CN112669292B - Method for detecting and classifying defects on painted surface of aircraft skin - Google Patents
Method for detecting and classifying defects on painted surface of aircraft skin Download PDFInfo
- Publication number
- CN112669292B CN112669292B CN202011626199.0A CN202011626199A CN112669292B CN 112669292 B CN112669292 B CN 112669292B CN 202011626199 A CN202011626199 A CN 202011626199A CN 112669292 B CN112669292 B CN 112669292B
- Authority
- CN
- China
- Prior art keywords
- convolution kernels
- inclusion
- module
- classification
- neural network
- 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.)
- Active
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The invention discloses a method for detecting and classifying defects on a painted surface of an aircraft skin, which comprises the following steps: s1, collecting an image of the painted surface of the aircraft skin; s2, performing two-classification defect detection on the image acquired in the step S1 based on the surface smoothness; s3, multi-classification detection is carried out on the surface defects of the aircraft skin paint spraying on the basis of the simplified GoogLeNet convolutional neural network model; and S4, outputting a multi-classification result. The method for detecting and classifying the defects on the surface of the aircraft skin sprayed paint integrates a traditional image defect detection two-classification method and a defect multi-classification frame of a GoogLeNet convolution neural network model based on deep learning, and combines two-classification and multi-classification, so that the method not only effectively improves the operation efficiency and robustness of the algorithm, but also effectively shortens the execution time of the algorithm, improves the accuracy of the detection of the defects on the surface of the aircraft skin sprayed paint, and can effectively meet the requirements of intelligent detection and classification of the defects on the surface of the aircraft skin sprayed paint.
Description
Technical Field
The invention relates to a method for realizing detection and classification of defects on a painted surface of an aircraft skin, and belongs to the technical field of defect detection.
Background
The surface coating process of the aircraft skin is complex, once the coating and the matrix material are debonded and cracks are generated in the using process, the appearance is influenced, and the operation safety of the aircraft is also seriously influenced, so that the detection of the surface defects of the aircraft skin is an important means for inspecting the delivery quality of the aircraft. The surface defects of the aircraft skin are various, and the defects typically comprise particles, scratches, garbage, flowing, orange peel, extremely thick flanging and the like. In most cases, the size of the surface defect of the aircraft skin is measured in millimeters, and the defect area is small and is not uniformly dispersed. Traditional defect detection mainly relies on artifical visual inspection, not only has limitations such as the cost of labor is high, work efficiency is low, the real-time is poor, and the effect of examining is easily influenced by artifical subjective factors such as eyesight, health. With the improvement of the automation degree of aircraft production and manufacturing, the manual visual detection method cannot meet the development requirement, so that the automatic intellectualization of the detection method for the surface defects of the aircraft skin is urgently needed in the field.
At present, the automatic detection of the surface defects of the aircraft skin is still in a starting stage in China, the existing detection methods have the problems of large calculated amount, poor accuracy, low reliability and the like, and the industrial requirements are difficult to meet, so that the technical problem of how to shorten the execution time of the algorithm and improve the detection accuracy of the surface defects of the aircraft skin is urgently needed to be solved.
Disclosure of Invention
Aiming at the problems in the prior art, the invention aims to provide a method for detecting and classifying the defects of the painted surface of the aircraft skin.
