Copper pipe processing surface defect detection system

Abstract

The invention relates to the technical field of copper pipe surface defect identification, in particular to a copper pipe processing surface defect detection system which comprises the following steps: performing image acquisition on the surface of a to-be-processed copper pipe to obtain a first image set; constructing a first surface recognition model to recognize the first image set to obtain a first classification image and first abnormal data; the copper pipe to be machined is subjected to paint spraying machining, and a machined copper pipe is obtained; performing image acquisition on the surface of the processed copper pipe to obtain a second image set; dividing the second image set according to the division mode of the first classification image in the first image set to obtain a second classification image; constructing a second surface recognition model to recognize the second classification image to obtain second abnormal data; and constructing a third surface recognition model to train the second abnormal data of the second classification image and the first abnormal data corresponding to the first classification image to obtain third abnormal data. According to the invention, the defect of paint spraying on the surface of the copper pipe is accurately identified through the third abnormal data.

Description

Technical Field

[0001] The invention relates to the technical field of copper tube surface defect recognition, in particular to a copper tube processing surface defect detection system. Background Art

[0002] Copper tube is a kind of non-ferrous metal tube, which is produced by extrusion, drawing, rolling and cold heading. Copper tube has excellent properties, such as high temperature resistance, fatigue resistance, safety and hygiene, and is widely used in construction facilities, electronic and electrical equipment, automotive industry, medical equipment and other fields.

[0003] Since the surface of copper tubes is relatively soft and easily scratched and damaged, if they are in a corrosive environment for a long time, the surface is prone to corrosion, causing damage to the copper tubes. The above problems can be effectively addressed by spraying paint on the surface of copper tubes.

[0004] Spraying paint on the copper tube surface can enhance the wear resistance of the copper tube surface, prevent the corrosive substances in the environment from damaging the copper tube, and extend the service life. In the process of copper tube spraying, the early cleaning and polishing of the copper tube play an important role, which can remove impurities and oxides on the surface of the copper tube and ensure the flatness and uniformity of the copper tube surface.

[0005] If the surface treatment of the copper tube is not perfect, it will lead to uneven painting, bubbles, sagging, paint falling off and rough paint surface, which will affect the protective effect of the paint; therefore, it is necessary to identify defects in the copper tube painting process.

[0006] In the prior art, defects on the processed surface of copper tubes after painting are mainly identified through image recognition methods, which can achieve recognition effect to a certain extent; however, the painting effect of the copper tube surface is also affected by the surface defects of the copper tube before painting. The prior art does not link the defects before and after painting, resulting in inaccurate identification of the paint defect area.

[0007] Therefore, a copper tube processing surface defect detection system is proposed. Summary of the invention

[0008] The object of the present invention is to provide a copper tube processing surface defect detection system, which comprises the following steps: performing image acquisition on the surface of a copper tube to be processed to obtain a first image set; constructing a first surface recognition model to recognize the first image set to obtain a first classified image and first abnormal data; performing paint processing on the copper tube to be processed to obtain a processed copper tube; performing image acquisition on the surface of the processed copper tube to obtain a second image set; dividing the second image set according to the division method of the first classified image in the first image set to obtain a second classified image; constructing a second surface recognition model to recognize the second classified image to obtain second abnormal data; constructing a third surface recognition model to train the second abnormal data of the second classified image and the first abnormal data corresponding to the first classified image to obtain third abnormal data; and accurately identifying the defects of the paint spraying on the copper tube surface through the third abnormal data.

[0009] To achieve the above object, the present invention provides a copper tube processing surface defect detection system, comprising:

[0010] A first data acquisition module pre-processes the copper tube to obtain a copper tube to be processed; and collects images of the surface of the copper tube to be processed to obtain a first image set;

[0011] A first defect recognition module is configured to construct a first surface recognition model, recognize the first image set by using the first surface recognition model, divide the first image set into regions according to the contaminant features, flatness features, and roughness features of the first image set, and obtain a plurality of first classified images and corresponding first abnormal data;

[0012] A second data acquisition module sprays paint on the copper tube to be processed to obtain a processed copper tube; and collects images of the surface of the processed copper tube to obtain a second image set;

[0013] A data processing module obtains a second image set, and divides the second image set according to the division method of the first classification image in the first image set to obtain a plurality of second classification images;

[0014] A second defect recognition module constructs a second surface recognition model; recognizes the second classified image through the second surface recognition model, and obtains second abnormal data through texture feature and color feature recognition;

[0015] The defect warning module constructs a third surface recognition model, inputs the second abnormal data of the second classification image and the first abnormal data corresponding to the first classification image into the third surface recognition model to obtain third abnormal data; and identifies the defective area of ​​the processed surface according to the third abnormal data.

