaning.

[0158] Based on this embodiment, a further variation of the fourth variation of the dirt estimation method of the first embodiment described above includes step ST4A. In step ST4A, the prompt information indicating whether the air conditioning device or visible part needs cleaning, generated by any one of steps ST5, ST5A, ST5B, and ST5C, is output together with the corresponding image or partial image and the dirt information or partial dirt information associated with the image or partial image.

[0159] Based on this embodiment, a further modification of the fifth variation of the dirt estimation method of the first embodiment described above includes step ST5D instead of step ST5. In step ST5D, at least one of the following is used to indicate whether the air conditioning unit or visible parts need cleaning: the image or partial image obtained in step ST1, the dirt information or partial dirt information obtained in step ST2AA, and the type information, usage duration information, and usage environment information of the air conditioning unit.

[0160] Furthermore, this application also provides a dirt warning module according to one embodiment, which includes an input unit, an acquisition unit, and a warning unit. The input unit is configured to accept an image of a visible portion of an air conditioning unit as input. The acquisition unit is configured to acquire dirt information associated with the image, characterizing the dirt condition of the visible portion. The estimation unit is configured to provide a warning based on the image and the dirt information, indicating whether the air conditioning unit or the visible portion needs cleaning.

[0161] In another aspect, this application also provides a computer-readable storage medium, which may be included in the computer device described in the above embodiments and their variations, or it may exist independently and not assembled into the computer device. The computer-readable storage medium carries one or more programs that, when executed by the computer device, cause the computer system to implement the methods as described in the above embodiments and their variations. For example, the computer system can implement the steps shown in the figures.

[0162] According to one aspect of this application, a computer program product is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in various optional implementations of the above embodiments.

[0163] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method of estimating dirtiness of an air conditioning device, characterized by, comprises: a first step of acquiring an image of a visible portion of the air conditioning device; a second step of acquiring dirt information associated with the image and characterizing a dirt condition of the visible portion; and a third step of inferring a dirt condition of a portion other than the visible portion of the air conditioning device based on the image and the dirt information.

2. The dirt inference method according to claim 1, wherein in the first step, a plurality of partial images respectively showing local regions of the visible portion are acquired, in the second step, partial dirt information associated with each partial image and characterizing a dirt condition of the local region is acquired, and in the third step, a dirt condition of a portion other than the visible portion is inferred based on the plurality of partial images and the partial dirt information associated with each partial image.

3. The dirt inference method according to claim 2, further comprising a positioning step in which, based on a template image corresponding to the visible portion and pre-stored, the position information of the local region shown by each partial image on the visible portion is determined, the template image including position tags for indicating different regions, in the third step, a dirt condition of a portion other than the visible portion is inferred based on the partial image, the position information corresponding to the partial image, and the partial dirt information associated with the partial image.

4. The dirt inference method according to claim 2 or 3, wherein acquiring the partial dirt information comprises: inputting each of the partial images into a first correlation model pre-trained, and causing the first correlation model to output, based on each of the input partial images, the partial dirt information associated with the partial image, wherein the first correlation model is obtained based on a first training data set, the first training data set including a plurality of groups of first training data, each group of first training data including first sample data and first label data, the first sample data includes a partial image showing a local region of the visible portion, the first label data includes partial dirt information associated with the partial image and characterizing a dirt condition of the local region of the visible portion.

5. The dirt inference method according to claim 1, wherein the visible portion is a portion that can be observed from at least one of an air inlet and an air outlet of the air conditioning device, the portion other than the visible portion is a non-visible portion that cannot be observed from both the air inlet and the air outlet.

6. The dirt inference method according to claim 5, wherein the air conditioning device is a ducted air conditioning device including a heat exchanger and a fan scroll, the visible portion is at least a portion of the heat exchanger observed from the air outlet or at least a portion of the fan scroll observed from the air inlet, the non-visible portion is another portion of the fan scroll that cannot be observed from the air inlet and the air outlet.

7. The dirt inference method according to claim 1, wherein ​ ​ In the second step, at least one of type information, usage time length, and usage environment of the air conditioning device is also acquired, In the third step, based on the image, the dirt information, and at least one of the type information, the usage time length, and the usage environment, a dirt condition of a portion other than the visible portion is estimated.

