Air conditioner condenser surface mildew spot characterization intelligent identification and cleanliness early warning method

Abstract

The invention discloses an air conditioner condenser surface mildew spot characterization intelligent identification and cleaning early warning method, and relates to the technical field of intelligent identification and cleaning early warning. Based on the image processing and machine learning technology, the efficient, rapid and low-cost air conditioner condenser surface mildew spot characterization intelligent identification and cleanliness early warning method is established, the air conditioner condenser surface mildew spot identification efficiency is greatly improved, the condenser cleanliness early warning system can automatically set a corresponding threshold value, and the air conditioner condenser surface mildew spot characterization intelligent identification and cleanliness early warning efficiency is improved. Therefore, early warning of different degrees of cleanliness is carried out.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent identification and cleaning early warning, and specifically to an intelligent identification and cleaning early warning method for mildew characteristics on the surface of an air-conditioning condenser. Background Art

[0002] At present, the intelligent identification and cleanliness warning methods for the characterization of mildew on the surface of air-conditioning condensers mainly use a system and method for detecting and processing mildew by an intelligent mobile robot. The system includes an intelligent mobile robot and an intelligent terminal. The intelligent mobile robot includes a mildew detection module, a mildew processing module and a chip. The mildew detection module is connected to the chip to collect environmental images and send them to the chip for mildew detection. The mildew processing module is connected to the chip to process mildew. The chip is used to process the data of the mildew detection module and the mildew processing module and to control the work of the mildew detection module and the mildew processing module. The intelligent terminal is installed with software that can communicate with the robot. The software is used to receive and save notifications sent by the robot and send instructions to the robot.

[0003] The existing intelligent mobile robot system for detecting and treating mildew has the disadvantages of poor detection accuracy and adaptability, high hardware equipment cost, difficult maintenance, insufficient operation speed and real-time performance, limited multi-tasking execution and mobility capabilities, low intelligence level, and poor environmental adaptability.

[0004] Therefore, a new solution to the above problems needs to be proposed. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent identification method for mildew characterization on the surface of an air-conditioning condenser and a cleanliness early warning method to solve the technical problems raised in the background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: an intelligent identification method for characterizing mildew on the surface of an air conditioner condenser, comprising at least the following steps:

[0007] S1: Obtain image data of mildew on the surface of air conditioner condenser components through various micro high-definition cameras or other means to form a condenser mildew characterization database;

[0008] S2: Combine machine vision and learning algorithms to build an intelligent recognition model for condenser mildew characterization;

[0009] S3: Inputting the condenser mildew image to be processed into the established condenser mildew characterization intelligent recognition model to obtain a mildew characterization image recognition result.

[0010] Furthermore, before the image to be processed is input into the established intelligent recognition model for characterizing mildew spots on condensers, the following steps are also included:

[0011] Acquire and form a training data set of condenser mildew characterization images from a condenser mildew characterization database, and divide the training data set into a training set, a test set, and a validation set in proportion;

[0012] Classify the morphological features of mold spots in the training data set and generate multiple sets of training images;

[0013] Based on the image morphological features, the image recognition model is trained.

[0014] Furthermore, the construction of the intelligent recognition model for characterizing mildew spots on condensers includes at least the following steps:

[0015] A feature extraction model based on deep convolutional neural network (CNN) is used to cut the condenser mildew image, and ResNet is used as the backbone network to extract image features.

[0016] Use FPN (Feature Pyramid Network) structure for multi-scale feature fusion;

[0017] Image prediction and output are completed by a classifier, wherein the classifier includes at least Faster R-CNN;

[0018] And use VIA tool to annotate the condenser mildew characterization training images, generate annotated data files, and convert them into CSV format;

[0019] Finally, the labeled data files are trained through the training framework, an intelligent recognition model for condenser mildew is constructed, and the mildew recognition is realized through the detection algorithm.

[0020] Furthermore, the model building algorithm of the condenser mildew characterization intelligent recognition model includes:

[0021] The convolution layer extracts the mildew features from the input condenser mildew image. Given a condenser mildew input image I and a convolution kernel K, the convolution kernel is equivalent to a specific feature detector used to scan every position of the input mildew image. The convolution operation can be expressed as:

[0022] O(i,j)=∑ m ∑ n I(i+m,j+n)·K(m,n)

[0023] I(i,j) represents the pixel value of the input mildew image at position (i,j);

[0024] K(m,n) represents the value of the convolution kernel at position (m,n);

[0025] O(i,j) represents the value of the output mildew feature map at position (i,j);

[0026] Activation function, using activation function (ReLU) to introduce nonlinearity, reduce recognition training time and computing cost, thereby improving the training efficiency and performance of the condenser mildew characterization intelligent recognition model. The ReLU function is expressed as:

[0027] ReLU(x)=max(0,x)

[0028] If x is greater than 0, the ReLU function outputs x, if x is less than 0, the ReLU function outputs 0;