[0016] The first surface recognition model includes a first data input layer, a first feature extraction layer, a pollutant feature recognition layer, a flatness feature recognition layer, a roughness feature recognition layer and an image segmentation layer;

[0017] The first data input layer is used to input the first image set into the model and perform image preprocessing;

[0018] The first feature extraction layer extracts features from the preprocessed first image set to obtain a first feature data set;

[0019] The pollutant feature recognition layer is used to recognize the first feature data set to obtain pollutant distribution data;

[0020] The flatness feature recognition layer is used to recognize the first feature data set to obtain flatness distribution data;

[0021] The roughness feature recognition layer is used to recognize the first feature data set to obtain roughness distribution data;

[0022] The image division layer divides the first image set according to the pollutant distribution data, flatness distribution data and roughness distribution data to obtain a first classified image; and uses the pollutant distribution data, flatness distribution data and roughness distribution data as first abnormal data.

[0023] The training process of the pollutant feature recognition layer is as follows:

[0024] Acquire a first data set, which is a surface image of a copper tube to be processed with a pollutant area marked thereon; split the first data set to obtain a first training set and a first test set; and use the first training set and the first test set to train to obtain a pollutant feature recognition layer;

[0025] The training process of the flatness feature recognition layer is as follows:

[0026] Acquire a second data set, which is a surface image of a copper tube to be processed with an abnormal flatness area marked thereon; split the second data set to obtain a second training set and a second test set; and use the second training set and the second test set to train to obtain a flatness feature recognition layer;

[0027] The training process of the roughness feature recognition layer is:

[0028] A third data set is obtained, wherein the third data set is a surface image of the copper tube to be processed with annotated roughness abnormal areas; the third data set is split to obtain a third training set and a third test set; and the third training set and the third test set are used to train to obtain a roughness feature recognition layer.

[0029] The process of dividing the second image set to obtain the second classified image is as follows:

[0030] Establishing a three-dimensional model of the copper tube according to the copper tube to be processed, and acquiring a first image set and a first classified image;

[0031] Restoring the first image set to the copper tube three-dimensional model to obtain the regional distribution of the first classified image on the copper tube three-dimensional model, and obtaining a first classified three-dimensional model;

[0032] The second image set is restored to the first classified three-dimensional model; and the second image set is divided according to the regional distribution of the first classified images on the first classified three-dimensional model to obtain second classified images.

[0033] The second surface recognition model includes a second data input layer, a second feature extraction layer, a texture anomaly recognition layer, a color anomaly recognition layer and an anomaly output layer;

[0034] The second data input layer is used to input the second classified image into the model and perform image preprocessing;

[0035] The second feature extraction layer performs feature extraction on the preprocessed second classification image to obtain a second feature data set;

[0036] The texture anomaly recognition layer recognizes the texture features in the second feature data set to obtain texture anomaly data;

[0037] The color anomaly recognition layer recognizes the color features in the second feature data set to obtain color anomaly data;

[0038] The abnormality output layer is used to organize the texture abnormality data and the color abnormality data in the second classified image into second abnormality data, and output the second abnormality data.

[0039] The training process of the texture anomaly recognition layer is as follows: obtaining a fourth data set, wherein the fourth data set is a surface image of a processed copper tube with an anomaly texture area marked thereon; splitting the fourth data set to obtain a fourth training set and a fourth test set; and using the fourth training set and the fourth test set to train the texture anomaly recognition layer;

[0040] The training process of the color anomaly recognition layer is as follows: obtaining a fifth data set, wherein the fifth data set is an image of a processed copper tube surface with color anomaly areas marked; splitting the fifth data set to obtain a fifth training set and a fifth test set; and using the fifth training set and the fifth test set to train the color anomaly recognition layer.

[0041] The third surface recognition model includes a third data input layer, a third feature extraction layer, a defect feature recognition layer and a result output layer;

[0042] The third data input layer inputs the second abnormal data of the second classified image and the first abnormal data corresponding to the first classified image into the model;

[0043] The third feature extraction layer extracts features from the second abnormal data and the first abnormal data to obtain abnormal feature data;

[0044] The defect feature recognition layer is used to recognize the abnormal feature data to obtain third abnormal data;

[0045] The result output layer is used to output the third abnormal data.

[0046] The training process of the third surface recognition model is:

[0047] A sixth data set is obtained, wherein the sixth data set includes historical second classified images, historical second abnormal data, historical first abnormal data, and defect labels; the sixth data set is split to obtain a sixth training set and a sixth test set; the initialized third surface recognition model is trained using the sixth training set, and then the trained third surface recognition model is optimized and parameters are adjusted using the sixth test set.

