CN112668628A - Quality detection and visualization method for air conditioner outdoor unit - Google Patents
Quality detection and visualization method for air conditioner outdoor unit Download PDFInfo
- Publication number
- CN112668628A CN112668628A CN202011545107.6A CN202011545107A CN112668628A CN 112668628 A CN112668628 A CN 112668628A CN 202011545107 A CN202011545107 A CN 202011545107A CN 112668628 A CN112668628 A CN 112668628A
- Authority
- CN
- China
- Prior art keywords
- air conditioner
- detected
- outdoor unit
- feature
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention relates to a quality detection and visualization method for an air conditioner outdoor unit, which comprises the following steps: (1) acquiring initial data; (2) classification and data labeling: classifying pictures of the air conditioner outdoor unit to be detected, labeling data according to determined key characteristics, and dividing the labeled data into a training set and a test set; (3) and (3) treatment: selecting a proper anchor value through a K-means clustering algorithm; (4) improving a yolov3 model, adjusting a yolov3 model, setting the improved model as a Yolo model, and setting proper training parameters; (5) training a Yolo model; (6) inputting a picture of the air conditioner outdoor unit to be detected into a well-trained Yolo model to obtain a detection result; (7) and (4) displaying the detection result obtained in the step (6) in real time on a terminal. The invention can clearly visualize the detected key parts in the captured image and obtain obvious effect.
Description
Technical Field
The invention belongs to the technical field of deep learning and quality detection, and particularly relates to an application method of image processing, target detection and edge calculation.
Background
After the air conditioner outdoor unit is manufactured and before the air conditioner outdoor unit is packaged and stored in a warehouse formally, in order to ensure that the produced air conditioner outdoor unit meets the quality standard, all the accessories of the outdoor unit are added correctly, and the appearance of the air conditioner outdoor unit needs to be detected. In the quality detection of the external unit, the most traditional mode is a manual detection method, and in addition, the modes of template matching algorithm, machine learning and the like in the traditional algorithm are also provided.
The manual detection mode is mainly used for determining the quality of the air conditioner outdoor unit in a mode of manually identifying a detection target as the name suggests. However, with the increase of labor cost, workers cannot keep attention for a long time and have certain misjudgment and cannot keep the smoothness of production line operation, and the like, and the mode is gradually eliminated.
With the popularization of industrial automation technology, in order to apply an automated process, a machine intelligent image detection mode is adopted to complete quality detection, and the detection mode is mainly a template matching mode. The template matching is divided into two types, the first mode is that the whole qualified outdoor unit image is used as a template and matched with the image acquired on site, and the image is judged to be positive after the matching degree reaches a certain proportion; the other matching mode is that the key parts to be detected are determined firstly, templates of qualified samples of the key parts are made, then the recognition main body is cut out, template matching is carried out on the templates of the key parts and cut images, and whether the images contain the key parts or not is detected. The algorithm of the first mode has the lowest complexity, is best to realize, and can be quickly identified on a host with poor performance, but the identification accuracy is poor; the complexity of the algorithm of the second mode is slightly low, and meanwhile, a large amount of image preprocessing needs to be carried out before detection, so that the difficulty in realization is greatly improved, and the recognition rate is obviously improved. The template matching algorithm can be used as a relatively ideal mode when the performance of the host computer is limited, but relatively speaking, the template matching algorithm does not meet the requirement of industrial production in many cases in terms of accuracy.
Machine learning is another common detection method, which converts the appearance detection problem of the air conditioner into a classification problem of images, and classifies the images by using a series of classification methods in machine learning, so as to obtain a detection result, and the main flow methods include KNN, SVM, BP neural network, CNN, transfer learning and the like. KNN is called K-nearest Neighbor (K-nearest Neighbor) classification algorithm, and the basic idea is that if most of K nearest Neighbor samples of a sample in a feature space belong to a certain class, the sample also belongs to the class; the SVM is called a support vector machine (SupportVectormachine) in its entirety, and is a generalized linear classifier for binary classification of data, and uses a nonlinear function to map sample data from a low-dimensional space to a high-dimensional space, and the main task is to solve a maximum margin hyperplane for classifying the data samples; the BP neural NetWork (Back-propagation neural NetWork) is the most traditional neural NetWork. The learning process of the BP neural network consists of forward propagation and backward propagation. The samples are transmitted from the input layer, and transmitted to the output layer after passing through the hidden layer, because an error exists between the actual output and the expected output, the back propagation is needed, the output is back propagated to the input layer through the hidden layer in a certain mode, and the error is distributed to each hidden layer, so that error signals of each layer are obtained, and the weight is corrected. The BP neural network has a certain limitation in the classification field, and requires a long training time and is easy to fall into a local minimum. Taking the idea of the BP Neural network as a reference, the Convolutional Neural Network (CNN) is beginning to be widely used and becomes one of the representative algorithms for deep learning. The CNN input layer can process multidimensional data, a gradient descent algorithm is used for learning, meanwhile, the hidden layer comprises a convolution layer, a pooling layer, a full connection layer and other structures, various neural network algorithms are generated according to different structures, the series of algorithms have great requirements on calculated amount, the performances of different algorithms on different tasks have great differences, and specific design is needed on specific problems.