8. The dirt estimation method according to any one of claims 1 to 7, wherein In the third step, includes: inputting the image and the dirt information into a second correlation model that is pre-trained; and causing the second correlation model to output, based on the input image and the dirt information, an estimation result that represents a dirt condition of a portion other than the visible portion, wherein the second correlation model is obtained by training based on a second training data set, the second training data set including a plurality of second training data sets, each second training data set including second sample data and second label data, the second sample data includes an image showing the visible portion and dirt information associated with the image and representing a dirt condition of the visible portion, the second label data includes data associated with the second sample data and representing a dirt condition of a portion other than the visible portion.

9. The dirt estimation method according to any one of claims 1 to 7, further comprising an output step in which the estimation result that represents a dirt condition of the invisible portion is output in association with the image and the dirt information associated with the image. includes:

10. A dirtiness estimation module of an air conditioning device, characterized by, an input unit configured to input an image of a visible portion of the air conditioning device; an acquisition unit configured to acquire dirt information associated with the image and representing a dirt condition of the visible portion; an estimation unit configured to estimate, based on the image and the dirt information, a dirt condition of a portion other than the visible portion of the air conditioning device as an estimation result; and an output unit configured to output the estimation result. includes: an input module configured to input an image showing a visible portion of the air conditioning device and dirt information associated with the image and representing a dirt condition of the visible portion; 11. An estimation model for estimating a dirty condition of a portion of an air conditioning apparatus other than a visible portion, characterized by, a processing module configured to generate, based on the input image and the dirt information, an estimation result that represents a dirt condition of a portion other than the visible portion; and an output module configured to output the estimation result, wherein the estimation model is obtained by training based on a training data set, the training data set including a plurality of training data sets, each training data set including sample data and label data, the sample data includes an image showing the visible portion and dirt information associated with the image and representing a dirt condition of the visible portion, the label data includes data associated with the sample data and representing a dirt condition of a portion other than the visible portion. includes: a first step of acquiring an image of a visible portion of the air conditioning device; ​ 12. A dirty prompt method of an air conditioning apparatus, characterized by, ​ ​ a second step of acquiring dirt information associated with the image and characterizing a dirt condition of the visible part; and a fourth step of prompting whether the air conditioning device or the visible part needs cleaning based on the image and the dirt information.

13. The dirt prompting method according to claim 12, wherein in the fourth step, comprising: inputting the image and the dirt information into a third correlation model trained in advance; and causing the third correlation model to output prompting information for prompting whether the air conditioning device or the visible part needs cleaning based on the input image and the dirt information, wherein the third correlation model is obtained based on a third training data set, the third training data set including a plurality of third training data, each of the third training data including third sample data and third label data, the third sample data including an image showing the visible part and dirt information associated with the image and characterizing a dirt condition of the visible part, the third label data including prompting information associated with the third sample data and prompting that the air conditioning device or the visible part needs cleaning.

14. The dirt prompting method according to claim 12 or 13, wherein in the first step, a plurality of partial images showing partial regions of the visible part, respectively, are acquired, in the second step, partial dirt information associated with each partial image and characterizing a dirt condition of the partial region is acquired, and in the third step, whether the air conditioning device or the visible part needs cleaning is prompted based on the plurality of partial images and the partial dirt information associated with each partial image.

15. The dirt prompting method according to claim 14, wherein further comprising a positioning step in which, based on a template image corresponding to the visible part and stored in advance, position information of a partial region shown by each partial image on the visible part is determined, the template image including position labels for indicating different regions, in the fourth step, whether the air conditioning device or the visible part needs cleaning is prompted based on the partial image, the position information corresponding to the partial image, and the partial dirt information associated with the partial image.

16. A dirty prompt module of an air conditioning apparatus, characterized by comprising: comprising: an input unit configured to input an image of a visible part of the air conditioning device; an acquisition unit configured to acquire dirt information associated with the image and characterizing a dirt condition of the visible part; and a prompting unit configured to prompt whether the air conditioning device or the visible part needs cleaning based on the image and the dirt information.

17. A computer system comprising a processor and a memory storing a computer program, wherein the computer program is processed by the processor to implement the dirt estimation method according to any one of claims 1 to 9 or the dirt prompting method according to any one of claims 12 to 15.

18. A computer-readable storage medium storing a computer program, the computer program being processed by a processor to implement the dirtiness estimation method according to any one of claims 1-9 or the dirtiness prompting method according to any one of claims 12-15.

19. A computer program product comprising a computer program, the computer program being processed by a processor to implement the dirtiness estimation method according to any one of claims 1-9 or the dirtiness prompting method according to any one of claims 12-15. ​ ​

Images