[0048] The process of obtaining the defect label of the historical second classification image is as follows:

[0049] Determine that the historical second classification image corresponds to a second distribution area on the surface of the copper tube;

[0050] Divide the second distribution area into three equal parts to obtain a first verification area, a second verification area and a third verification area;

[0051] Performing uniformity identification on the first verification area; checking the thickness data of the spray paint by a thickness measuring device, and obtaining first verification data according to the difference between the thickness data and the standard thickness data;

[0052] Conducting an adhesion test on the second verification area; conducting a test on the second verification area using a peeling test device, and obtaining second verification data according to the peeling area and peeling volume of the spray paint;

[0053] Conducting a corrosion resistance test on the third verification area; obtaining a working place of the copper pipe, setting a corrosion environment according to the working place, and obtaining third verification data according to the corrosion area and corrosion rate of the spray paint in the corrosion environment;

[0054] A defect label is obtained by fitting according to the first verification data, the second verification data and the third verification data.

[0055] The warning process of the defect warning module is: obtain all the second classification images of the processed copper tube, and obtain the third abnormal data through the third surface recognition model; set the defect threshold according to the historical data; make an abnormal judgment on the third abnormal data through the defect threshold, and issue a warning message.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. The present invention constructs a first surface recognition model, identifies the collected first image set through dimensions such as pollutant characteristics, flatness characteristics and roughness characteristics, and divides the first image set according to the identified pollutant distribution data, flatness distribution data and roughness distribution data to obtain a first classified image and a first abnormal data, so as to accurately understand the defective area on the surface of the copper tube to be processed.

[0058] 2. The present invention divides the second image set in the same manner as the first image set divides the first classified image to obtain the second image set; constructs a second surface recognition model, identifies the texture features and color features of the second classified image through the second surface recognition model, and obtains texture abnormality data and color abnormality data; outputs the texture abnormality data and the color abnormality data as the second abnormality data; and accurately understands the defective area on the surface of the processed copper tube.

[0059] 3. The present invention obtains a third surface recognition model by training the historical second classification image, the historical second abnormal data, the historical first abnormal data and the defect label; the second abnormal data of the second classification area and the first abnormal data corresponding to the first classification image are identified by the third surface recognition model to obtain the third abnormal data; the defects of the paint spraying on the copper tube surface can be accurately and comprehensively identified by the third abnormal data. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a structural schematic diagram of a copper tube processing surface defect detection system of the present invention;

[0061] Figure 2 It is a schematic diagram of the structure of the first surface recognition model of the present invention;

[0062] Figure 3 is a schematic diagram of the structure of the second surface recognition model of the present invention;

[0063] Figure 4 It is a schematic diagram of the structure of the third surface recognition model of the present invention. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] In order to accurately identify the defective area of ​​the copper tube surface during the painting process, the present invention proposes a copper tube processing surface defect detection system.

[0066] Embodiment 1

[0067] The present invention proposes a copper tube processing surface defect detection system, the structure of the system is as follows Figure 1 As shown, it includes: a first data acquisition module, a first defect recognition module, a second data acquisition module, a data processing module, a second defect recognition module and a defect warning module.

[0068] The first data acquisition module pre-processes the copper tube to obtain the copper tube to be processed; and collects images of the surface of the copper tube to be processed to obtain a first image set.

[0069] The first defect recognition module constructs a first surface recognition model, recognizes the first image set through the first surface recognition model, divides the area according to the contaminant characteristics, flatness characteristics and roughness characteristics of the first image set, and obtains multiple first classified images and corresponding first abnormal data.

[0070] The present invention constructs a first surface recognition model based on a convolutional neural network model;

[0071] The structure of the first surface recognition model is as follows Figure 2 As shown, it includes a first data input layer, a first feature extraction layer, a pollutant feature recognition layer, a flatness feature recognition layer, a roughness feature recognition layer and an image segmentation layer;

[0072] The first data input layer is used to input the first image set into the model and perform image preprocessing;

[0073] The first feature extraction layer extracts features from the preprocessed first image set to obtain a first feature data set;

[0074] The pollutant feature recognition layer is used to recognize the first feature data set to obtain pollutant distribution data;

[0075] The flatness feature recognition layer is used to recognize the first feature data set to obtain flatness distribution data;

[0076] The roughness feature recognition layer is used to recognize the first feature data set to obtain roughness distribution data;

[0077] The image division layer divides the first image set according to the pollutant distribution data, flatness distribution data and roughness distribution data to obtain a first classified image; and uses the pollutant distribution data, flatness distribution data and roughness distribution data as first abnormal data.