In order to realize the purpose of the invention, the technical scheme adopted by the invention is as follows:
a method for realizing detection and classification of defects on a painted surface of an aircraft skin comprises the following steps:
s1, collecting an image of the paint spraying surface of the aircraft skin;
s2, performing two-classification defect detection on the image acquired in the step S1 based on the surface smoothness, namely: firstly, preprocessing an acquired image; then, block detection and positioning are carried out on the preprocessed image, and pixels with defects are found; then, carrying out surface smoothness estimation on the pixels with the defects and the pixels of the whole image by using an information entropy function, judging whether the image has the defects or not based on the surface smoothness, and dividing the image into a defective image and a non-defective image to realize two-classification defect detection on the image;
s3, multi-classification detection is carried out on the surface defects of the aircraft skin paint spraying on the basis of the simplified GoogLeNet convolution neural network model, namely: constructing a simplified GoogLeNet convolutional neural network model, and then performing multi-classification detection on the defect images obtained by two classification in the step S2 through the constructed simplified GoogLeNet convolutional neural network model; the GoogleLeNet convolutional neural network model comprises five layers, wherein the five layers are respectively a first layer, a second layer, a third layer, a fourth layer and a fifth layer, the first layer and the second layer are common convolutional layers, the third layer is composed of an inclusion-A module and an inclusion-B module, the fourth layer is composed of an inclusion-A, Inception-B, Inception-C, Inception-D module and an inclusion-E module, and the fifth layer is composed of an inclusion-A module and an inclusion-B module; the simplified GoogLeNet convolutional neural network model is a GoogLeNet convolutional neural network model with an addition-B module and an addition-C module removed; according to the invention, the simplified GoogLeNet convolutional neural network model has a faster operation speed on the premise that the accuracy is not reduced compared with the GoogLeNet convolutional neural network model, and the execution time of the algorithm can be effectively shortened.
And S4, outputting a multi-classification result.
In one embodiment, the preprocessing the image acquired in step S1 in step S2 includes: the image acquired in step S1 is subjected to preprocessing of gaussian filtering (to reduce noise of the image), histogram equalization (to enhance the image), Ostu adaptive threshold segmentation and erosion.
In one embodiment, in step S2, Blob is used to perform block detection on the preprocessed image.
In a preferred embodiment, in step S2, the operation of performing binary defect detection on the image acquired in step S1 based on the surface smoothness is as follows: firstly, preprocessing of Gaussian filtering, histogram equalization, Ostu self-adaptive threshold segmentation and corrosion is carried out on the image collected in the step S1; then performing block detection on the preprocessed image by using Blob to find out the pixels with defects; and then, performing surface smoothness estimation on the pixels with the defects and the pixels of the whole image by using an information entropy function, setting a surface smoothness threshold (the surface smoothness threshold represents the number of the pixels with the defects, and judging that the image is defective when the number of the pixels with the defects reaches the threshold) to judge whether the image has the defects or not, and dividing the image into a defective image and a non-defective image to realize binary defect detection on the image.
In one embodiment, step S3 specifically includes the following operations:
s31, data set generation: using the collected image of the paint spraying surface of the aircraft skin for manufacturing a data set, and dividing the data set into a training sample data set and a test sample data set;
s32, constructing a GoogleLeNet convolutional neural network model: the GoogLeNet convolutional neural network model comprises five layers, wherein the five layers are respectively a first layer, a second layer, a third layer, a fourth layer and a fifth layer, the first layer and the second layer are common convolutional layers, the third layer is composed of an inclusion-A module and an inclusion-B module, the fourth layer is composed of an inclusion-A, Inception-B, Inception-C, Inception-D module and an inclusion-E module, and the fifth layer is composed of an inclusion-A module and an inclusion-B module;
s33, constructing a simplified GoogLeNet convolutional neural network model: removing the inclusion-B and inclusion-C modules of the GoogLeNet convolutional neural network model in the step S32, and constructing a simplified GoogLeNet convolutional neural network model;
s34, training a simplified GoogLeNet convolutional neural network model: inputting the images in the training sample data set into a simplified GoogLeNet convolutional neural network model for feature recognition, and acquiring an optimal simplified GoogLeNet convolutional neural network model;
s35, testing a simplified GoogLeNet convolutional neural network model: inputting the test sample data set into a trained simplified GoogLeNet convolutional neural network model, and verifying the accuracy;
s36, multi-classification detection: and (4) outputting the defect images classified in the step (S2) into the simplified GoogLeNet convolutional neural network model after testing, and realizing multi-classification detection on the surface defects of the aircraft skin paint spraying through the simplified GoogLeNet convolutional neural network model.