However, in practical applications, the existing chinese patent publications such as CN108644988A are mainly used for detecting the internal quality of the air conditioner outdoor unit, i.e. the problem of cooling/heating effect; as shown in chinese patent publications CN107143976A and CN106765921A, the present invention mainly collects and overhauls internal data of air conditioners which have been installed and are in daily use; also, as in chinese patent publications CN106840549A and CN106813422A, the method is also used for detecting the internal quality of air conditioners. And the intelligent detection aspect is still blank for the appearance of the air conditioner (whether the shell has flaws or whether the label indicates that the pasting is correct) and whether the on-line pipes of the air conditioner internal and external units are correctly added (the sizes of the on-line pipes of the external units of different models are different) after the production is finished but before the air conditioner leaves the factory.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a quality detection and visualization method for an air conditioner outdoor unit.
The invention solves the problems of low classification accuracy, long time, low visualization degree and the like in the prior art and by using a general classification algorithm.
According to the method, a yolo algorithm for video target detection is improved and then applied to image detection and classification, on the premise that the time requirement is met and the classification accuracy is improved, key features (detection targets) needing to be identified in an external machine are marked through a target detection method, and the detection results are output and visualized.
Interpretation of terms:
1. the value of anchor can be understood as a multi-scale sliding window, and can be understood as the size of a priori candidate frame when the target is detected in a popular sense.
2. The Yolov3 model is composed of 75 convolutional layers (Conv), the convolutional layers mainly function to extract the features of the input image, one layer of BN layer (Batch Normalization) is arranged behind each convolutional layer and used for standardizing the matrix behind the convolutional layers, the purpose is to accelerate the training and convergence speed of the network, control the gradient and prevent overfitting, and the connected active layer (LeakyReLU is a specific active function) behind the BN layer is used for increasing the nonlinearity of the structure and enhancing the expression capability of the neural network. In the overall structure, jump connection (residual error network) can be carried out on every two convolutional layers, so that the problem of neural network degradation in a deep neural network is greatly relieved, further training is achieved, and meanwhile, an FPN (feature map pyramid) structure (feature map pyramid network) is added into the structure so as to realize target prediction on a plurality of feature levels. As shown in fig. 8, the DBL structure is the above-mentioned Conv layer, and the BN layer is combined with the leakage relu layer. The Res module in fig. 8 represents a residual network, a single residual network is subjected to jump connection every two convolutional layers to relieve the problem of neural network degradation in a deep neural network, so that deeper training is achieved, meanwhile, an FPN structure (a feature map pyramid network) is added into the structure, the original features are subjected to deep extraction and then to upsampling, and concat fusion is performed with features with the same dimensionality as a shallow layer, so that target prediction is achieved on multiple feature levels.
3. The main role of the K-means clustering algorithm is to automatically classify similar samples into one category. The algorithm step is given K value, confirm the central point of K initial classes (the central point is chosen from the existing sample at random or according to some particular mode generally); dividing other samples into the K categories according to the nearest neighbor principle; re-finding the center point of each category (finding the expectation) and updating the center point; and iterating the steps until the position of the central point is basically not changed or the specified iteration times are reached. Fig. 9 is a result diagram obtained by performing a K-means clustering algorithm on a specific air conditioner external unit model label box, where K is 9, and a specific output result is (113.2,125.9,137.6,150.4,168.5,163.6,206.7,189.3,365.8,127.1,395.1,128.3,411.2,154.7,304.8,362.4,349.0,391.1), where each two bits represent a rectangular box size, and the original drawing size is 2464 × 2056.
The technical scheme of the invention is as follows:
a quality detection and visualization method for an air conditioner outdoor unit comprises the following steps:
(1) initial data acquisition: determining key features needing to be detected, namely acquiring pictures of the air conditioner external unit needing to be detected and determining different key features needing to be detected by different types of air conditioner external units; taking fig. 3b as an example, the outdoor unit in the drawing is pasted with four different brand marks, which need to be detected respectively, and meanwhile, whether the cyclone net of the outdoor unit is installed correctly is also detected; the appearance difference between different models is large, and the key features to be detected also have large difference.