[0078] The present invention constructs a first surface recognition model based on a convolutional neural network model, identifies the collected first image set through dimensions such as pollutant characteristics, flatness characteristics and roughness characteristics, and divides the first image set according to the identified pollutant distribution data, flatness distribution data and roughness distribution data to obtain a first classified image; the first classified image can be used to accurately understand the defects on the surface of the copper tube to be processed.

[0079] The training process of the pollutant feature recognition layer is as follows:

[0080] Acquire a first data set, which is a surface image of a copper tube to be processed with pollutant areas marked thereon; the pollutant areas include grease areas, dust areas, water stain areas, and oxide areas;

[0081] Splitting the first data set according to a ratio to obtain a first training set and a first test set;

[0082] The first training set is used to train the initialized pollutant feature recognition layer, and the first test set is used to optimize and adjust the parameters of the trained pollutant feature recognition layer.

[0083] The training process of the flatness feature recognition layer:

[0084] Acquire a second data set, wherein the second data set is a surface image of the copper tube to be processed with flatness abnormality areas marked thereon; the flatness abnormality areas include scratch areas, convex areas, pit areas, and crack areas;

[0085] Splitting the second data set according to a ratio to obtain a second training set and a second test set;

[0086] The initialized flatness feature recognition layer is trained using the second training set, and then the trained flatness feature recognition layer is optimized and parameter adjusted using the second test set.

[0087] The training process of the roughness feature recognition layer is:

[0088] Acquire a third data set, wherein the third data set is a surface image of the copper tube to be processed with an abnormal roughness region marked thereon; the abnormal roughness region is identified by a roughness detection device;

[0089] Splitting the third data set according to a ratio to obtain a third training set and a third test set;

[0090] The initialized roughness feature recognition layer is trained using the third training set, and then the trained roughness feature recognition layer is optimized and parameter adjusted using the third test set.

[0091] During the training process of the first surface recognition model, the present invention trains and optimizes the contaminant recognition layer by using the surface image of the copper tube to be processed marked with the contaminant area; trains and optimizes the flatness recognition layer by using the surface image of the copper tube to be processed marked with the flatness abnormality area; trains and optimizes the roughness recognition layer by using the surface image of the copper tube to be processed marked with the roughness abnormality area; and ensures the recognition accuracy of the first surface recognition model for the surface defects of the copper tube to be processed.

[0092] The second data acquisition module sprays paint on the copper tube to be processed to obtain a processed copper tube; and collects images of the surface of the processed copper tube to obtain a second image set.

[0093] The data processing module obtains a second image set, and divides the second image set according to the division method of the first classification image in the first image set to obtain a plurality of second classification images.

[0094] The process of dividing the second image set to obtain the second classified image is as follows:

[0095] Establishing a three-dimensional model of the copper tube according to the copper tube to be processed, and acquiring a first image set and a first classified image;

[0096] Restoring the first image set to the copper tube three-dimensional model to obtain the regional distribution of the first classified image on the copper tube three-dimensional model, and obtaining a first classified three-dimensional model;

[0097] The second image set is restored to the first classified three-dimensional model; and the second image set is divided according to the regional distribution of the first classified images on the first classified three-dimensional model to obtain second classified images.

[0098] The present invention constructs a copper tube three-dimensional model according to the copper tube to be processed, and restores the division method of the first classification image in the first image set to the copper tube three-dimensional model to obtain the first classification three-dimensional model; then restores the second image set to the first classification three-dimensional model, and segments the second image set according to the division method of the first classification image to obtain a second classification image, thereby ensuring the segmentation consistency of the second classification image and the first classification image.

[0099] The second defect recognition module constructs a second surface recognition model; recognizes the second classified image through the second surface recognition model, and obtains second abnormal data through texture feature and color feature recognition.

[0100] The present invention constructs a second surface recognition model based on a convolutional neural network model;

[0101] The structure of the second surface recognition model is as follows Figure 3 As shown, it includes a second data input layer, a second feature extraction layer, a texture anomaly recognition layer, a color anomaly recognition layer and an anomaly output layer;

[0102] The second data input layer is used to input the second classified image into the model and perform image preprocessing;

[0103] The second feature extraction layer performs feature extraction on the preprocessed second classification image to obtain a second feature data set;

[0104] The texture anomaly recognition layer recognizes the texture features in the second feature data set to obtain texture anomaly data;

[0105] The color anomaly recognition layer recognizes the color features in the second feature data set to obtain color anomaly data;

[0106] The abnormality output layer is used to organize the texture abnormality data and the color abnormality data in the second classified image into second abnormality data, and output the second abnormality data.