[0029] The pooling layer downsamples the output of the convolutional layer and uses maximum pooling to reduce the spatial dimension of the mildew feature map and extract the main feature texture information of the image, which is expressed as:

[0030]

[0031] pool represents the size of the pooling area of ​​the mold feature map;

[0032] For each position in the pooling area, the maximum value is taken, i.e., maximum pooling, to generate the main feature texture information of the output mildew image;

[0033] The fully connected layer integrates the high-dimensional feature vector into a global feature expression by weighted summing the condenser mildew features and introducing a nonlinear activation function. It uses trainable weights and biases to map the features to the sample label space and outputs the predicted probability distribution of each category, thereby completing the mildew classification task. It effectively enhances the model's discrimination ability and the ability to fit complex features, and ensures the correlation between the features and the classification target, which can be expressed as:

[0034] y=ReLU(wx+b)

[0035] y is the output result of the fully connected layer, which represents the output vector after weighted summation and activation function processing, corresponding to the predicted value or probability distribution of each mold image category;

[0036] w is a weight parameter matrix with a size of n×m, which reflects the relationship between input features and output results, where n is the dimension of the mold feature map input, and m is the dimension of the mold feature map output, which is related to the number of classification categories of the mold image;

[0037] x is the input vector, which represents the feature vector of mildew obtained after being processed by the previous layer;

[0038] b is a bias vector with a size of 1×m. Introducing a bias can enhance the expressiveness of the model and help adjust the baseline value of the output;

[0039] Loss function,In the training process of convolutional neural network (CNN), the cross entropy function is used to measure the gap between the prediction results of the condenser mildew characterization intelligent recognition model and the actual mildew classification results. It can effectively quantify the inconsistency between the probability distribution predicted by the model and the true label distribution. The model can gradually optimize the parameters, thereby improving the classification accuracy and generalization ability of the mildew category. The cross entropy loss function is expressed as:

[0040]

[0041] u i It is the result of actual mold classification. is the model's predicted probability for the i-th type of mold stain;

[0042] The optimization algorithm trains the CNN model by minimizing the loss function. The most basic optimization algorithm, the gradient descent method, is used. The gradient of the loss function to the condenser mildew feature recognition model parameters is calculated to guide the parameters to be updated in the direction of reducing the loss. Through repeated iterative optimization, the minimum value of the objective function is gradually approached, thereby optimizing the model parameters so that the mildew recognition can be completed more accurately. The update rule is:

[0043]

[0044] W is the weight parameter matrix of the mold spot feature recognition model, which is used to represent the trainable parameters of the model; η is the learning rate, which is used to control the step size of each weight update and affects the optimization convergence speed; It is the gradient of the loss function L with respect to the weight parameter matrix W, which represents the directional derivative of the current model on the weight W and is used to guide the direction of weight update. Each update of W gradually reduces the value of the loss function L and eventually approaches the optimal value, thereby optimizing the performance of the condenser mildew feature recognition model.

[0045] Furthermore, the S3 at least includes the following steps:

[0046] Recording with a high-definition camera device to obtain an image of the mildew on the surface of the air-conditioning condenser component to be processed;

[0047] The image is adjusted to an input resolution of the condenser mildew characterization intelligent recognition model, and is input into the condenser mildew characterization intelligent recognition model for processing;

[0048] The mold stain characterization image recognition results are obtained; and the recognition results are verified and analyzed through high-precision detection means.

[0049] Furthermore, verification, evaluation and optimization are also included, and the verification, evaluation and optimization at least include the following steps:

[0050] Verify and analyze the identification results through high-precision detection means, and the verification also includes verifying the mildew stain characterization through other high-precision detection means;

[0051] Compare the results of the validation process with the output of the intelligent recognition model for condenser mildew characterization to evaluate and optimize the reliability of the model and ensure its accuracy and stability;

[0052] The optimized model will be added to the database of typical features of condenser mildew to improve the accuracy and efficiency of subsequent identification;

[0053] In order to improve training speed and recognition efficiency, the size of the model and input image in the machine vision learning algorithm should be optimized with reference to the computer configuration.

[0054] The cleanliness early warning method for intelligently identifying mildew characteristics on the surface of an air conditioner condenser comprises at least the following steps:

[0055] Build a risk prediction model for condenser mildew characterization pollution;

[0056] An intelligent cleanliness early warning system is formed by a condenser mildew characterization pollution risk prediction model and a prompt module, wherein the prompt module at least includes an indicator light, an LCD screen, and a mobile phone prompt;

[0057] The condenser mildew characterization pollution risk prediction model is used to monitor and analyze the component condenser mildew characterization pollution risk in real time. When the mildew pollution condition of the air-conditioning system condenser reaches the threshold, the prompt module will issue a cleanliness warning to remind the user that they should take measures to clean the air-conditioning condenser components.

[0058] Furthermore, the construction of the condenser mildew characterization pollution risk prediction model includes at least the following steps:

[0059] Input the pre-processed condenser mildew image;

[0060] The contamination changes of condenser mildew characterization were modeled by combining the deep learning algorithm convolutional neural network (CNN);

[0061] Evaluate the accuracy of the established model and classify the risk level;

[0062] Formation of a contamination risk prediction model for condenser mildew characterization.