In a preferred embodiment, the data set created in step S31 includes images of various common defect types on the painted surface of the aircraft skin, including but not limited to bubbles, particles, debris, drooling, and flares. In the process of making the data set, the distinction of the specific defect types of the images can be distinguished by surface painting professionals.
According to a preferable scheme, the simplified GoogleLeNet convolutional neural network model constructed in the step S33 sequentially comprises a first convolutional layer, a second convolutional layer, a third inclusion-A module, a fourth inclusion-D module, a fourth inclusion-E module and a fifth inclusion-A module; the operation of each convolutional layer and module is as follows:
the first convolution layer firstly uses convolution kernels of 7 × 7, the number of the convolution kernels is 64, then 3 × 3 kernels are used for maximum pooling, and the sliding step length is 2;
the second convolution layer firstly uses convolution kernels with the number of 3 x 3, the number of the convolution kernels is 192, the sliding step length is 1, and then the convolution kernels with the number of 3 x 3 are used for maximum pooling, and the sliding step length is 2;
the third inclusion-a module includes four branches, which are specifically: using convolution kernels with 1 x 1, wherein the number of the convolution kernels is 64, and the sliding step is 1; 2) firstly, using 1 × 1 convolution kernel, wherein the number of the convolution kernels is 96, then using 3 × 3 convolution kernels, wherein the number of the convolution kernels is 128, and the sliding step length is 1; 3) firstly, using 1 × 1 convolution kernel, wherein the number of the convolution kernels is 16, then using 5 × 5 convolution kernels, wherein the number of the convolution kernels is 32, and the sliding step length is 1; 4) firstly, performing maximum pooling by using 3 × 3 kernels, then using 1 × 1 convolution kernels, wherein the number of the convolution kernels is 32, and the sliding step length is 1; finally, connecting the output results of the four branches, and continuing the next step;
the fourth inclusion-a module is identical to the third inclusion-a module;
the fourth inclusion-D module is identical to the third inclusion-a module;
the fourth inclusion-E module includes five branches, which specifically are: 1) using 1 × 1 convolution kernels, wherein the number of the convolution kernels is 64, and the sliding step length is 1; 2) firstly, 1 × 1 convolution kernels are used, the number of the convolution kernels is 96, then 3 × 3 convolution kernels are used, the number of the convolution kernels is 128, and the sliding step length is 1; 3) firstly, using 1 × 1 convolution kernel, wherein the number of the convolution kernels is 16, then using 5 × 5 convolution kernels, wherein the number of the convolution kernels is 32, and the sliding step length is 1; 4) firstly, performing maximum pooling by using 3 × 3 kernels, then using 1 × 1 convolution kernels, wherein the number of the convolution kernels is 32, and the sliding step length is 1; 5) firstly, carrying out average pooling by using 5-by-5 kernels, wherein the sliding step length is 3, then using 1-by-1 convolution kernels, the number of the convolution kernels is 32, the sliding step length is 1, then using two full-connection layers, and then using a Softmax activation function to assist classification; finally, connecting the output results of the first five branches, and continuing the next step;
the fifth inclusion-a module is identical to the third inclusion-a module;
in each convolutional layer and module, the ReLU activation function is used after the convolutional kernel and pooling operations are used.
Compared with the prior art, the invention has the following remarkable beneficial effects:
according to the method for detecting and classifying the paint spraying surface defects of the aircraft skin, the two classification defect detection is firstly carried out on the collected images of the paint spraying surface of the aircraft skin based on the surface smoothness, so that the accurate positioning of the defect area in the paint spraying surface defect detection of the aircraft skin is realized, and the defect part and the normal part of the paint spraying surface of the aircraft skin are accurately distinguished; the method has the advantages that the method is simple in structure, convenient to operate, high in accuracy and capable of achieving classification and identification of different defect types, the defects of known types can be intelligently detected or identified by a common deep learning network, the unknown defects are high in tolerance, flexibility and adaptability are high, the operation efficiency is improved while the network learning performance is met, the algorithm execution time can be effectively shortened, and the accuracy is high; the method provided by the invention can overcome the defects of the traditional manual visual detection, has important significance for reducing raw material waste in the aircraft production process, improving the protective performance, prolonging the service life of aircraft skin, reducing the later maintenance cost and ensuring the safety and reliability, also fills the blank of the surface paint spraying defect detection research of international large airliners, and obtains remarkable progress and unexpected effect compared with the prior art.