[0107] The present invention constructs a second surface recognition model with a convolutional neural network model, recognizes the texture features and color features of the second classified image, obtains texture abnormality data and color abnormality data; outputs the texture abnormality data and the color abnormality data as second abnormality data; and can accurately understand the defective area on the surface of the processed copper tube.

[0108] Among them, the training process of the texture anomaly recognition layer is:

[0109] Acquire a fourth data set, wherein the fourth data set is a surface image of a processed copper tube with annotated texture abnormality areas; split the fourth data set according to a ratio to obtain a fourth training set and a fourth test set;

[0110] The initialized texture anomaly recognition layer is trained using the fourth training set, and then the trained texture anomaly recognition layer is optimized and parameters are adjusted using the fourth test set.

[0111] The training process of the color anomaly recognition layer is:

[0112] Acquire a fifth data set, wherein the fifth data set is a surface image of a processed copper tube with an abnormal color area marked; split the fifth data set according to a ratio to obtain a fifth training set and a fifth test set;

[0113] The initialized color anomaly recognition layer is trained using the fifth training set, and then the trained color anomaly recognition layer is optimized and parameter adjusted using the fifth test set.

[0114] During the training process of the second surface recognition model, the present invention trains and optimizes the texture anomaly recognition layer by using the processed copper tube surface image marked with texture anomaly areas; and trains and optimizes the color anomaly recognition layer by using the processed copper tube surface image marked with color anomaly areas; thereby ensuring the recognition accuracy of the second surface recognition model for surface defects of the processed copper tube.

[0115] The defect warning module constructs a third surface recognition model, inputs the second abnormal data of the second classification image and the first abnormal data corresponding to the first classification image into the third surface recognition model to obtain third abnormal data; and identifies the defective area of ​​the processed surface according to the third abnormal data.

[0116] The present invention constructs a third surface recognition model based on a deep neural network;

[0117] The structure of the third surface recognition model is as follows Figure 4 As shown, it includes a third data input layer, a third feature extraction layer, a defect feature recognition layer and a result output layer;

[0118] The third data input layer inputs the second abnormal data of the second classified image and the first abnormal data corresponding to the first classified image into the model;

[0119] The third feature extraction layer extracts features from the second abnormal data and the first abnormal data to obtain abnormal feature data;

[0120] The defect feature recognition layer is used to recognize the abnormal feature data to obtain third abnormal data;

[0121] The result output layer is used to output the third abnormal data.

[0122] The present invention constructs a third surface recognition model based on a deep neural network. The second abnormal data of the second classification area and the first abnormal data corresponding to the first classification image are identified by the third surface recognition model to obtain third abnormal data. The third abnormal data can be used to accurately identify and predict the defects of the paint on the surface of the copper tube.

[0123] The training process of the third surface recognition model is:

[0124] Acquire a sixth data set, wherein the sixth data set includes a historical second classified image, historical second abnormal data, historical first abnormal data, and a defect label; the historical second classified image corresponds to the historical second abnormal data and the historical first abnormal data;

[0125] Splitting the sixth data set according to a ratio to obtain a sixth training set and a sixth test set;

[0126] The initialized third surface recognition model is trained using the sixth training set, and then the trained third surface recognition model is optimized and parameters are adjusted using the sixth test set.

[0127] During the training process of the third surface recognition model, the present invention trains and optimizes the third surface recognition model through the second abnormal data, the first abnormal data and the defect label of the second classified image, and ensures the recognition accuracy of the third surface recognition model for the first abnormal data and the second abnormal data.

[0128] The process of obtaining the defect label of the historical second classification image is as follows:

[0129] Determine that the historical second classification image corresponds to a second distribution area on the surface of the copper tube;

[0130] Divide the second distribution area into three equal parts to obtain a first verification area, a second verification area and a third verification area;

[0131] Performing uniformity identification on the first verification area; checking the thickness data of the spray paint by a thickness measuring device, and obtaining first verification data according to the difference between the thickness data and the standard thickness data;

[0132] Conducting an adhesion test on the second verification area; conducting a test on the second verification area using a peeling test device, and obtaining second verification data according to the peeling area and peeling volume of the spray paint;

[0133] Conducting a corrosion resistance test on the third verification area; obtaining a working place of the copper pipe, setting a corrosion environment according to the working place, and obtaining third verification data according to the corrosion area and corrosion rate of the spray paint in the corrosion environment;

[0134] A defect label is obtained by fitting according to the first verification data, the second verification data and the third verification data.