[0063] Furthermore, the condenser mildew characterization risk prediction model adopts decision tree and random forest algorithms, which are suitable for processing classification problems of mildew characterization with complex nonlinear relationships, can process a large number of input features, and automatically perform feature selection;

[0064] The decision tree algorithm achieves classification or regression of condenser mildew features by recursively dividing the mildew image data space;

[0065] The condenser mildew characterization risk prediction model selects the features that can best distinguish mildew data, and constructs a tree model with a branching structure layer by layer. In each division process, an indicator is used to measure the importance of the feature. The indicator includes at least information gain, information gain ratio or Gini index. The feature with the largest information gain or the feature with the smallest Gini index is preferentially selected for data division.

[0066] The partitioning process continues until a preset stop condition is met, wherein the stop condition includes at least the depth of the tree, the number of samples of the leaf node, or the information gain threshold, thereby generating an explanatory and efficient decision rule to complete the classification or prediction task of the mold spot characteristics;

[0067] The node partitioning formula is based on information gain or Gini index;

[0068] Information Gain:

[0069]

[0070]

[0071] IG(T,A) is used to measure the effectiveness of selecting the mold feature A for classifying the mold image dataset T. The larger the value, the more important the feature A is for classification.

[0072] Entropy(T) is the entropy of the mold image dataset T, which measures the uncertainty or confusion of T. v is a possible value of feature A;

[0073] Values(A) represents the set of all possible values ​​of feature A. v It is a subset of the feature A value v in the mold image dataset T;

[0074] is a subset T of feature values ​​v v The ratio relative to the overall data set T indicates the weight after division;

[0075] Entropy v ) represents the subset T v The entropy of , which indicates the purity or uncertainty of the subset;

[0076] c represents the total number of classification categories;

[0077] p i is the probability of the mold image category i, that is, the proportion of samples belonging to the i-th category in the mold image dataset;

[0078] Gini Index:

[0079]

[0080] Gini(T) represents the Gini index of the mold image dataset, which measures the accuracy of the dataset. The smaller the value, the higher the accuracy of the dataset.

[0081] pi is the probability of the mold image category i;

[0082] The random forest algorithm predicts the risk of mildew characterization of condensers by constructing multiple decision trees. Each decision tree uses a subset generated by random sampling during training, thereby introducing the randomness of data and features to enhance the robustness and generalization ability of the model. The prediction results of all decision trees are voted (classification task) or averaged (regression task) to obtain a comprehensive prediction result. In addition, the model performance is evaluated through "out-of-bag data" to effectively reduce the variance of the model and avoid overfitting, thereby improving the accuracy and stability of mildew characterization risk prediction;

[0083] The classification of the final prediction formula is:

[0084]

[0085] The final prediction formula for regression is:

[0086]

[0087] in is the prediction result of the mold spot characterization risk of the i-th decision tree, and N is the number of decision trees.

[0088] Compared with the prior art, the present invention has the following beneficial effects:

[0089] 1. Based on image processing and machine learning technology, the present invention establishes an efficient, fast and low-cost intelligent recognition and cleanliness warning method for mildew characterization on the surface of air-conditioning condensers, which greatly improves the efficiency of mildew recognition on the surface of air-conditioning condensers. The condenser cleanliness warning system can set corresponding thresholds by itself, thereby providing warnings of different degrees of cleanliness;

[0090] 2. The present invention can intelligently identify the mildew spots on the surface of the condenser, and capture the distribution and growth of the mildew spots in real time. The identification process has high accuracy and can distinguish the type, color, morphology and other characteristics of the mildew spots. This precise positioning provides a reliable basis for mildew spot treatment and can effectively avoid false detection and missed detection. When the changing trend of mildew spots on the condenser is detected, the system can issue an early warning to remind cleaning and maintenance operations. This early warning function not only helps to prevent mold growth in advance, but also can take measures before mildew spots form to ensure the cleanliness of the air-conditioning condenser. The intelligent identification and early warning method of mildew spot characterization has important advantages in the clean management of condensers through technical means such as automation, high precision, and real-time response. This technology can identify the formation trend of mildew spots on condensers in advance, provide a scientific basis for the maintenance and cleaning of air-conditioning equipment, extend the service life of air conditioners, improve the operating efficiency of equipment, improve the life and performance of air-conditioning systems, and effectively ensure the continuous stability of air quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0092] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0093] Figure 2 This is a schematic diagram of the intelligent identification process of mildew characterization of air conditioner condensers according to the present invention;

[0094] Figure 3 This is a schematic diagram of the risk prediction and cleanliness early warning process of mildew characterization of air conditioner condenser according to the present invention;

[0095] Figure 4 Schematic diagram of the hardware system of the present invention. DETAILED DESCRIPTION

[0096] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0097] Embodiment 1:

[0098] See also Figure 1 - Figure 2 , the intelligent identification method of mildew characterization on the surface of air conditioner condenser includes at least the following steps:

[0099] S1: Obtain image data of mildew on the surface of air conditioner condenser components through various micro high-definition cameras or other means to form a condenser mildew characterization database;

[0100] S2: Combine machine vision and learning algorithms to build an intelligent recognition model for condenser mildew characterization;

[0101] S3: Inputting the condenser mildew image to be processed into the established condenser mildew characterization intelligent recognition model to obtain a mildew characterization image recognition result.