(2) Classification and data labeling: classifying pictures of the air conditioner outdoor unit to be detected, labeling data according to determined key characteristics, and dividing the labeled data into a training set and a test set;
(3) and (3) treatment: analyzing image samples of air conditioner external units of different models, and selecting a proper anchor value through a K-means clustering algorithm;
(4) modified yolov3 model: adjusting and improving a yolov3 model, wherein the model obtained by adjusting and improving a yolov3 model is a Yolo model, and setting proper training parameters;
(5) training a Yolo model;
(6) inputting a picture of the air conditioner outdoor unit to be detected into a well-trained Yolo model to obtain a detection result;
(7) and (4) displaying the detection result obtained in the step (6) in real time on a terminal.
In the image displayed in real time, firstly, key features in the image are marked and names are displayed, as shown in fig. 6 and 7, whether the key features correspond to the key information of a specific model is detected, if the key features of the air conditioner external unit of the model are complete and have no other features which should not exist, the result is judged to be positive, the production line is automatically controlled to continue to work in the next step, if the key features are detected to be lacked or the features have errors, the result is judged to be negative, the production line is controlled to reject the air conditioner external unit, and the air conditioner external unit is returned to the factory for re-assembly.
According to the optimization method, in the step (4), the yolov3 model is improved on the basis of ensuring the accuracy, mainly considering that the yolov3 model has a deep network structure, high learning complexity and large calculation amount required by learning and detection, and a general industrial scene possibly has the problems of strong delay requirement and insufficient calculation force; the Yolo model comprises 24 convolutional layers, a BN layer is connected behind each convolutional layer, an activation layer is connected behind the BN layer, a residual error network is connected in a jumping mode between every two convolutional layers, an FPN structure is added into the Yolo model, only once up-sampling is conducted, feature fusion is conducted on an up-sampling result and shallow layer information of the same dimensionality, prediction information under the dimensionality of 13 x 13 and the dimensionality of 26 x 26 is finally obtained, and the prediction result is obtained by combining the two prediction information.
The structure of the Yolo model is as shown in fig. 2, firstly, performing dimensionality reduction feature extraction on an input image, wherein numbers in each module represent the dimensionality of the feature of the layer, simplifying the convolution layer from 75 layers to 24 layers, simultaneously still adopting a residual error network to prevent network degradation, adding an FPN structure into the model, but only performing one-time upsampling, performing feature fusion on a result and shallow layer information of the same dimensionality, and finally obtaining prediction information under two dimensions of 13 × 13 and 26 × 26, and combining the two to obtain a final prediction result.
Preferably, in step (5), the Yolo model is trained, which includes the following steps:
a. counting the similarity between the pictures of the air conditioner external units of different models processed in the step (3), namely dividing the pictures of the air conditioner external units of different models with the same key characteristics larger than n into one type; n is more than 4;
b. carrying out unified naming on key features to be detected of picture samples of the air conditioner external units of different types, and uniformly adjusting the brightness and the size of the picture; for example, the size is uniformly adjusted to 416 × 416.
c. And c, randomly dividing the picture samples which are divided into one type of air conditioner external units with different models and processed in the step b into a training set and a testing set, and training the Yolo model.
According to the present invention, preferably, in the step (2), classifying the pictures of the air conditioner external unit to be detected includes:
firstly, classifying according to different types of air conditioner outdoor units;
secondly, whether it is complete to detect key feature according to the outer machine of different models air conditioners, for example, whether a certain model whirlwind net is installed correctly, whether the online pipe of different diameters is complete, whether different brand marks are posted correctly, divide the positive sample and the negative sample of the outer machine of different models air conditioners, positive sample is that the outer machine of air conditioners detects key feature complete, the negative sample is that lack some or some outer machine of air conditioners detects key feature promptly. And selecting pictures with different light factors and different shooting angles from the classified pictures as a data set.
According to the optimization of the invention, the data are labeled according to the determined key characteristics, and the labeled data are divided into a training set and a test set, which comprises the following steps:
for the external machines of different models, the detection key features have great difference, so different key features are determined for the external machines of different models in the training set, and the different key features are subjected to unified naming;
A. performing category labeling on key features in a data set, namely after determining the key features to be labeled, framing each key feature in an image by using a rectangular frame respectively, and attaching uniformly named feature names so as to extract position information of the key features, wherein the position information of the key features comprises feature names, starting coordinates of the feature in the u direction of a horizontal coordinate, starting coordinates of the feature in the v direction of a vertical coordinate, ending coordinates of the feature in the u direction and ending coordinates of the feature in the v direction; the coordinate system is a pixel coordinate system, the origin of the coordinate system is a pixel point at the upper left corner, the unit of the coordinate is a pixel, the u direction is the horizontal direction, the maximum value is the number of columns in the image array, the v direction is the vertical direction, and the maximum value is the number of rows in the image array; for example, (logo, 5,4, 20,40) is a key feature name, and the starting point coordinate of the position is (5,4), that is, the starting point is a pixel point in the fifth column and the fourth row, and the ending point coordinate is (20,40), and the ending position is a rectangular frame in which the area where the pixel point in the 20 th column and the 40 th row is located is a pixel point in the u direction 15 and the v direction is a pixel point 36.