[0135] The present invention obtains a second distribution area on the surface of a processed copper tube corresponding to a second classification image, and equally divides the second distribution area into a first verification area, a second verification area and a third verification area; performs uniformity identification, adhesion experiment and corrosion resistance experiment respectively to obtain first verification data, second verification data and third verification data; and obtains defect labels according to the first verification data, the second verification data and the third verification data, so as to accurately reflect and predict the quality of the paint spraying on the copper tube surface.

[0136] The warning process of the defect warning module is: obtain all the second classification images of the processed copper tube, and obtain the third abnormal data through the third surface recognition model; set the defect threshold according to the historical data; make an abnormal judgment on the third abnormal data through the defect threshold, and issue a warning message.

[0137] The present invention determines a defect threshold value according to historical data; judges the third abnormal data identified by the defect threshold value, and issues a warning message when the third abnormal data is abnormal; and accurately issues a warning for abnormal situations.

[0138] The present invention collects images of the surface of a copper tube to be processed to obtain a first image set; constructs a first surface recognition model to recognize the first image set to obtain a first classified image and first abnormal data; spray-paints the copper tube to obtain a processed copper tube; collects images of the surface of the processed copper tube to obtain a second image set; divides the second image set according to the division method of the first classified image in the first image set to obtain a second classified image; constructs a second surface recognition model to recognize the second classified image to obtain second abnormal data; constructs a third surface recognition model to train the second abnormal data of the second classified image and the first abnormal data corresponding to the first classified image to obtain third abnormal data; and accurately recognizes the defects of spray-painting on the surface of the copper tube through the third abnormal data.

[0139] Embodiment 2

[0140] Copper tube processing factory A is mainly engaged in the production and reprocessing of copper tubes, which includes spray painting the surface of copper tubes to ensure the working quality and service life of copper tubes. In order to better identify the processing effect of copper tube spray painting, copper tube processing factory A adopts a copper tube processing surface defect detection system described in the present invention to identify the spray painting process.

[0141] The structure of the system is as follows Figure 1 As shown, it includes: a first data acquisition module, a first defect recognition module, a second data acquisition module, a data processing module, a second defect recognition module and a defect warning module.

[0142] The first data acquisition module pre-processes the copper tube to obtain the copper tube to be processed; and collects images of the surface of the copper tube to be processed to obtain a first image set.

[0143] The first defect recognition module constructs a first surface recognition model, recognizes the first image set through the first surface recognition model, divides the area according to the contaminant characteristics, flatness characteristics and roughness characteristics of the first image set, and obtains multiple first classified images and corresponding first abnormal data.

[0144] The first surface recognition model is constructed based on a convolutional neural network model, and includes a first data input layer, a first feature extraction layer, a pollutant feature recognition layer, a flatness feature recognition layer, a roughness feature recognition layer, and an image segmentation layer; it is used to identify surface defects of copper tubes before painting, and the defect types include pollutant areas, abnormal flatness areas, and abnormal roughness areas.

[0145] The second data acquisition module sprays paint on the copper tube to be processed to obtain a processed copper tube; and collects images of the surface of the processed copper tube to obtain a second image set.

[0146] The data processing module obtains a second image set, and divides the second image set according to the division method of the first classification image in the first image set to obtain a plurality of second classification images.

[0147] The process of dividing the second image set to obtain the second classified image is as follows:

[0148] Establishing a three-dimensional model of the copper tube according to the copper tube to be processed, and acquiring a first image set and a first classified image;

[0149] Restoring the first image set to the copper tube three-dimensional model to obtain the regional distribution of the first classified image on the copper tube three-dimensional model, and obtaining a first classified three-dimensional model;

[0150] The second image set is restored to the first classified three-dimensional model; and the second image set is divided according to the regional distribution of the first classified images on the first classified three-dimensional model to obtain second classified images.

[0151] The second defect recognition module constructs a second surface recognition model; recognizes the second classified image through the second surface recognition model, and obtains second abnormal data through texture feature and color feature recognition.

[0152] The second surface recognition model is constructed based on a convolutional neural network model, and includes a second data input layer, a second feature extraction layer, a texture anomaly recognition layer, a color anomaly recognition layer and an anomaly output layer; and is used to identify defective areas on the surface of a processed copper tube.

[0153] The defect warning module constructs a third surface recognition model, inputs the second abnormal data of the second classification image and the first abnormal data corresponding to the first classification image into the third surface recognition model to obtain third abnormal data; and identifies the defective area of ​​the processed surface according to the third abnormal data.

[0154] The third surface recognition model is constructed based on a deep neural network, and includes a third data input layer, a third feature extraction layer, a defect feature recognition layer and a result output layer; it is used to identify comprehensive paint defects based on the defect data of the copper tube surface before and after painting.