[0102] Before the image to be processed is input into the established intelligent recognition model for characterizing mildew spots on condensers, the following steps are also included:

[0103] Acquire and form a training data set of condenser mildew characterization images from a condenser mildew characterization database, and divide the training data set into a training set, a test set, and a validation set in proportion;

[0104] Classify the morphological features of mold spots in the training data set and generate multiple sets of training images;

[0105] Based on the image morphological features, the image recognition model is trained.

[0106] The construction of the intelligent recognition model for condenser mildew characterization includes at least the following steps:

[0107] A feature extraction model based on deep convolutional neural network (CNN) is used to cut the condenser mildew image, and ResNet is used as the backbone network to extract image features.

[0108] Use FPN (Feature Pyramid Network) structure for multi-scale feature fusion;

[0109] Image prediction and output are completed through classifiers, which include at least Faster R-CNN;

[0110] And use VIA tool to annotate the condenser mildew characterization training images, generate annotated data files, and convert them into CSV format;

[0111] Finally, the labeled data files are trained through the training framework, an intelligent recognition model for condenser mildew is constructed, and the mildew recognition is realized through the detection algorithm.

[0112] The model building algorithm of the intelligent recognition model for condenser mildew characterization includes:

[0113] The convolution layer extracts the mildew features from the input condenser mildew image. Given a condenser mildew input image I and a convolution kernel K, the convolution kernel is equivalent to a specific feature detector used to scan every position of the input mildew image. The convolution operation can be expressed as:

[0114] O(i,j)=∑ m ∑ n I(i+m,j+n)·K(m,n)

[0115] I(i,j) represents the pixel value of the input mildew image at position (i,j);

[0116] K(m,n) represents the value of the convolution kernel at position (m,n);

[0117] O(i,j) represents the value of the output mildew feature map at position (i,j);

[0118] Activation function, using activation function (ReLU) to introduce nonlinearity, reduce recognition training time and computing cost, thereby improving the training efficiency and performance of the condenser mildew characterization intelligent recognition model. The ReLU function is expressed as:

[0119] ReLU(x)=max(0,x)

[0120] If x is greater than 0, the ReLU function outputs x, if x is less than 0, the ReLU function outputs 0;

[0121] The pooling layer downsamples the output of the convolutional layer and uses maximum pooling to reduce the spatial dimension of the mildew feature map and extract the main feature texture information of the image, which is expressed as:

[0122]

[0123] Pool represents the size of the pooling area of ​​the mildew feature map. The pooling window is equivalent to a rectangular area of ​​the mildew feature map. The 2x2 or 3x3 pooling window slides on the feature map, and the sliding stride is 2 or 3 units each time.

[0124] For each position in the pooling area, the maximum value is taken, i.e., maximum pooling, to generate the main feature texture information of the output mildew image;

[0125] The fully connected layer integrates the high-dimensional feature vector into a global feature expression by weighted summing the condenser mildew features and introducing a nonlinear activation function. It uses trainable weights and biases to map the features to the sample label space and outputs the predicted probability distribution of each category, thereby completing the mildew classification task. It effectively enhances the model's discrimination ability and the ability to fit complex features, and ensures the correlation between the features and the classification target, which can be expressed as:

[0126] y=ReLU(wx+b)

[0127] y is the output result of the fully connected layer, which represents the output vector after weighted summation and activation function processing, corresponding to the predicted value or probability distribution of each mold image category;

[0128] w is a weight parameter matrix with a size of n×m, which reflects the relationship between input features and output results, where n is the dimension of the mold feature map input, and m is the dimension of the mold feature map output, which is related to the number of classification categories of the mold image;

[0129] x is the input vector, which represents the feature vector of mildew obtained after being processed by the previous layer;

[0130] b is a bias vector with a size of 1×m. Introducing a bias can enhance the expressiveness of the model and help adjust the baseline value of the output;

[0131] Loss function,In the training process of convolutional neural network (CNN), the cross entropy function is used to measure the gap between the prediction results of the condenser mildew characterization intelligent recognition model and the actual mildew classification results. It can effectively quantify the inconsistency between the probability distribution predicted by the model and the true label distribution. The model can gradually optimize the parameters, thereby improving the classification accuracy and generalization ability of the mildew category. The cross entropy loss function is expressed as:

[0132]

[0133] u i It is the result of actual mold classification. is the model's predicted probability for the i-th type of mold stain;

[0134] The optimization algorithm trains the CNN model by minimizing the loss function. The most basic optimization algorithm, the gradient descent method, is used. The gradient of the loss function to the condenser mildew feature recognition model parameters is calculated to guide the parameters to be updated in the direction of reducing the loss. Through repeated iterative optimization, the minimum value of the objective function is gradually approached, thereby optimizing the model parameters so that the mildew recognition can be completed more accurately. The update rule is:

[0135]

[0136] W is the weight parameter matrix of the mold spot feature recognition model, which is used to represent the trainable parameters of the model; η is the learning rate, which is used to control the step size of each weight update and affects the optimization convergence speed; It is the gradient of the loss function L with respect to the weight parameter matrix W, which represents the directional derivative of the current model on the weight W and is used to guide the direction of weight update. Each update of W gradually reduces the value of the loss function L and eventually approaches the optimal value, thereby optimizing the performance of the condenser mildew feature recognition model.

[0137] S3 includes at least the following steps:

[0138] Recording with a high-definition camera device to obtain an image of the mildew on the surface of the air-conditioning condenser component to be processed;

[0139] The image is adjusted to an input resolution of the condenser mildew characterization intelligent recognition model, and is input into the condenser mildew characterization intelligent recognition model for processing;

[0140] The mold stain characterization image recognition results are obtained; and the recognition results are verified and analyzed through high-precision detection means.

[0141] It also includes verification, evaluation and optimization, which at least include the following steps:

[0142] Verify and analyze the identification results through high-precision detection means, and the verification also includes verifying the mildew stain characterization through other high-precision detection means;

[0143] Compare the results of the validation process with the output of the intelligent recognition model for condenser mildew characterization to evaluate and optimize the reliability of the model and ensure its accuracy and stability;

[0144] The optimized model will be added to the database of typical features of condenser mildew to improve the accuracy and efficiency of subsequent identification;

[0145] In order to improve training speed and recognition efficiency, the size of the model and input image in the machine vision learning algorithm should be optimized with reference to the computer configuration.

[0146] Embodiment 2:

[0147] See also Figure 3 Based on the above embodiment, this embodiment proposes a cleanliness warning method for intelligently identifying mildew characteristics on the surface of an air conditioner condenser;

[0148] The cleanliness early warning method for intelligently identifying mildew characteristics on the surface of an air conditioner condenser comprises at least the following steps:

[0149] Build a risk prediction model for condenser mildew characterization pollution;

[0150] An intelligent cleanliness early warning system is formed by a condenser mildew characterization pollution risk prediction model and a prompt module, wherein the prompt module at least includes an indicator light, an LCD screen, and a mobile phone prompt;

[0151] The condenser mildew characterization pollution risk prediction model is used to monitor and analyze the component condenser mildew characterization pollution risk in real time. When the mildew pollution condition of the air-conditioning system condenser reaches the threshold, the prompt module will issue a cleanliness warning to remind the user that they should take measures to clean the air-conditioning condenser components.

[0152] The establishment of a condenser mildew characterization pollution risk prediction model includes at least the following steps:

[0153] Input the pre-processed condenser mildew image;

[0154] The contamination changes of condenser mildew characterization were modeled by combining the deep learning algorithm convolutional neural network (CNN);

[0155] Evaluate the accuracy of the established model and classify the risk level;

[0156] Formation of a contamination risk prediction model for condenser mildew characterization.

[0157] The decision tree and random forest algorithms are used in the condenser mildew characterization risk prediction model, which is suitable for processing the classification problem of mildew characterization with complex nonlinear relationships, can handle a large number of input features, and automatically perform feature selection;

[0158] The decision tree algorithm achieves classification or regression of condenser mildew features by recursively dividing the mildew image data space;

[0159] The condenser mildew characterization risk prediction model selects the features that can best distinguish mildew data, and constructs a tree model with a branching structure layer by layer. In each division process, an indicator is used to measure the importance of the feature. The indicator includes at least information gain, information gain ratio or Gini index. The feature with the largest information gain or the smallest Gini index is preferentially selected for data division.

[0160] The partitioning process continues until the preset stopping condition is met, which includes at least the tree depth, the number of samples of leaf nodes or the information gain threshold, thereby generating explanatory and efficient decision rules to complete the classification or prediction task of mold spot characteristics;

[0161] The node partitioning formula is based on information gain or Gini index;

[0162] Information Gain:

[0163]

[0164]

[0165] IG(T,A) is used to measure the effectiveness of selecting the mold feature A for classifying the mold image dataset T. The larger the value, the more important the feature A is for classification.

[0166] Entropy(T) is the entropy of the mold image dataset T, which measures the uncertainty or confusion of T. v is a possible value of feature A;

[0167] Values(A) represents the set of all possible values ​​of feature A. v It is a subset of the feature A value v in the mold image dataset T;

[0168] is a subset T of feature values ​​v v The ratio relative to the overall data set T indicates the weight after division;

[0169] Entropy v ) represents the subset T v The entropy of , which indicates the purity or uncertainty of the subset;

[0170] c represents the total number of classification categories;

[0171] p i is the probability of the mold image category i, that is, the proportion of samples belonging to the i-th category in the mold image dataset;

[0172] Gini Index:

[0173]

[0174] Gini(T) represents the Gini index of the mold image dataset, which measures the accuracy of the dataset. The smaller the value, the higher the accuracy of the dataset.