Drawings
FIG. 1 is a diagram of a defect image detected by two-classification according to an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further detailed and completely explained by combining specific embodiments.
Examples
The invention provides a method for realizing the detection and classification of the paint spraying surface defects of an aircraft skin, which comprises the following steps:
s1, acquiring images of the skin painting surface of the airplane, for example, acquiring images of the skin painting surface of a civil airplane of a C919 type on the market;
s2, performing two-classification defect detection on the image acquired in the step S1 based on the surface smoothness, specifically as follows:
firstly, preprocessing of Gaussian filtering, histogram equalization, Ostu self-adaptive threshold segmentation and corrosion is carried out on the image collected in the step S1; then performing block detection on the preprocessed image by using Blob to find out the pixels with defects; and then, performing surface smoothness estimation on the pixels with the defects and the pixels of the whole image by using an information entropy function, setting a surface smoothness threshold (the surface smoothness threshold represents the number of the pixels with the defects, and judging that the image is defective when the number of the pixels with the defects reaches the threshold) to judge whether the image has the defects or not, and dividing the image into a defective image and a non-defective image to realize binary defect detection on the image. The defect images detected by two classifications are shown in fig. 1 and include common defects such as bubbles, particles, trash, flowing, flanging and the like, and in the step, only various images in fig. 1 are judged to be defect images, and specific defect types are confirmed by subsequent multiple classifications.
One of the key and difficulty of the paint spraying surface defect detection of the aircraft skin lies in the accurate positioning of the defect area.
The second key and difficulty of the detection of the paint spraying surface defects of the aircraft skin is how to distinguish the paint spraying surface defects of the aircraft skin from the normal parts, and the method introduces an image surface smoothness evaluation method based on information entropy, wherein the information entropy can be used for measuring the richness of image information, and the information entropy based on the richness of the image information is adopted to evaluate the surface smoothness of the image, so that a standard is provided for measuring the paint spraying surface defects of the aircraft skin, and the paint spraying surface defects of the aircraft skin can be accurately distinguished from the normal parts; in the embodiment, the accuracy rate of the two-classification test is 98.3%, the accuracy rate is high, and the two-classification result is accurate.
S3, multi-classification detection is carried out on the surface defects of the aircraft skin paint spraying on the basis of the simplified GoogLeNet convolution neural network model: the method specifically comprises the following steps:
s31, data set generation: using the acquired image of the painted surface of the aircraft skin for manufacturing a data set, and dividing the data set into a training sample data set and a test sample data set; in the manufacturing process of the data set, a professional worker for surface painting can distinguish the defect types of the collected images, wherein the defect types comprise common defect types such as bubbles, particles, garbage, flowing, flanging and the like, and then the well-divided images are respectively divided into a training sample data set and a test sample data set;
s32, constructing a GoogleLeNet convolutional neural network model: the GoogLeNet convolutional neural network model comprises five layers, wherein the five layers are respectively a first layer, a second layer, a third layer, a fourth layer and a fifth layer, the first layer and the second layer are common convolutional layers, the third layer is composed of an inclusion-A module and an inclusion-B module, the fourth layer is composed of an inclusion-A, Inception-B, Inception-C, Inception-D module and an inclusion-E module, and the fifth layer is composed of an inclusion-A module and an inclusion-B module;
