A foreign body detection method for carbon fiber composite materials based on feature matching

Abstract

The present invention relates to the field of foreign body detection technology, and in particular to a foreign body detection method for carbon fiber composite materials based on feature matching, comprising the following steps: collecting and saving images of materials to be detected; dividing the image collected in step S1 into n image blocks; extracting feature points from the n equally divided image blocks; describing the feature points extracted in S3; after extracting and describing the feature points, pairing each image block with the remaining n-1 image blocks to determine whether there is a foreign body. This detection method can effectively solve the problem of low efficiency and inaccuracy in foreign body detection.

Description

Technical Field

[0001] The present invention relates to the technical field of foreign body detection, and in particular to a foreign body detection method of carbon fiber composite materials based on feature matching. Background Art

[0002] In the manufacturing process of resin-based carbon fiber composite materials, there is a risk of foreign matter inclusion in each layer of material. The presence of foreign matter will have a serious impact on the quality of the composite product, causing the product performance to fail to meet the design requirements and resulting in parts being scrapped.

[0003] In actual production, due to the large area of ​​the product and the color of the material, it is obviously inefficient and impractical to judge whether the material is attached with foreign matter by human eyes. In addition, since the types, shapes and sizes of unknown foreign matter are diverse and unknown, it is difficult for computers to learn all the types of foreign matter and classify them effectively, resulting in a gap in how to efficiently detect unknown foreign matter that may exist on the material. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a foreign body detection method for carbon fiber composite materials based on feature matching, which can effectively solve the problem of low efficiency and inaccuracy in foreign body detection, and is particularly suitable for detecting unknown foreign bodies.

[0005] The present invention is achieved by adopting the following technical solutions:

[0006] A foreign body detection method for carbon fiber composite materials based on feature matching, characterized in that it comprises the following steps:

[0007] Step S1. Collect and save the image of the material to be tested;

[0008] Step S2. Divide the image collected in step S1 into n image blocks;

[0009] Step S3. Extract feature points from the n equally divided image blocks respectively;

[0010] Step S4. Perform feature description on the feature points extracted in S3;

[0011] Step S5. After extracting and describing the feature points, feature pair each image block with the remaining n-1 image blocks to determine whether there is a foreign object.

[0012] The step S4 specifically refers to: describing the information of the extracted feature points using a set of feature vectors.

[0013] The step S5 specifically refers to: after extracting and describing the feature points, each image block is feature-paired with the remaining n-1 image blocks, the distance of the feature vectors is calculated, and whether there is a foreign body is determined based on whether the distance of the feature vectors exceeds a set threshold.

[0014] The specific calculation method of the distance of the feature vector is:

[0015]

[0016] Among them, R i is the key point descriptor in the template image patch, R i =(r i1 ,r i2 ,...,r i128 );S i is the key point descriptor in the image block to be matched with the template, S i =(s i1 ,s i2 ,...,s i128 ).

[0017] The step S5 of judging whether there is foreign matter specifically includes the following steps: performing feature pairing on the xth image block and the remaining nx image blocks, and calculating the distance of the feature vectors; if the distances between the feature vectors of the xth image block and the nx image blocks are both less than the set threshold, judging that there is no foreign matter in the remaining image blocks, and ending directly; if the distances between the feature vectors of the xth image block and some of the image blocks in the nx image blocks are less than the threshold, and the distances to the feature vectors of some of the image blocks are greater than the threshold, marking the image blocks whose distances to the feature vectors of the xth image block are greater than the threshold as image blocks containing foreign matter, and ending; if the distances between the feature vectors of the xth image block and the nx image blocks are both greater than the set threshold, judging that there is foreign matter in the xth image block, and if x+1<n, continuing to judge whether there is foreign matter in the x+1th image block.

[0018] The step S3 specifically comprises the following steps:

[0019] Step S 31 .Apply Gaussian blur to image blocks at different scales;

[0020] Step S 32 .Downsample the Gaussian blurred image block to obtain a Gaussian pyramid image;

[0021] Step S 33 . Generate DOG Gaussian difference pyramid image using Gaussian difference kernels of different scales and convolution;

[0022] Step S 34. Perform local extrema detection on the DOG Gaussian difference pyramid image to obtain feature points.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] 1. The present invention can efficiently detect foreign matter, save manual detection time to a certain extent, and improve the production efficiency of the stacking process, thereby avoiding the negative impact of foreign matter on subsequent use.

[0025] 2. The present invention adopts a feature matching method to extract, describe and pair features of the equally divided image blocks, and judges whether there is a foreign body based on whether the distance of the feature vector exceeds a set threshold. Therefore, it has good adaptability and accuracy to the type and size of unknown foreign bodies, and to a certain extent fills the gap in the current field for unknown foreign body detection, and has a wide range of applications.

[0026] 3. The generation of DOG Gaussian difference pyramid image can effectively extract stable key points. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, wherein:

[0028] Figure 1 It is a schematic diagram of the process of the present invention;

[0029] Figure 2 It is a schematic diagram of the foreign body judgment process in the present invention. DETAILED DESCRIPTION

[0030] Example 1

[0031] As a basic embodiment of the present invention, the present invention includes a foreign body detection method for carbon fiber composite materials based on feature matching, comprising the following steps:

[0032] Step S1. Collect and save images of the material to be tested.