[0175] pi is the probability of the mold image category i;

[0176] The random forest algorithm predicts the risk of mildew characterization of condensers by constructing multiple decision trees. Each decision tree uses a subset generated by random sampling during training, thereby introducing the randomness of data and features to enhance the robustness and generalization ability of the model. The prediction results of all decision trees are voted (classification task) or averaged (regression task) to obtain a comprehensive prediction result. In addition, the model performance is evaluated through "out-of-bag data" to effectively reduce the variance of the model and avoid overfitting, thereby improving the accuracy and stability of mildew characterization risk prediction;

[0177] The classification of the final prediction formula is:

[0178]

[0179] The final prediction formula for regression is:

[0180]

[0181] in is the prediction result of the mold spot characterization risk of the i-th decision tree, and N is the number of decision trees.

[0182] Embodiment three:

[0183] See also Figure 4 , based on the above two embodiments, a composite system is proposed;

[0184] The system includes a mildew quantity intelligent identification system, a mildew data communication module, and a cleanliness early warning module;

[0185] The intelligent identification system for the amount of mildew spots includes a mildew spot image acquisition module, an image data communication module and a data storage and processing system;

[0186] The image data communication module is used to transmit the collected images of the mildew image acquisition module to the data storage and processing system, and the condenser mildew characterization intelligent recognition model in the data storage and processing system is used to achieve recognition. The recognized data is transmitted to the cleanliness warning module through the mildew data communication module. The cleanliness warning module is equipped with an intelligent cleanliness warning system. The condenser mildew characterization risk prediction model in the intelligent cleanliness warning system is used to judge whether the threshold is reached, and then determine whether an alarm is needed.

[0187] In summary, this case obtains image information of mildew spots on air conditioner condensers based on experiments or related methods, combines and processes them, and thus establishes a typical feature database of mildew spots on air conditioner condensers. By introducing integrated machine vision and learning algorithms, a basic mildew spot image recognition model is established. After processing the feature information and image information of mildew spots on condensers, they are input into the mildew spot basic image recognition model for targeted training, forming an intelligent recognition basic model for mildew spots on condensers.

[0188] Subsequently, the model is used to perform image recognition tasks and generate recognition results. The parameters of the basic mildew image recognition model are set according to the equipment configuration and image size, and finally a basic image intelligent recognition model is formed that can accurately identify the typical characteristics of condenser mildew in the database and has a certain recognition ability for new images.

[0189] Based on the basic model for intelligent recognition of mildew on condensers, when the training result reaches a sufficient confidence level, the image to be processed is obtained and input into the intelligent recognition model for mildew on condensers for recognition, so as to obtain the recognition result of the image to be processed; at the same time, the recognition results are randomly sampled for verification analysis to eliminate the influence of uncertain factors, improve the accuracy of the intelligent recognition results of mildew, verify the reliability of the model, and complete the iteration.

[0190] Based on the basic model and key algorithms of intelligent recognition, the changing characteristics of condenser mildew contamination are modeled in combination with the deep learning algorithm convolutional neural network (CNN). The condenser mildew contamination risk prediction model is constructed using the random forest ensemble learning method to improve the accuracy and stability of the model. Through intelligent recognition and real-time monitoring, the mold and mildew spots on the surface of the condenser are intelligently identified after image processing, and the distribution and growth of the mildew spots are captured in real time.

[0191] The early warning system allows different mildew growth thresholds to be set, thereby triggering different degrees of cleanliness warnings. When the mildew contamination condition of the air-conditioning system condenser reaches a certain threshold, an early warning will be issued through an indicator light or SMS notification, reminding the user that measures should be taken to clean the air-conditioning condenser components.

[0192] Therefore, efficient image processing algorithms and intelligent recognition technology can be used to intelligently identify the characteristics of mildew spots on the condenser surface, and by analyzing the distribution and changes of mildew spots, real-time monitoring and early warning of the condenser cleanliness can be achieved.

[0193] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. An intelligent identification method for characterizing mildew on the surface of an air conditioner condenser, characterized by: At least the following steps are included: S1: Obtain image data of mildew on the surface of air conditioner condenser components through various micro high-definition cameras or other means to form a condenser mildew characterization database; S2: Combine machine vision and learning algorithms to build an intelligent recognition model for condenser mildew characterization; S3: Inputting the condenser mildew image to be processed into the established condenser mildew characterization intelligent recognition model to obtain a mildew characterization image recognition result.

2. The intelligent identification method for mildew characterization on the surface of an air conditioner condenser according to claim 1 is characterized in that: Before the image to be processed is input into the established intelligent recognition model for characterizing mildew spots on condensers, the following steps are also included: Acquire and form a training data set of condenser mildew characterization images from a condenser mildew characterization database, and divide the training data set into a training set, a test set, and a validation set in proportion; Classify the morphological features of mold spots in the training data set and generate multiple sets of training images; Based on the image morphological features, the image recognition model is trained.

3. The intelligent identification method for mildew characterization on the surface of an air conditioner condenser according to claim 2 is characterized in that: The construction of the intelligent recognition model for condenser mildew characterization comprises at least the following steps: A feature extraction model based on deep convolutional neural network is used to cut the condenser mildew image, and ResNet is used as the backbone network to extract image features. Use FPN structure for multi-scale feature fusion; Image prediction and output are completed by a classifier, wherein the classifier includes at least Faster R-CNN; And use VIA tool to annotate the condenser mildew characterization training images, generate annotated data files, and convert them into CSV format; Finally, the labeled data files are trained through the training framework, an intelligent recognition model for condenser mildew is constructed, and the mildew recognition is realized through the detection algorithm.

4. The intelligent identification method for mildew characterization on the surface of an air conditioner condenser according to claim 3 is characterized in that: The model building algorithm of the condenser mildew characterization intelligent recognition model includes: The convolution layer extracts the mildew features from the input condenser mildew image. Given a condenser mildew input image I and a convolution kernel K, the convolution kernel is equivalent to a specific feature detector used to scan every position of the input mildew image. The convolution operation can be expressed as: O(i,j)=∑ m ∑ n I(i+m,j+n)·K(m,n) I(i,j) represents the pixel value of the input mildew image at position (i,j); K(m,n) represents the value of the convolution kernel at position (m,n); O(i,j) represents the value of the output mildew feature map at position (i,j); Activation function, using the activation function, namely the ReLU function, introduces nonlinearity, reduces the recognition training time and computing cost, thereby improving the training efficiency and performance of the condenser mildew characterization intelligent recognition model. The ReLU function is expressed as: ReLU(x)=max(0,x) If x is greater than 0, the ReLU function outputs x, if x is less than 0, the ReLU function outputs 0; The pooling layer downsamples the output of the convolutional layer and uses maximum pooling to reduce the spatial dimension of the mildew feature map and extract the main feature texture information of the image, which is expressed as: pool represents the size of the pooling area of ​​the mold feature map; For each position in the pooling area, the maximum value is taken, i.e., maximum pooling, to generate the main feature texture information of the output mildew image; The fully connected layer integrates the high-dimensional feature vector into a global feature expression by weighted summing the condenser mildew features and introducing a nonlinear activation function. It uses trainable weights and biases to map the features to the sample label space and outputs the predicted probability distribution of each category, thereby completing the mildew classification task. It effectively enhances the model's discrimination ability and the ability to fit complex features, and ensures the correlation between the features and the classification target, which can be expressed as: y=ReLU(wx+b) y is the output result of the fully connected layer, which represents the output vector after weighted summation and activation function processing, corresponding to the predicted value or probability distribution of each mold image category; w is a weight parameter matrix with a size of n×m, which reflects the relationship between input features and output results, where n is the dimension of the mold feature map input, and m is the dimension of the mold feature map output, which is related to the number of categories of the mold image classification; x is the input vector, which represents the feature vector of mildew spots obtained after being processed by the previous layer; b is a bias vector with a size of 1×m. Introducing a bias can enhance the expressiveness of the model and help adjust the baseline value of the output; Loss function,In the training process of convolutional neural network (CNN), the cross entropy function is used to measure the gap between the prediction results of the condenser mildew characterization intelligent recognition model and the actual mildew classification results. It can effectively quantify the inconsistency between the probability distribution predicted by the model and the true label distribution. The model can gradually optimize the parameters, thereby improving the classification accuracy and generalization ability of the mildew category. The cross entropy loss function is expressed as: u i It is the result of actual mold classification. is the model's predicted probability for the i-th type of mold stain; The optimization algorithm trains the CNN model by minimizing the loss function. The most basic optimization algorithm, the gradient descent method, is used. The gradient of the loss function to the condenser mildew feature recognition model parameters is calculated to guide the parameters to be updated in the direction of reducing the loss. Through repeated iterative optimization, the minimum value of the objective function is gradually approached, thereby optimizing the model parameters so that the mildew recognition can be completed more accurately. The update rule is: W is the weight parameter matrix of the mold spot feature recognition model, which is used to represent the trainable parameters of the model; η is the learning rate, which is used to control the step size of each weight update and affects the optimization convergence speed; It is the gradient of the loss function L with respect to the weight parameter matrix W, which represents the directional derivative of the current model on the weight W and is used to guide the direction of weight update. Each update of W gradually reduces the value of the loss function L and eventually approaches the optimal value, thereby optimizing the performance of the condenser mildew feature recognition model.