[0155] The training process of the third surface recognition model is:

[0156] A sixth data set is obtained, wherein the sixth data set includes historical second classified images, historical second abnormal data, historical first abnormal data, and defect labels; the sixth data set is split to obtain a sixth training set and a sixth test set; the initialized third surface recognition model is trained using the sixth training set, and then the trained third surface recognition model is optimized and parameters are adjusted using the sixth test set.

[0157] The process of obtaining the defect label of the historical second classification image is as follows:

[0158] Determine that the historical second classification image corresponds to a second distribution area on the surface of the copper tube;

[0159] Divide the second distribution area into three equal parts to obtain a first verification area, a second verification area and a third verification area;

[0160] Performing uniformity identification on the first verification area; checking the thickness data of the spray paint by a thickness measuring device, and obtaining first verification data according to the difference between the thickness data and the standard thickness data;

[0161] Conducting an adhesion test on the second verification area; conducting a test on the second verification area using a peeling test device, and obtaining second verification data according to the peeling area and peeling volume of the spray paint;

[0162] Conducting a corrosion resistance test on the third verification area; obtaining a working place of the copper pipe, setting a corrosion environment according to the working place, and obtaining third verification data according to the corrosion area and corrosion rate of the spray paint in the corrosion environment;

[0163] A defect label is obtained by fitting according to the first verification data, the second verification data and the third verification data.

[0164] The present invention collects images of the surface of a copper tube to be processed to obtain a first image set; constructs a first surface recognition model to recognize the first image set to obtain a first classified image and first abnormal data; spray-paints the copper tube to obtain a processed copper tube; collects images of the surface of the processed copper tube to obtain a second image set; divides the second image set according to the division method of the first classified image in the first image set to obtain a second classified image; constructs a second surface recognition model to recognize the second classified image to obtain second abnormal data; constructs a third surface recognition model to train the second abnormal data of the second classified image and the first abnormal data corresponding to the first classified image to obtain third abnormal data; and accurately recognizes the defects of spray-painting on the surface of the copper tube through the third abnormal data.

[0165] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A copper tube processing surface defect detection system, characterized in that: include: A first data acquisition module pre-processes the copper tube to obtain the copper tube to be processed; Capturing images of the surface of the copper tube to be processed to obtain a first image set; A first defect recognition module is configured to construct a first surface recognition model, recognize the first image set by using the first surface recognition model, divide the first image set into regions according to the contaminant features, flatness features, and roughness features of the first image set, and obtain a plurality of first classified images and corresponding first abnormal data; A second data acquisition module sprays paint on the copper tube to be processed to obtain a processed copper tube; and collects images of the surface of the processed copper tube to obtain a second image set; A data processing module obtains a second image set, and divides the second image set according to the division method of the first classification image in the first image set to obtain a plurality of second classification images; A second defect recognition module constructs a second surface recognition model; recognizes the second classified image through the second surface recognition model, and obtains second abnormal data through texture feature and color feature recognition; The defect warning module constructs a third surface recognition model, inputs the second abnormal data of the second classification image and the first abnormal data corresponding to the first classification image into the third surface recognition model to obtain third abnormal data; and identifies the defective area of ​​the processed surface according to the third abnormal data.

2. A copper tube processing surface defect detection system according to claim 1, characterized in that: The first surface recognition model includes a first data input layer, a first feature extraction layer, a pollutant feature recognition layer, a flatness feature recognition layer, a roughness feature recognition layer and an image segmentation layer; The first data input layer is used to input the first image set into the model and perform image preprocessing; The first feature extraction layer extracts features from the preprocessed first image set to obtain a first feature data set; The pollutant feature recognition layer is used to recognize the first feature data set to obtain pollutant distribution data; The flatness feature recognition layer is used to recognize the first feature data set to obtain flatness distribution data; The roughness feature recognition layer is used to recognize the first feature data set to obtain roughness distribution data; The image division layer divides the first image set according to the pollutant distribution data, flatness distribution data and roughness distribution data to obtain a first classified image; and uses the pollutant distribution data, flatness distribution data and roughness distribution data as first abnormal data.

3. A copper tube processing surface defect detection system according to claim 2, characterized in that: The training process of the pollutant feature recognition layer is as follows: Acquire a first data set, which is a surface image of a copper tube to be processed with a pollutant area marked thereon; split the first data set to obtain a first training set and a first test set; and use the first training set and the first test set to train to obtain a pollutant feature recognition layer; The training process of the flatness feature recognition layer is as follows: Acquire a second data set, which is a surface image of a copper tube to be processed with an abnormal flatness area marked thereon; split the second data set to obtain a second training set and a second test set; and use the second training set and the second test set to train to obtain a flatness feature recognition layer; The training process of the roughness feature recognition layer is: A third data set is obtained, wherein the third data set is a surface image of the copper tube to be processed with annotated roughness abnormal areas; the third data set is split to obtain a third training set and a third test set; and the third training set and the third test set are used to train to obtain a roughness feature recognition layer.