B. Labeling each image to generate a corresponding labeled file, wherein the labeled file comprises an original image file name, a storage position, a pixel size, a channel number, labeled feature names and corresponding position information, reading the labeled file, uniformly sequencing the feature names, sequentially converting the names into numbers, such as 0,1,2,3 and 4, representing five features to be detected, and respectively assigning different digital codes to the different features; if the number of the features to be detected is six, the different features are sequenced and then converted into six numerical codes of 0,1,2,3,4 and 5; reading the size of an image pixel in the label file, and then normalizing the coordinates and lengths contained in different characteristics to convert the coordinates and lengths into relative coordinates and relative lengths; for example, a certain feature is (logo,300,400,350,450) that the name of the feature is logo, the start position is column 300, row 400, the end position is column 350, row 450, the size of the rectangular frame is 50 pixels in the horizontal direction, 50 pixels in the vertical direction, after the pixel size (for example, 2000 × 1000) of the markup file is obtained, assuming that the logo is the first feature after sorting, the final conversion result is (0,0.15,0.4,0.025,0.05) that the feature is 0 th, the image length and width are respectively 1, the start position is (0.15,0.4), and the sizes of the rectangular frame in the horizontal and vertical directions are respectively 0.025 and 0.05. Then, adjusting the pixels of the picture;
C. the marked pictures are divided into a training set and a test set, and the data volume is large enough, so that the marked pictures are divided into 90% of training sets and 10% of test sets.
Further preferably, the pixels of the picture are adjusted to 416 × 416.
The beneficial effects of the invention are:
in practical project application, the invention achieves very outstanding effect. For an air conditioner outdoor unit without an online pipe, the accuracy rate of 99.98% is obtained, and the traditional algorithm can only be about 70%; for the air conditioner outdoor unit added with the on-line pipe, the traditional algorithm performance is extremely poor, the accuracy rate is extremely low, the accuracy rates of different models are maintained to be more than 85% after the technical scheme of the invention is used, the pure algorithm time for detecting one machine is reduced to 0.2 second on average, and the detected key part is clearly visualized in a captured image
Drawings
FIG. 1 is a schematic flow chart of a method for detecting and visualizing the quality of an air conditioner outdoor unit;
FIG. 2 is a schematic structural diagram of the Yolo model;
FIG. 3a is a schematic model view (with the broad category of inline pipes);
FIG. 3b is a schematic model view (without the linking pipe category);
FIG. 4 is a schematic loss of training process;
FIG. 5a is a schematic diagram of a P-R curve of a key feature Haier text token after 20000 iterations in a training process;
FIG. 5b is a schematic diagram of the P-R curve of the key feature Haier image flag after 20000 iterations in the training process;
FIG. 5c is a schematic diagram of the P-R curve of the key feature white online tube head after 20000 iterations in the training process;
FIG. 5d is a schematic diagram of the P-R curve of the black online tip of the key feature after 20000 iterations in the training process;
FIG. 6 is a schematic diagram of the output results (with the online pipe category);
FIG. 7 is a schematic diagram of the output results (no-line pipe class);
FIG. 8 is a schematic structural diagram of the Yolov3 model;
fig. 9 is a schematic diagram of a result obtained by applying a K-means clustering algorithm to a specific air conditioner external unit model number marking box.
Detailed Description
The invention is further defined in the following, but not limited to, the figures and examples in the description.
Example 1
A method for detecting and visualizing the quality of an air conditioner outdoor unit, as shown in FIG. 1, comprises the following steps:
(1) initial data acquisition: determining key features needing to be detected, namely acquiring pictures of the air conditioner external unit needing to be detected and determining different key features needing to be detected by different types of air conditioner external units; taking fig. 3b as an example, the outdoor unit in the drawing is pasted with four different brand marks, which need to be detected respectively, and meanwhile, whether the cyclone net of the outdoor unit is installed correctly is also detected; the appearance difference between different models is large, and the key features to be detected also have large difference.
(2) Classification and data labeling: classifying pictures of the air conditioner outdoor unit to be detected, labeling data according to determined key characteristics, and dividing the labeled data into a training set and a test set;
in the step (2), classifying the pictures of the air conditioner external unit to be detected, which means that:
firstly, classifying according to different types of air conditioner outdoor units; in this embodiment, the outdoor unit of the air conditioner is divided into 32 different models according to the difference between the appearance and the detected detection target.