s33, constructing a simplified GoogLeNet convolutional neural network model: removing the inclusion-B and inclusion-C modules of the GoogLeNet convolutional neural network model in the step S32, and constructing a simplified GoogLeNet convolutional neural network model;
therefore, the simplified google lenet convolutional neural network model constructed in the embodiment sequentially comprises: the winding device comprises a first winding layer, a second winding layer, a third inclusion-A module, a fourth inclusion-D module, a fourth inclusion-E module and a fifth inclusion-A module; the operation of each convolutional layer and module is as follows:
the first convolution layer firstly uses convolution kernels of 7 × 7, the number of the convolution kernels is 64, then 3 × 3 kernels are used for maximum pooling, and the sliding step length is 2;
the second convolution layer firstly uses 3 × 3 convolution kernels, the number of the convolution kernels is 192, the sliding step length is 1, then 3 × 3 kernels are used for maximum pooling, and the sliding step length is 2;
the third inclusion-a module includes four branches, which specifically are: using 1 × 1 convolution kernels, wherein the number of the convolution kernels is 64, and the sliding step length is 1; 2) firstly, using 1 × 1 convolution kernel, wherein the number of the convolution kernels is 96, then using 3 × 3 convolution kernels, wherein the number of the convolution kernels is 128, and the sliding step length is 1; 3) firstly, using 1 × 1 convolution kernel, wherein the number of the convolution kernels is 16, then using 5 × 5 convolution kernels, wherein the number of the convolution kernels is 32, and the sliding step length is 1; 4) firstly, performing maximum pooling by using 3 × 3 kernels, then using 1 × 1 convolution kernels, wherein the number of the convolution kernels is 32, and the sliding step length is 1; finally, connecting the output results of the four branches, and continuing the next step;
the fourth inclusion-a module is identical to the third inclusion-a module;
the fourth inclusion-D module is identical to the third inclusion-a module;
the fourth inclusion-E module includes five branches, which specifically are: 1) using convolution kernels with 1 x 1, wherein the number of the convolution kernels is 64, and the sliding step is 1; 2) firstly, using 1 × 1 convolution kernel, wherein the number of the convolution kernels is 96, then using 3 × 3 convolution kernels, wherein the number of the convolution kernels is 128, and the sliding step length is 1; 3) firstly, using 1 × 1 convolution kernel, wherein the number of the convolution kernels is 16, then using 5 × 5 convolution kernels, wherein the number of the convolution kernels is 32, and the sliding step length is 1; 4) firstly, performing maximum pooling by using 3 × 3 kernels, then using 1 × 1 convolution kernels, wherein the number of the convolution kernels is 32, and the sliding step length is 1; 5) firstly, carrying out average pooling by using 5-by-5 kernels, wherein the sliding step length is 3, then using 1-by-1 convolution kernels, the number of the convolution kernels is 32, the sliding step length is 1, then using two full-connection layers, and then using a Softmax activation function to assist classification; finally, connecting the output results of the first five branches, and continuing the next step;
the fifth inclusion-a module is identical to the third inclusion-a module;
in each convolution layer and module, a ReLU activation function is used after convolution kernels and pooling operation are used;
s34, training a simplified GoogLeNet convolutional neural network model: inputting the images in the training sample data set into a simplified GoogLeNet convolutional neural network model for feature recognition, and acquiring an optimal simplified GoogLeNet convolutional neural network model;
s35, testing a simplified GoogLeNet convolutional neural network model: inputting a test sample data set into a trained simplified GoogLeNet convolutional neural network model, and verifying the accuracy;
s36, multi-classification detection: and (4) outputting the defect images classified in the step (S2) into the simplified GoogLeNet convolutional neural network model after the test, realizing multi-classification detection on the surface defects of the aircraft skin paint spraying through the simplified GoogLeNet convolutional neural network model, and further judging the specific defect types of the defect images classified in the step (S2), such as whether the defect images are bubble defects or particle, garbage, flowing or flanging defects.