4. A copper tube processing surface defect detection system according to claim 1, characterized in that: The process of dividing the second image set to obtain the second classified image is as follows: Establishing a three-dimensional model of the copper tube according to the copper tube to be processed, and acquiring a first image set and a first classified image; Restoring the first image set to the copper tube three-dimensional model to obtain the regional distribution of the first classified image on the copper tube three-dimensional model, and obtaining a first classified three-dimensional model; The second image set is restored to the first classified three-dimensional model; and the second image set is divided according to the regional distribution of the first classified images on the first classified three-dimensional model to obtain second classified images.

5. A copper tube processing surface defect detection system according to claim 1, characterized in that: The second surface recognition model includes a second data input layer, a second feature extraction layer, a texture anomaly recognition layer, a color anomaly recognition layer and an anomaly output layer; The second data input layer is used to input the second classified image into the model and perform image preprocessing; The second feature extraction layer performs feature extraction on the preprocessed second classification image to obtain a second feature data set; The texture anomaly recognition layer recognizes the texture features in the second feature data set to obtain texture anomaly data; The color anomaly recognition layer recognizes the color features in the second feature data set to obtain color anomaly data; The abnormality output layer is used to organize the texture abnormality data and the color abnormality data in the second classified image into second abnormality data, and output the second abnormality data.

6. A copper tube processing surface defect detection system according to claim 5, characterized in that: The training process of the texture anomaly recognition layer is as follows: obtaining a fourth data set, the fourth data set being a surface image of a processed copper tube with annotated texture anomaly areas; splitting the fourth data set to obtain a fourth training set and a fourth test set; and using the fourth training set and the fourth test set to train the texture anomaly recognition layer; The training process of the color anomaly recognition layer is as follows: obtaining a fifth data set, wherein the fifth data set is an image of a processed copper tube surface with color anomaly areas marked; splitting the fifth data set to obtain a fifth training set and a fifth test set; and using the fifth training set and the fifth test set to train the color anomaly recognition layer.

7. A copper tube processing surface defect detection system according to claim 1, characterized in that: The third surface recognition model includes a third data input layer, a third feature extraction layer, a defect feature recognition layer and a result output layer; The third data input layer inputs the second abnormal data of the second classified image and the first abnormal data corresponding to the first classified image into the model; The third feature extraction layer extracts features from the second abnormal data and the first abnormal data to obtain abnormal feature data; The defect feature recognition layer is used to recognize the abnormal feature data to obtain third abnormal data; The result output layer is used to output the third abnormal data.

8. A copper tube processing surface defect detection system according to claim 7, characterized in that: The training process of the third surface recognition model is: A sixth data set is obtained, wherein the sixth data set includes historical second classified images, historical second abnormal data, historical first abnormal data, and defect labels; the sixth data set is split to obtain a sixth training set and a sixth test set; the initialized third surface recognition model is trained using the sixth training set, and then the trained third surface recognition model is optimized and parameters are adjusted using the sixth test set.

9. A copper tube processing surface defect detection system according to claim 8, characterized in that: The process of obtaining the defect label of the historical second classification image is as follows: Determine that the historical second classification image corresponds to a second distribution area on the surface of the copper tube; Divide the second distribution area into three equal parts to obtain a first verification area, a second verification area and a third verification area; Performing uniformity identification on the first verification area; checking the thickness data of the spray paint by a thickness measuring device, and obtaining first verification data according to the difference between the thickness data and the standard thickness data; Conducting an adhesion test on the second verification area; conducting a test on the second verification area using a peeling test device, and obtaining second verification data according to the peeling area and peeling volume of the spray paint; Conducting a corrosion resistance test on the third verification area; obtaining a working place of the copper pipe, setting a corrosion environment according to the working place, and obtaining third verification data according to the corrosion area and corrosion rate of the spray paint in the corrosion environment; A defect label is obtained by fitting according to the first verification data, the second verification data and the third verification data.

10. A copper tube processing surface defect detection system according to claim 1, characterized in that: The early warning process of the defect early warning module is as follows: All second classification images of the processed copper tubes are obtained, and third abnormal data are obtained through the third surface recognition model; a defect threshold is set according to historical data; an abnormality judgment is made on the third abnormal data through the defect threshold, and an early warning message is issued.

Images