5. The intelligent identification method for mildew characterization on the surface of an air conditioner condenser according to claim 1, characterized in that: The S3 at least includes the following steps: Recording with a high-definition camera device to obtain an image of the mildew on the surface of the air-conditioning condenser component to be processed; The image is adjusted to an input resolution of the condenser mildew characterization intelligent recognition model, and is input into the condenser mildew characterization intelligent recognition model for processing; The mold stain characterization image recognition results are obtained; and the recognition results are verified and analyzed through high-precision detection means.

6. The intelligent identification method for mildew characterization on the surface of an air conditioner condenser according to claim 1, characterized in that: It also includes verification, evaluation and optimization, which at least include the following steps: Verify and analyze the identification results through high-precision detection means, and the verification also includes verifying the mildew stain characterization through other high-precision detection means; Compare the results of the validation process with the output of the intelligent recognition model for condenser mildew characterization to evaluate and optimize the reliability of the model and ensure its accuracy and stability; The optimized model will be added to the database of typical features of condenser mildew to improve the accuracy and efficiency of subsequent identification; In order to improve training speed and recognition efficiency, the size of the model and input image in the machine vision learning algorithm should be optimized with reference to the computer configuration.

7. A cleanliness warning method for intelligently identifying mildew on the surface of an air conditioner condenser, applied to the method for intelligently identifying mildew on the surface of an air conditioner condenser according to any one of claims 1 to 6, characterized in that: At least the following steps are included: Build a risk prediction model for condenser mildew characterization pollution; An intelligent cleanliness early warning system is formed by a condenser mildew characterization pollution risk prediction model and a prompt module, wherein the prompt module at least includes an indicator light, an LCD screen, and a mobile phone prompt; The condenser mildew characterization pollution risk prediction model is used to monitor and analyze the component condenser mildew characterization pollution risk in real time. When the mildew pollution condition of the air-conditioning system condenser reaches the threshold, the prompt module will issue a cleanliness warning to remind the user that they should take measures to clean the air-conditioning condenser components.

8. The cleanliness warning method for intelligently identifying mildew characteristics on the surface of an air conditioner condenser according to claim 7 is characterized in that: The construction of the condenser mildew characterization pollution risk prediction model comprises at least the following steps: Input the pre-processed condenser mildew image; The contamination changes of condenser mildew characterization were modeled by combining the deep learning algorithm convolutional neural network (CNN); Evaluate the accuracy of the established model and classify the risk level; Formation of a contamination risk prediction model for condenser mildew characterization.

9. The cleanliness warning method for intelligently identifying mildew characteristics on the surface of an air conditioner condenser according to claim 8, characterized in that: The condenser mildew characterization risk prediction model adopts decision tree and random forest algorithms, which are suitable for processing classification problems of mildew characterization with complex nonlinear relationships, can process a large number of input features, and automatically perform feature selection; The decision tree algorithm achieves classification or regression of condenser mildew features by recursively dividing the mildew image data space; The condenser mildew characterization risk prediction model selects the features that can best distinguish mildew data, and constructs a tree model with a branching structure layer by layer. In each division process, an indicator is used to measure the importance of the feature. The indicator includes at least information gain, information gain ratio or Gini index. The feature with the largest information gain or the feature with the smallest Gini index is preferentially selected for data division. The partitioning process continues until a preset stop condition is met, wherein the stop condition includes at least the depth of the tree, the number of samples of the leaf node, or the information gain threshold, thereby generating an explanatory and efficient decision rule to complete the classification or prediction task of the mold spot characteristics; The node partitioning formula is based on information gain or Gini index; Information Gain: IG(T,A) is used to measure the effectiveness of selecting the mold feature A for classifying the mold image dataset T. The larger the value, the more important the feature A is for classification. Entropy(T) is the entropy of the mold image dataset T, which measures the uncertainty or confusion of T. v is a possible value of feature A; Values(A) represents the set of all possible values ​​of feature A. v It is a subset of the feature A value v in the mold image dataset T; is a subset T of feature values ​​v v The ratio relative to the overall data set T indicates the weight after division; Entropy(T v ) represents the subset T v The entropy of , which indicates the purity or uncertainty of the subset; c represents the total number of classification categories; p i is the probability of the mold image category i, that is, the proportion of samples belonging to the i-th category in the mold image dataset; Gini Index: Gini(T) represents the Gini index of the mold image dataset, which measures the accuracy of the dataset. The smaller the value, the higher the accuracy of the dataset. pi is the probability of the mold image category i; The random forest algorithm predicts the risk of mildew characterization of condensers by constructing multiple decision trees. Each decision tree uses a subset generated by random sampling during training, thereby introducing the randomness of data and features to enhance the robustness and generalization ability of the model. The prediction results of all decision trees are voted (classification task) or averaged (regression task) to obtain a comprehensive prediction result. In addition, the model performance is evaluated through "out-of-bag data" to effectively reduce the variance of the model and avoid overfitting, thereby improving the accuracy and stability of mildew characterization risk prediction; The classification of the final prediction formula is: The final prediction formula for regression is: in is the prediction result of the mold spot characterization risk of the i-th decision tree, and N is the number of decision trees.

Images