The invention adopts a simplified GoogLeNet network based on increment-V4 to realize the classification and identification of different defect types, the whole frame can solve the problem of the intelligent detection or identification of the known type of the defects of the common deep learning network, has stronger tolerance to unknown defects, higher flexibility and adaptability and high accuracy, improves the operation efficiency while meeting the network learning performance, and effectively shortens the algorithm execution time. In the embodiment, the multi-classification test accuracy is 99.7%, the accuracy is high, and the multi-classification result is accurate.
And S4, outputting multi-classification results, and finishing the defect detection and classification of the aircraft skin paint spraying surface.
The method for realizing the paint spraying surface defect detection and classification of the aircraft skin integrates a defect multi-classification framework based on a traditional image defect detection two-classification method and a simplified GoogLeNet convolutional neural network model based on deep learning, the classification result is shown as the attached figure 1, wherein the classification result has five defect types of bubbles, flanging, particles, flowing and garbage, the two-classification and the multi-classification are combined, the arithmetic efficiency and the robustness (Robust) of the algorithm are effectively improved, thereby effectively shortening the execution time of the algorithm, improving the accuracy of the surface defect detection of the aircraft skin, effectively meeting the intelligent detection and classification requirements of the surface defect of the aircraft skin sprayed paint, the method has important significance for reducing the waste of raw materials in the production process of the airplane, improving the protection performance, prolonging the service life of the airplane skin, reducing the later maintenance cost and ensuring the safety and reliability.
Finally, it should be pointed out here that: the above are only some of the preferred embodiments of the present invention, and should not be construed as limiting the scope of the present invention, and the insubstantial modifications and adaptations made by those skilled in the art based on the above description of the present invention are within the scope of the present invention.
Claims (1)
1. A method for realizing the detection and classification of the paint spraying surface defects of an aircraft skin is characterized by comprising the following steps:
s1, collecting an image of the paint spraying surface of the aircraft skin;
s2, performing two-classification defect detection on the image acquired in the step S1 based on the surface smoothness, specifically as follows:
firstly, preprocessing of Gaussian filtering, histogram equalization, Ostu self-adaptive threshold segmentation and corrosion is carried out on the image collected in the step S1; then performing block detection on the preprocessed image by using Blob to find out the pixels with defects; then, carrying out surface smoothness estimation on the pixels with the defects and the pixels of the whole image by using an information entropy function, setting a surface smoothness threshold value to judge whether the image has the defects or not, and dividing the image into a defective image and a non-defective image to realize two-classification defect detection on the image;
s3, multi-classification detection is carried out on the surface defects of the aircraft skin paint spraying on the basis of the simplified GoogLeNet convolution neural network model: the method specifically comprises the following steps:
s31, data set generation: using the collected image of the paint spraying surface of the aircraft skin for manufacturing a data set, and dividing the data set into a training sample data set and a test sample data set; in the manufacturing process of the data set, firstly, a professional worker for surface painting distinguishes the types of defects of the collected images, wherein the types of defects comprise bubbles, particles, garbage, flowing and flanging, and then the distinguished images are respectively divided into a training sample data set and a test sample data set;
s32, constructing a GoogleLeNet convolutional neural network model: the GoogLeNet convolutional neural network model comprises five layers, wherein the five layers are a first layer, a second layer, a third layer, a fourth layer and a fifth layer respectively, the first layer and the second layer are convolutional layers, the third layer is composed of an inclusion-A module and an inclusion-B module, the fourth layer is composed of an inclusion-A, Inception-B, Inception-C, Inception-D module and an inclusion-E module, and the fifth layer is composed of an inclusion-A module and an inclusion-B module;