Secondly, whether it is complete to detect key feature according to the outer machine of different models air conditioners, for example, whether a certain model whirlwind net is installed correctly, whether the online pipe of different diameters is complete, whether different brand marks are posted correctly, divide the positive sample and the negative sample of the outer machine of different models air conditioners, positive sample is that the outer machine of air conditioners detects key feature complete, the negative sample is that lack some or some outer machine of air conditioners detects key feature promptly. And selecting pictures with different light factors and different shooting angles from the classified pictures as a data set.
Taking fig. 3a as an example, two logos and two on-line pipe caps on the right side of the air conditioner outdoor unit must be included, one of the two on-line pipe caps must be a white pipe cap, and the other one must be a blue pipe cap; taking fig. 3b as an example, it is required that all four signs on the right side of the air conditioner external unit are correctly pasted and the cyclone net is correctly installed, and the missing point is a negative sample. Meanwhile, after the criteria are determined, the detection targets need to be named uniformly, the character mark in fig. 3a is named as logo _ red1, the image mark is named as logo _ red3, the white cap is named as white, and the blue cap is named as blue. If the same detection target as that in fig. 3a is assumed in the subsequent models, the names are the same, and if a new detection target appears, the names are uniformly named and then the names are used, and the reference sample is shown in fig. 6 and table 1, for example.
TABLE 1
(3) And (3) treatment: analyzing image samples of air conditioner external units of different models, and selecting a proper anchor value through a K-means clustering algorithm;
(4) modified yolov3 model: adjusting and improving a yolov3 model, wherein the model obtained by adjusting and improving a yolov3 model is a Yolo model, and setting proper training parameters;
(5) training a Yolo model;
(6) inputting a picture of the air conditioner outdoor unit to be detected into a well-trained Yolo model to obtain a detection result;
(7) and (4) displaying the detection result obtained in the step (6) in real time on a terminal.
In the image displayed in real time, firstly, key features in the image are marked and names are displayed, as shown in fig. 6 and 7, whether the key features correspond to the key information of a specific model is detected, if the key features of the air conditioner external unit of the model are complete and have no other features which should not exist, the result is judged to be positive, the production line is automatically controlled to continue to work in the next step, if the key features are detected to be lacked or the features have errors, the result is judged to be negative, the production line is controlled to reject the air conditioner external unit, and the air conditioner external unit is returned to the factory for re-assembly.
Example 2
The method for detecting and visualizing the quality of the air conditioner outdoor unit in the embodiment 1 is characterized in that:
in the step (4), the yolov3 model is improved on the basis of ensuring the accuracy by mainly considering that the yolov3 model has a deep network structure, high learning complexity and large calculation amount required by learning and detection, and a general industrial scene may have the problems of strong time delay requirement and insufficient calculation force; the Yolo model comprises 24 convolutional layers, each layer of convolutional layer is connected with a BN layer, the BN layer is connected with an activation layer, a residual error network is connected in a jumping mode every two convolutional layers, the Yolo model is added with an FPN structure, only once up-sampling is conducted, feature fusion is conducted on an up-sampling result and shallow layer information of the same dimensionality, prediction information under the two dimensionalities of 13 x 13 and 26 x 26 is finally obtained, and the prediction result is obtained by combining the two dimensionalities.
The structure of the Yolo model is as shown in fig. 2, firstly, performing dimensionality reduction feature extraction on an input image, wherein numbers in each module represent the dimensionality of the feature of the layer, simplifying the convolution layer from 75 layers to 24 layers, simultaneously still adopting a residual error network to prevent network degradation, adding an FPN structure into the model, but only performing one-time upsampling, performing feature fusion on a result and shallow layer information of the same dimensionality, and finally obtaining prediction information under two dimensions of 13 × 13 and 26 × 26, and combining the two to obtain a final prediction result.
Feature extraction is carried out on the image through a convolutional neural network, dimension reduction is carried out through a pooling layer, feature extraction is continuously carried out after dimension reduction reaches 13 × 13, meanwhile, a residual error network and an FPN (feature pyramid) framework are applied, feature fusion is finally carried out, and prediction is carried out through two independent prediction branches. Fig. 8 shows the complete yolov3 network architecture used, in fig. 8, conv represents the convolutional layer, BN represents batch standardization, leakareu represents the active layer, and the three constitute the DBL layer; res is a residual structure, the input of each residual structure is added with the original input (matrix addition) after passing through two DBL layers to form a new characteristic matrix, the residual structure greatly keeps the characteristic information of the image, the characteristic disappearance phenomenon is greatly relieved, and the network can be trained more deeply; meanwhile, an FPN architecture is added in fig. 8, and is composed of a bottom-up line, a top-down line and a transverse connection, in fig. 8, the part before the output of 13 × 27 is the bottom-up, and then the part enters the top-down line from the upper sampling of DBL +, wherein concat is the transverse connection; the framework enables prediction to be carried out in feature maps of three dimensions at the same time, and the prediction capability of small targets is greatly improved.