[0033] Step S2: Divide the image acquired in step S1 into n equal image blocks.

[0034] Step S3: Extract feature points from the n equally divided image blocks respectively.

[0035] Step S4: Describe the feature points extracted in S3.

[0036] Step S5. After extracting and describing the feature points, feature pair each image block with the remaining n-1 image blocks to determine whether there is a foreign object.

[0037] Since the image blocks are from the same piece of aviation resin-based carbon fiber composite material, the feature differences between the image blocks without foreign matter are small, while the image blocks containing foreign matter will have a large difference from other image blocks when matching features (even two image blocks containing foreign matter will have a large difference due to the different properties and distribution of the foreign matter). This embodiment divides the same material into several equal blocks, and compares the blocks to effectively identify foreign matter.

[0038] This embodiment is not limited to the matching method of image features. Since each feature point has a descriptor that uniquely identifies its identity and characteristics, conventional technical methods in the field can be used, such as calculating the Hamming distance between two descriptors for matching, or using the RANSAC (random sampling consensus) algorithm, etc.

[0039] Example 2

[0040] As a preferred embodiment of the present invention, the present invention includes a foreign body detection method for carbon fiber composite materials based on feature matching, comprising the following steps:

[0041] Step S1. Collect and save images of the material to be tested.

[0042] Step S2: Divide the image acquired in step S1 into n equal image blocks.

[0043] Step S3: Extract feature points from the n equally divided image blocks. Specifically, it includes:

[0044] Step S 31 .Apply Gaussian blur to image blocks at different scales;

[0045] Step S 32 .Downsample the Gaussian blurred image block to obtain a Gaussian pyramid image;

[0046] Step S 33 . Generate DOG Gaussian difference pyramid image using Gaussian difference kernels of different scales and convolution;

[0047] Step S 34 . Perform local extrema detection on the DOG Gaussian difference pyramid image to obtain feature points.

[0048] Step S4: Describe the feature points extracted in S3.

[0049] Step S5. After extracting and describing the feature points, feature pair each image block with the remaining n-1 image blocks to determine whether there is a foreign object.

[0050] Example 3

[0051] As the best embodiment of the present invention, the present invention includes a foreign body detection method for carbon fiber composite materials based on feature matching, wherein the feature matching is a low-level visual processing technology that directly extracts and matches features of the image itself. Feature matching uses the features extracted from the image as conjugate entities, and the proposed feature attributes or description parameters as matching entities, and calculates the similarity measure between the matching entities to achieve an image matching method for conjugate entity registration.

[0052] This embodiment uses a feature matching method to extract, describe and match features of the equally divided image patches, and determines whether there is a foreign object based on whether the distance of the feature vector exceeds a set threshold. Therefore, this embodiment is also suitable for unknown foreign objects. Figure 1 , the detection method specifically comprises the following steps:

[0053] Step S1. Collect and save the image of the material to be detected. In this embodiment, the image can be collected from aviation resin-based carbon fiber composite materials.

[0054] Step S2: divide the image collected in step S1 into n image blocks. Where n is an input variable, and a number can be inputted from the keyboard as the value of n to divide the image into n blocks. The value of n can be adjusted according to the size of the image to be tested.

[0055] Step S3. Extract feature points from the n equally divided image blocks respectively, so that each image block can be identified and distinguished by the extracted features, specifically including the following steps:

[0056] Step S 31 .Apply Gaussian blur to image blocks at different scales.

[0057] Step S 32 .Downsample the image block after Gaussian blur to obtain a Gaussian pyramid image. Specifically, the original image is continuously downsampled to obtain a series of images of different sizes, forming a tower model from large to small and from bottom to top.

[0058] Step S 33 .In order to effectively extract stable key points, Gaussian difference kernels of different scales and convolution are used to generate DOG Gaussian difference pyramid images.

[0059] Step S 34 . Perform local extrema detection on the DOG Gaussian difference pyramid image to obtain feature points. Feature points are composed of local extrema in the DOG space. In order to find the extrema of the DOG function, each pixel is compared with all its neighboring points to see whether it is larger or smaller than its neighboring points in the image domain and scale domain.

[0060] Step S4. Perform feature description on the feature points extracted in S3. Specifically, a set of feature vectors may be used to describe the information of the extracted feature points, so as to facilitate the distance calculation of the feature vectors when performing feature pairing in step S5.

[0061] Step S5. After extracting and describing the feature points, each image block is paired with the remaining n-1 image blocks, the distance of the feature vector is calculated, and whether there is a foreign object is determined based on whether the distance of the feature vector exceeds a set threshold.

[0062] The distance of feature vectors is a method for judging differences in image processing. The smaller the distance, the smaller the difference. Images from the same piece of material should have small differences in the absence of foreign matter. When there are unknown foreign matter on the material, it means that the difference becomes larger and the feature vector distance increases. This method uses the Euclidean distance method to calculate the distance of feature vectors. The Euclidean distance in two-dimensional, three-dimensional or multi-dimensional space is the straight-line distance between two points. In n-dimensional space, it is the actual distance between two points. The Euclidean distance is used to calculate the similarity of vectors. The smaller the Euclidean distance, the greater the similarity.

[0063] The specific calculation method of the distance of the feature vector is:

[0064]

[0065] Among them, R i is the key point descriptor in the template image patch, R i =(r i1 ,r i2 ,...,r i128 );S i is the key point descriptor in the image block to be matched with the template, S i =(s i1 ,s i2 ,...,s i128 );d(R i ,S i ) is to calculate the Euclidean distance of these key point descriptors, that is, the actual distance between the points. If the calculated d(R i ,S i ) is less than the set threshold (Threshold Value), it proves that the feature difference between the two image blocks is small.

[0066] Since the image blocks are from the same piece of aviation resin-based carbon fiber composite material, the feature differences between image blocks that do not contain foreign matter are small, that is, the feature vector distance is small, and the image blocks containing foreign matter will have a larger feature vector distance with other image blocks during feature matching, that is, they have a greater difference (even two image blocks containing foreign matter will have a large difference due to the different properties and distribution of foreign matter). Under normal circumstances, the proportion of image blocks containing foreign matter is very small, because materials with a large number of foreign matter can be easily identified by the naked eye without further inspection, so it can be considered that the image blocks with feature vector distances greater than the set threshold from the majority of other image blocks contain foreign matter.

[0067] For more details, please refer to the attached manual. Figure 2 , the xth (initially 1) image block and the remaining nx image blocks can be feature paired to calculate the distance of the feature vectors; if the distances between the feature vectors of the xth image block and the nx image blocks are both less than the set threshold, the difference is small, and it is judged that there is no foreign matter in the remaining image blocks, and the process ends directly; if the distance between the feature vectors of the xth image block and some of the image blocks in the nx image blocks is less than the threshold, and the distance to the feature vectors of another part of the image blocks is greater than the threshold, the image blocks whose distances to the feature vectors of the xth image block are greater than the threshold are marked as image blocks containing foreign matter and the process ends; if the distances between the feature vectors of the xth image block and the nx image blocks are both greater than the set threshold, the difference is large, and it is judged that the xth image block contains foreign matter. If x+1<n, it continues to judge whether the x+1th image block contains foreign matter until all the image blocks are judged.

[0068] In summary, after reading the present invention document, ordinary technicians in this field can make various other corresponding transformation schemes based on the technical scheme and technical concept of the present invention without creative mental labor, which all fall within the scope of protection of the present invention.

Claims

1. A foreign body detection method for carbon fiber composite materials based on feature matching, characterized in that: The following steps are involved: Step S1. Collecting and saving images of the material to be tested; Step S2. Divide the image collected in step S1 into n image blocks; Step S3. Extract feature points from the n equally divided image blocks respectively; Step S4. Describing the feature points extracted in S3; Step S5. After extracting and describing the feature points, each image block is paired with the remaining n-1 image blocks, the distance of the feature vectors is calculated, and whether there is a foreign object is determined based on whether the distance of the feature vector exceeds a set threshold; The step S5 of judging whether there is foreign matter specifically includes the following steps: performing feature pairing on the xth image block and the remaining nx image blocks, and calculating the distance of the feature vectors; if the distances between the feature vectors of the xth image block and the nx image blocks are both less than the set threshold, judging that there is no foreign matter in the remaining image blocks, and ending directly; if the distances between the feature vectors of the xth image block and some of the image blocks in the nx image blocks are less than the threshold, and the distances to the feature vectors of some of the image blocks are greater than the threshold, marking the image blocks whose distances to the feature vectors of the xth image block are greater than the threshold as image blocks containing foreign matter, and ending; if the distances between the feature vectors of the xth image block and the nx image blocks are both greater than the set threshold, judging that there is foreign matter in the xth image block, and if x+1<n, continuing to judge whether there is foreign matter in the x+1th image block.

2. The method for detecting foreign matter in carbon fiber composite materials based on feature matching according to claim 1, characterized in that: The step S4 specifically refers to: describing the information of the extracted feature points using a set of feature vectors.

3. A method for detecting foreign matter in carbon fiber composite materials based on feature matching according to claim 1 or 2, characterized in that: The specific calculation method of the distance of the feature vector is: Among them, R i is the key point descriptor in the template image patch, , S i is the key point descriptor in the image block to be matched with the template, .

4. The method for detecting foreign matter in carbon fiber composite materials based on feature matching according to claim 1, characterized in that: The step S3 specifically comprises the following steps: Step S 31 . Apply Gaussian blur of different scales to image blocks; Step S 32 . Downsample the Gaussian blurred image block to obtain a Gaussian pyramid image; Step S 33 . Generate DOG Gaussian difference pyramid image using Gaussian difference kernels of different scales and convolution; Step S 34 . Perform local extrema detection on the DOG Gaussian difference pyramid image to obtain feature points.

Images