s33, constructing a simplified GoogLeNet convolutional neural network model: removing the inclusion-B and inclusion-C modules of the GoogLeNet convolutional neural network model in the step S32, and constructing a simplified GoogLeNet convolutional neural network model;
the constructed simplified GoogLeNet convolutional neural network model sequentially comprises the following steps: the module comprises a first coiling layer, a second coiling layer, a third inclusion-A module, a fourth inclusion-D module, a fourth inclusion-E module and a fifth inclusion-A module; the operation of each convolutional layer and module is as follows:
the first convolution layer firstly uses convolution kernels of 7 × 7, the number of the convolution kernels is 64, then 3 × 3 kernels are used for maximum pooling, and the sliding step length is 2;
the second convolution layer firstly uses 3 × 3 convolution kernels, the number of the convolution kernels is 192, the sliding step length is 1, then 3 × 3 kernels are used for maximum pooling, and the sliding step length is 2;
the third inclusion-a module includes four branches, which are specifically: 1) using 1 × 1 convolution kernels, wherein the number of the convolution kernels is 64, and the sliding step length is 1; 2) firstly, using 1 × 1 convolution kernel, wherein the number of the convolution kernels is 96, then using 3 × 3 convolution kernels, wherein the number of the convolution kernels is 128, and the sliding step length is 1; 3) firstly, using 1 × 1 convolution kernel, wherein the number of the convolution kernels is 16, then using 5 × 5 convolution kernels, wherein the number of the convolution kernels is 32, and the sliding step length is 1; 4) firstly, performing maximum pooling by using 3 × 3 kernels, then using 1 × 1 convolution kernels, wherein the number of the convolution kernels is 32, and the sliding step length is 1; finally, connecting the output results of the four branches, and continuing the next step;
the fourth inclusion-a module is identical to the third inclusion-a module;
the fourth inclusion-D module is identical to the third inclusion-a module;
the fourth inclusion-E module includes five branches, which specifically are: 1) using 1 × 1 convolution kernels, wherein the number of the convolution kernels is 64, and the sliding step length is 1; 2) firstly, 1 × 1 convolution kernels are used, the number of the convolution kernels is 96, then 3 × 3 convolution kernels are used, the number of the convolution kernels is 128, and the sliding step length is 1; 3) firstly, using 1 × 1 convolution kernel, wherein the number of the convolution kernels is 16, then using 5 × 5 convolution kernels, wherein the number of the convolution kernels is 32, and the sliding step length is 1; 4) firstly, performing maximum pooling by using 3 × 3 kernels, then using 1 × 1 convolution kernels, wherein the number of the convolution kernels is 32, and the sliding step length is 1; 5) firstly, carrying out average pooling by using 5-by-5 kernels, wherein the sliding step length is 3, then using 1-by-1 convolution kernels, the number of the convolution kernels is 32, the sliding step length is 1, then using two full-connection layers, and then using a Softmax activation function to assist classification; finally, connecting the output results of the first five branches, and continuing the next step;
the fifth inclusion-a module is identical to the third inclusion-a module;
in each convolution layer and module, a ReLU activation function is used after convolution kernels and pooling operation are used;
s34, training a simplified GoogLeNet convolutional neural network model: inputting the images in the training sample data set into a simplified GoogLeNet convolutional neural network model for feature recognition, and acquiring the simplified GoogLeNet convolutional neural network model;
s35, testing a simplified GoogLeNet convolutional neural network model: inputting a test sample data set into a trained simplified GoogLeNet convolutional neural network model, and verifying the accuracy;
s36, multi-classification detection: inputting the defect images classified in the step S2 into the simplified GoogLeNet convolutional neural network model after testing, and realizing multi-classification detection on the paint spraying surface defects of the aircraft skin through the simplified GoogLeNet convolutional neural network model;
and S4, outputting multi-classification results, and finishing the defect detection and classification of the aircraft skin paint spraying surface.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011626199.0A CN112669292B (en) | 2020-12-31 | 2020-12-31 | Method for detecting and classifying defects on painted surface of aircraft skin |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011626199.0A CN112669292B (en) | 2020-12-31 | 2020-12-31 | Method for detecting and classifying defects on painted surface of aircraft skin |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112669292A CN112669292A (en) | 2021-04-16 |