Example 3
The method for detecting and visualizing the quality of the air conditioner outdoor unit in the embodiment 1 is characterized in that:
in the step (5), the Yolo model is trained, which comprises the following steps:
a. counting the similarity between the pictures of the air conditioner external units of different models processed in the step (3), namely dividing the pictures of the air conditioner external units of different models with the same key characteristics larger than n into one type; n is more than 4;
b. carrying out unified naming on key features to be detected of picture samples of the air conditioner external units of different types, and uniformly adjusting the brightness and the size of the picture; for example, the size is uniformly adjusted to 416 × 416.
c. And c, randomly dividing the picture samples which are divided into one type of air conditioner external units with different models and processed in the step b into a training set and a testing set, and training the Yolo model.
When the loss falls to a certain thresholdAnd stopping the training of the model after the value is basically not reduced any more. Fig. 5a, 5b, 5c, and 5d are P-R curves of a certain type detection target after 20000 iterations. In FIGS. 5a to 5d, the abscissa of the P-R curve is recall (recall), and the ordinate is precision (precision), recall
Precision ratio
TP represents the number of correctly classified positive samples, FN represents the number of incorrectly classified positive samples as negative samples, and FP represents the number of incorrectly classified negative samples as positive samples.
Example 4
The method for detecting and visualizing the quality of the air conditioner outdoor unit in the embodiment 1 is characterized in that:
labeling data according to the determined key features, and dividing the labeled data into a training set and a test set, wherein the method comprises the following steps:
for the external machines of different models, the detection key features have great difference, so different key features are determined for the external machines of different models in the training set, and the different key features are subjected to unified naming;
A. performing category labeling on key features in a data set, namely after determining the key features to be labeled, framing each key feature in an image by using a rectangular frame respectively, and attaching uniformly named feature names so as to extract position information of the key features, wherein the position information of the key features comprises feature names, starting coordinates of the feature in the u direction of a horizontal coordinate, starting coordinates of the feature in the v direction of a vertical coordinate, ending coordinates of the feature in the u direction and ending coordinates of the feature in the v direction; the coordinate system is a pixel coordinate system, the origin of the coordinate system is a pixel point at the upper left corner, the unit of the coordinate is a pixel, the u direction is the horizontal direction, the maximum value is the number of columns in the image array, the v direction is the vertical direction, and the maximum value is the number of rows in the image array; for example, (logo, 5,4, 20,40) is a key feature name, and the starting point coordinate of the position is (5,4), that is, the starting point is a pixel point in the fifth column and the fourth row, and the ending point coordinate is (20,40), and the ending position is a rectangular frame in which the area where the pixel point in the 20 th column and the 40 th row is located is a pixel point in the u direction 15 and the v direction is a pixel point 36.
B. Labeling each image to generate a corresponding labeled file, wherein the labeled file comprises an original image file name, a storage position, a pixel size, a channel number, labeled feature names and corresponding position information, reading the labeled file, uniformly sequencing the feature names, sequentially converting the names into numbers, such as 0,1,2,3 and 4, representing five features to be detected, and respectively assigning different digital codes to the different features; if the number of the features to be detected is six, the different features are sequenced and then converted into six numerical codes of 0,1,2,3,4 and 5; reading the size of an image pixel in the label file, and then normalizing the coordinates and lengths contained in different characteristics to convert the coordinates and lengths into relative coordinates and relative lengths; for example, a certain feature is (logo,300,400,350,450) that the name of the feature is logo, the start position is column 300, row 400, the end position is column 350, row 450, the size of the rectangular frame is 50 pixels in the horizontal direction, 50 pixels in the vertical direction, after the pixel size (for example, 2000 × 1000) of the markup file is obtained, assuming that the logo is the first feature after sorting, the final conversion result is (0,0.15,0.4,0.025,0.05) that the feature is 0 th, the image length and width are respectively 1, the start position is (0.15,0.4), and the sizes of the rectangular frame in the horizontal and vertical directions are respectively 0.025 and 0.05. Then, adjusting the pixels of the picture; fig. 4 is an example of the process of labeling an image. The abscissa iters represents the number of iterations, and the ordinate loss represents the loss at the current number of iterations (the smaller the loss, the better); generating a corresponding table 1 label file for each image according to the above, then reading five key information in the label file, converting the target name into numbers of 0,1,2, etc., normalizing the coordinates and the length into relative coordinates and relative length, then performing pixel adjustment on the images, uniformly adjusting the images into a fixed size, and then converting an xml file (fig. 4) generated after labeling into a txt format (table 1) capable of being trained;
C. the marked pictures are divided into a training set and a test set, and the data volume is large enough, so that the marked pictures are divided into 90% of training sets and 10% of test sets. The pixels for the picture are adjusted to 416 x 416.
Claims (6)
1. A quality detection and visualization method for an air conditioner outdoor unit is characterized by comprising the following steps:
(1) initial data acquisition: determining key features needing to be detected, namely acquiring pictures of the air conditioner external unit needing to be detected and determining different key features needing to be detected by different types of air conditioner external units;
(2) classification and data labeling: classifying pictures of the air conditioner outdoor unit to be detected, labeling data according to determined key characteristics, and dividing the labeled data into a training set and a test set;
(3) and (3) treatment: analyzing image samples of air conditioner external units of different models, and selecting a proper anchor value through a K-means clustering algorithm;
(4) modified yolov3 model: adjusting and improving a yolov3 model, wherein the model obtained by adjusting and improving a yolov3 model is a Yolo model, and setting proper training parameters;
(5) training a Yolo model;
(6) inputting a picture of the air conditioner outdoor unit to be detected into a well-trained Yolo model to obtain a detection result;
(7) and (4) displaying the detection result obtained in the step (6) in real time on a terminal.
2. The method for detecting and visualizing the quality of the outdoor unit of the air conditioner as claimed in claim 1, wherein in the step (4), the Yolo model includes 24 convolutional layers, a BN layer is connected behind each convolutional layer, an activation layer is connected behind the BN layer, a residual network is connected in a jumping manner between every two convolutional layers, the Yolo model adds an FPN structure, upsampling is performed only once, and feature fusion is performed on upsampled results and shallow layer information of the same dimension, so that prediction information of two dimensions 13 × 13 and 26 × 26 is finally obtained, and the prediction results are combined to obtain the final prediction result.
3. The method for detecting and visualizing the quality of the outdoor unit of the air conditioner as claimed in claim 1, wherein the step (5) of training a Yolo model comprises the steps of:
a. counting the similarity between the pictures of the air conditioner external units of different models processed in the step (3), namely dividing the pictures of the air conditioner external units of different models with the same key characteristics larger than n into one type; n is more than 4;
b. carrying out unified naming on key features to be detected of picture samples of the air conditioner external units of different types, and uniformly adjusting the brightness and the size of the picture;
c. and c, randomly dividing the picture samples which are divided into one type of air conditioner external units with different models and processed in the step b into a training set and a testing set, and training the Yolo model.
4. The method for detecting and visualizing the quality of the outdoor unit of the air conditioner as claimed in claim 1, wherein in the step (2), the step of classifying the pictures of the outdoor unit of the air conditioner to be detected includes:
firstly, classifying according to different types of air conditioner outdoor units;
secondly, according to whether the key features to be detected of the air conditioner external units of different models are complete or not, positive samples and negative samples of the air conditioner external units of different models are separated, the positive samples are the key features to be detected of the air conditioner external units to be complete, and the negative samples are the key features to be detected of the air conditioner external units to be short of one or more items.
5. The method for detecting and visualizing the quality of the outdoor unit of the air conditioner as claimed in any one of claims 1 to 4, wherein the data are labeled according to the determined key features, and the labeled data are divided into a training set and a test set, comprising the steps of:
A. performing category labeling on key features in a data set, namely after determining the key features to be labeled, framing each key feature in an image by using a rectangular frame respectively, and attaching uniformly named feature names so as to extract position information of the key features, wherein the position information of the key features comprises feature names, starting coordinates of the feature in the u direction of a horizontal coordinate, starting coordinates of the feature in the v direction of a vertical coordinate, ending coordinates of the feature in the u direction and ending coordinates of the feature in the v direction; the coordinate system is a pixel coordinate system, the origin of the coordinate system is a pixel point at the upper left corner, the unit of the coordinate is a pixel, the u direction is the horizontal direction, the maximum value is the number of columns in the image array, the v direction is the vertical direction, and the maximum value is the number of rows in the image array;
B. labeling each image to generate a corresponding labeled file, wherein the labeled file comprises an original image file name, a storage position, a pixel size, a channel number, labeled feature names and corresponding position information, reading the labeled file, uniformly sequencing the feature names, sequentially converting the names into numbers, reading the image pixel size in the labeled file, and then normalizing coordinates and lengths contained in different features to convert the coordinates and the lengths into relative coordinates and relative lengths; then, adjusting the pixels of the picture;
C. and dividing the marked pictures into a training set and a test set, wherein 90% of the training set and 10% of the test set are used as the training set.
6. The method as claimed in claim 5, wherein the pixels of the picture are adjusted to 416 x 416.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011545107.6A CN112668628A (en) | 2020-12-24 | 2020-12-24 | Quality detection and visualization method for air conditioner outdoor unit |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011545107.6A CN112668628A (en) | 2020-12-24 | 2020-12-24 | Quality detection and visualization method for air conditioner outdoor unit |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112668628A true CN112668628A (en) | 2021-04-16 |
Family
ID=75408175
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011545107.6A Pending CN112668628A (en) | 2020-12-24 | 2020-12-24 | Quality detection and visualization method for air conditioner outdoor unit |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112668628A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113409314A (en) * | 2021-08-18 | 2021-09-17 | 南京市特种设备安全监督检验研究院 | Unmanned aerial vehicle visual detection and evaluation method and system for corrosion of high-altitude steel structure |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109064461A (en) * | 2018-08-06 | 2018-12-21 | 长沙理工大学 | A kind of detection method of surface flaw of steel rail based on deep learning network |
| CN110660052A (en) * | 2019-09-23 | 2020-01-07 | 武汉科技大学 | Hot-rolled strip steel surface defect detection method based on deep learning |
| CN111222474A (en) * | 2020-01-09 | 2020-06-02 | 电子科技大学 | Method for detecting small target of high-resolution image with any scale |
| CN111612784A (en) * | 2020-06-01 | 2020-09-01 | 南通大学 | Steel plate surface defect detection method based on classification-first YOLO network |
| CN111881730A (en) * | 2020-06-16 | 2020-11-03 | 北京华电天仁电力控制技术有限公司 | Wearing detection method for on-site safety helmet of thermal power plant |
-
2020
- 2020-12-24 CN CN202011545107.6A patent/CN112668628A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109064461A (en) * | 2018-08-06 | 2018-12-21 | 长沙理工大学 | A kind of detection method of surface flaw of steel rail based on deep learning network |
| CN110660052A (en) * | 2019-09-23 | 2020-01-07 | 武汉科技大学 | Hot-rolled strip steel surface defect detection method based on deep learning |
| CN111222474A (en) * | 2020-01-09 | 2020-06-02 | 电子科技大学 | Method for detecting small target of high-resolution image with any scale |
| CN111612784A (en) * | 2020-06-01 | 2020-09-01 | 南通大学 | Steel plate surface defect detection method based on classification-first YOLO network |
| CN111881730A (en) * | 2020-06-16 | 2020-11-03 | 北京华电天仁电力控制技术有限公司 | Wearing detection method for on-site safety helmet of thermal power plant |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113409314A (en) * | 2021-08-18 | 2021-09-17 | 南京市特种设备安全监督检验研究院 | Unmanned aerial vehicle visual detection and evaluation method and system for corrosion of high-altitude steel structure |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2022147965A1 (en) | Arithmetic question marking system based on mixnet-yolov3 and convolutional recurrent neural network (crnn) | |
| Busta et al. | Deep textspotter: An end-to-end trainable scene text localization and recognition framework | |
| CN110321967B (en) | Image classification improvement method based on convolutional neural network | |
| CN111325203A (en) | American license plate recognition method and system based on image correction | |
| US20120195495A1 (en) | Hierarchical Tree AAM | |
| CN102385592B (en) | Image concept detection method and device | |
| CN108664975B (en) | Uyghur handwritten letter recognition method and system and electronic equipment | |
| WO2021022571A1 (en) | Interactive modeling-based multi-label distance metric learning method | |
| CN109902609A (en) | A kind of road traffic sign detection and recognition methods based on YOLOv3 | |
| CN110751232A (en) | Chinese complex scene text detection and identification method | |
| CN112818951A (en) | Ticket identification method | |
| CN113129284B (en) | Appearance detection method based on 5G cloud edge cooperation and implementation system | |
| CN111368703A (en) | Platform logo detection and identification method based on FPN | |
| CN112541491A (en) | End-to-end text detection and identification method based on image character region perception | |
| CN116385374A (en) | Cell counting method based on convolutional neural network | |
| CN114882204A (en) | Automatic ship name recognition method | |
| CN104809468A (en) | Multi-view classification method based on indefinite kernels | |
| CN112668628A (en) | Quality detection and visualization method for air conditioner outdoor unit | |
| CN111832497B (en) | Text detection post-processing method based on geometric features | |
| CN111144469B (en) | End-to-end multi-sequence text recognition method based on multi-dimensional associated time sequence classification neural network | |
| CN110827351B (en) | Automatic generation method of voice tag of new target for robot audio-visual collaborative learning | |
| Calefati et al. | Reading meter numbers in the wild | |
| CN113705713B (en) | Text recognition method based on global and local attention mechanisms | |
| CN115984968A (en) | Student time-space action recognition method and device, terminal equipment and medium | |
| CN111985397B (en) | Pupil oral calculation subject text picture content identification method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210416 |