| CN112669292B true CN112669292B (en) | 2022-09-30 |
Family
ID=75412543
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011626199.0A Active CN112669292B (en) | 2020-12-31 | 2020-12-31 | Method for detecting and classifying defects on painted surface of aircraft skin |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112669292B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106570516A (en) * | 2016-09-06 | 2017-04-19 | 国网重庆市电力公司电力科学研究院 | Obstacle recognition method using convolution neural network |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101719273A (en) * | 2009-10-21 | 2010-06-02 | 苏州有色金属研究院有限公司 | On-line self-adaptation extraction method of metallurgy strip surface defect based on one-dimension information entropy |
| CN108682003B (en) * | 2018-04-04 | 2021-10-08 | 睿视智觉（厦门）科技有限公司 | Product quality detection method |
| CN109886925A (en) * | 2019-01-19 | 2019-06-14 | 天津大学 | A kind of aluminium material surface defect inspection method that Active Learning is combined with deep learning |
| CN111582257A (en) * | 2019-02-15 | 2020-08-25 | 波音公司 | Method, device and system for detecting object to be detected |
| CN111415325B (en) * | 2019-11-11 | 2023-04-25 | 杭州电子科技大学 | Copper foil substrate defect detection method based on convolutional neural network |
| CN111340754B (en) * | 2020-01-18 | 2023-08-25 | 中国人民解放军国防科技大学 | Method for detecting and classifying surface defects of aircraft skin |
| CN111798419A (en) * | 2020-06-27 | 2020-10-20 | 上海工程技术大学 | Metal paint spraying surface defect detection method |
-
2020
- 2020-12-31 CN CN202011626199.0A patent/CN112669292B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106570516A (en) * | 2016-09-06 | 2017-04-19 | 国网重庆市电力公司电力科学研究院 | Obstacle recognition method using convolution neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112669292A (en) | 2021-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109239102B (en) | CNN-based flexible circuit board appearance defect detection method | |
| CN110992317B (en) | PCB defect detection method based on semantic segmentation | |
| CN108074231B (en) | Magnetic sheet surface defect detection method based on convolutional neural network | |
| CN111798419A (en) | Metal paint spraying surface defect detection method | |
| CN111667455A (en) | AI detection method for various defects of brush | |
| CN107607554A (en) | A kind of Defect Detection and sorting technique of the zinc-plated stamping parts based on full convolutional neural networks | |
| CN107657603A (en) | A kind of industrial appearance detecting method based on intelligent vision | |
| CN110308151A (en) | A kind of surface of workpiece defect identification method and device based on machine vision | |
| CN109657718B (en) | Data-driven SPI defect type intelligent identification method on SMT production line | |
| CN111852792B (en) | Fan blade defect self-diagnosis positioning method based on machine vision | |
| CN111815555A (en) | Metal additive manufacturing image detection method and device combining anti-neural network with local binary | |
| CN112529884B (en) | Welding spot quality evaluation method based on indentation characteristic image recognition | |
| CN111783544B (en) | Method for building diamond milling grinding head state monitoring system for processing ceramic mobile phone backboard | |
| CN112907561A (en) | Notebook appearance flaw detection method based on deep learning | |
| CN110349125A (en) | A kind of LED chip open defect detection method and system based on machine vision | |
| CN115018790A (en) | Workpiece surface defect detection method based on anomaly detection | |
| CN113588656A (en) | Novel bolt appearance size detection method | |
| CN112669292B (en) | Method for detecting and classifying defects on painted surface of aircraft skin | |
| CN115423741A (en) | Structural defect detection method based on deep learning | |
| CN115866502A (en) | Microphone part surface defect online detection process | |
| CN114841957A (en) | Steel plate surface defect detection method based on deep learning | |
| CN111028207A (en) | Button flaw detection method based on brain-like immediate-universal feature extraction network | |
| CN114049543A (en) | Automatic identification method for scrap steel unloading change area based on deep learning | |
| Zhu et al. | A surface fatal defect detection method for magnetic tiles based on semantic segmentation and object detection: IEEE ITAIC (ISSN: 2693-2865) | |
| CN108765391A (en) | A kind of plate glass foreign matter image analysis methods 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |