Laser paint removal detection method, device, equipment and readable storage medium

Abstract

The present invention provides a detection method, device, equipment and readable storage medium for laser paint removal, including: acquiring acoustic wave data of a sample collected by an acoustic wave collector during laser paint removal, and acquiring a visual image of the sample collected by an image collector during laser paint removal; performing dimensionality reduction processing on the acoustic wave data to generate reduced dimensionality data, and converting the reduced dimensionality data into an acoustic wave two-dimensional image aligned with the visual image through a GAF Gram angle field; calling and introducing a DPN network and a CBAM network optimized by a CBAM attention mechanism to extract features from the acoustic wave two-dimensional image and the visual image, and respectively generate acoustic wave features and visual features; splicing and fusing the acoustic wave features and the visual features to generate a fusion result, and generating a classification result of the sample according to the fusion result. The invention aims to solve the problem of insufficient detection capability of laser paint removal on sample surfaces in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of laser paint removal detection, and in particular to a laser paint removal detection method, device, equipment and readable storage medium. Background Art

[0002] During the laser cleaning process, both plasma acoustic waves and thermoelastic acoustic waves will cause the photoacoustic signal to change. The acoustic signal is very sensitive to the critical state changes of the cleaning sample and is not affected by subjective factors. The change in the intensity of the acoustic signal can accurately determine the cleanliness of the local area of ​​the sample and the state changes of the confusion area. Therefore, the state changes of the local area of ​​the sample during laser cleaning can be accurately determined by detecting the changes in the acoustic signal during laser cleaning;

[0003] The change in the acoustic signal is only obvious in a very small area of ​​the sample. When the overall state of the sample is detected, a large number of acoustic signals will significantly increase the difficulty of judging the overall sample. It has no advantage in judging the cleaning state of the entire sample and is difficult to meet complex detection tasks. In the process of single visual image detection and cleaning, due to the different thickness of the paint layer, materials, and possible heat accumulation effects in different areas, the contours of the areas change gradually during the laser cleaning process. The color, texture, and reflectivity of the area only change slightly. However, under microscopic detection, the morphology has changed significantly, that is, the sample state category has changed. This change brings about a transition interval and confuses the boundary contour, which is easily confusing for human eyes. Figure 1 As shown, by detecting the visual image of the sample, the overall sample state can be judged, but the perception of the microscopic changes of the confusing samples and critical states is not strong.

[0004] In view of this, this application is filed. Summary of the invention

[0005] The invention discloses a laser paint removal detection method, device, equipment and readable storage medium, aiming to solve the problem of insufficient detection capability of laser paint removal on sample surface in the prior art.

[0006] The first embodiment of the present invention discloses a method for detecting laser paint removal, comprising:

[0007] Acquiring acoustic wave data of the sample during the laser paint removal process collected by the acoustic wave collector, and visual images of the sample during the laser paint removal process collected by the image collector;

[0008] Performing dimensionality reduction processing on the acoustic wave data to generate reduced-dimensionality data, and converting the reduced-dimensionality data into an acoustic wave two-dimensional image aligned with the visual image through a GAF Gram angle field;

[0009] The DPN network and CBAM network optimized by the CBAM attention mechanism are called to extract features from the acoustic two-dimensional image and visual image respectively, and generate acoustic wave features and visual features respectively;

[0010] The acoustic wave features and the visual features are spliced ​​and fused to generate a fusion result, and a classification result of the sample is generated according to the fusion result.

[0011] Preferably, the dimensionality reduction processing is performed on the acoustic wave data to generate reduced-dimensionality data, and the reduced-dimensionality data is converted into an acoustic wave two-dimensional image aligned with the visual image through a GAF Gram angle field, specifically:

[0012] Performing segmented aggregation approximation on the acoustic wave data to generate dimension-reduced data;

[0013] The time series of the dimension-reduced data in the Cartesian coordinate system is normalized and scaled to [-1, 1], and the scaled time series is converted to the polar coordinate system;

[0014] Calling the arccosine function to convert the scaled time series into an angle in a polar coordinate system, and generating a Gram matrix according to the angle;

[0015] Each segment of sound wave data is converted according to the Gram matrix to generate a two-dimensional image of the sound wave aligned with the visual image.

[0016] Preferably, the DPN network optimized by the CBAM attention mechanism is called to extract features from the two-dimensional sound wave image to generate sound wave features, specifically:

[0017] The acoustic wave two-dimensional image is subjected to large-scale convolution and pooling and then input into four Dense Blocks to extract features of the acoustic wave two-dimensional image, wherein the Dense Blocks are connected by a Transition layer, each Dense Block includes multiple Bottlenecks, and each Bottleneck is embedded with a CBAM attention mechanism;

[0018] The features of the acoustic wave two-dimensional image are visualized to generate an acoustic wave signature.

[0019] Preferably, a CBAM network optimized by the CBAM attention mechanism is called to extract features from the visual image and generate visual features, specifically:

[0020] The visual image is subjected to large-scale convolution and pooling and then input into four blocks to extract features of the visual image, wherein each of the blocks includes a plurality of bottle necks for connecting with the CBAM attention mechanism, and each of the blocks has a transition layer;

[0021] Features of the visual image are visualized to generate visual features.

[0022] Preferably, the acoustic wave features and visual features are spliced ​​and fused to generate a fusion result, and a classification result of the sample is generated according to the fusion result, specifically:

[0023] Performing feature splicing and pooling on the acoustic wave features and visual features to generate a fusion result;

[0024] The fusion result is input into the fully connected layer and the Softmax function for classification prediction to generate the classification result of the sample.

[0025] A second embodiment of the present invention provides a laser paint removal detection device, comprising:

[0026] A data acquisition unit, used to acquire the acoustic wave data of the sample during the laser paint removal process collected by the acoustic wave collector, and the visual image of the sample during the laser paint removal process collected by the image collector;

[0027] A data dimension reduction unit, used for performing dimension reduction processing on the acoustic wave data to generate dimension reduction data, and converting the dimension reduction data into an acoustic wave two-dimensional image aligned with the visual image through a GAF Gram angle field;

[0028] The feature generation unit calls and introduces the DPN network and CBAM network optimized by the CBAM attention mechanism to extract features from the acoustic two-dimensional image and the visual image, and generates acoustic wave features and visual features respectively;

[0029] The classification result generating unit is used to concatenate and fuse the acoustic wave features and the visual features to generate a fusion result, and to generate a classification result of the sample according to the fusion result.

[0030] Preferably, the data dimension reduction unit is specifically used for:

[0031] Performing segmented aggregation approximation on the acoustic wave data to generate dimension-reduced data;

[0032] The time series of the dimension reduction data in the Cartesian coordinate system is normalized and scaled to [-1, 1], and the scaled time series is converted to a polar coordinate system;

[0033] Calling the arccosine function to convert the scaled time series into an angle in a polar coordinate system, and generating a Gram matrix according to the angle;

[0034] Each segment of sound wave data is converted according to the Gram matrix to generate a two-dimensional image of the sound wave aligned with the visual image.

[0035] Preferably, the classification result generating unit is specifically used for:

[0036] Performing feature splicing and pooling on the acoustic wave features and visual features to generate a fusion result;

[0037] The fusion result is input into the fully connected layer and the Softmax function for classification prediction to generate the classification result of the sample.

[0038] A third embodiment of the present invention provides a laser paint removal detection device, including a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement a laser paint removal detection method as described in any one of the above.

[0039] A fourth embodiment of the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program can be executed by a processor of a device where the computer-readable storage medium is located to implement a laser paint removal detection method as described in any one of the above.

[0040] A laser paint removal detection method, device, equipment and readable storage medium provided by the present invention collects the acoustic wave data and visual images of the sample during the laser paint removal process through an acoustic wave collector and an image collector, then performs dimensionality reduction processing on the acoustic wave data to generate reduced dimensionality data, and converts the reduced dimensionality data into an acoustic wave two-dimensional image aligned with the visual image through a GAF Gram angle field, calls and introduces a DPN network optimized by a CBAM attention mechanism to extract features from the acoustic wave two-dimensional image, calls and introduces a CBAM network optimized by a CBAM attention mechanism to extract features from the visual image, and generates acoustic wave features and visual features respectively; splices and fuses the acoustic wave features and the visual features to generate a fusion result, and generates a classification result of the sample according to the fusion result, thereby solving the problem of insufficient detection capability of the prior art during laser paint removal on the sample surface. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is the confusing sample image in the background technology;

[0042] Figure 2 It is a schematic flow chart of a laser paint removal detection method provided by the first embodiment of the present invention;

[0043] Figure 3 It is a multimodal data processing and alignment structure diagram provided by the present invention;

[0044] Figure 4 It is a schematic diagram of the distribution of data points after dimensionality reduction provided by the present invention;

[0045] Figure 5 It is a schematic diagram of three states corresponding to the sample provided by the present invention;

[0046] Figure 6 It is a structural diagram of a single Block embedded in CBAM provided by the present invention;

[0047] Figure 7 It is a CBAM-DPN feature extraction flow chart provided by the present invention;

[0048] Figure 8 It is a schematic diagram of visual comparison of visual picture features provided by the present invention;

[0049] Fig. 9 It is a structural diagram of a single DenseBlock embedded in CBAM provided by the present invention;

[0050] Fig.10 It is a CBAM-DenseNet121 network model diagram provided by the present invention;

[0051] Fig.11 It is a schematic diagram of visual comparison of visual picture features provided by the present invention;

[0052] Fig.12 It is a schematic diagram of the feature fusion structure provided by the present invention;

[0053] Fig.13 It is a schematic diagram of a module of a laser paint removal detection device provided in the second embodiment of the present invention. DETAILED DESCRIPTION

[0054] 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.

[0055] In order to better understand the technical solution of the present invention, the embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0056] It should be clear that the described embodiments are only some embodiments of the present invention, not all 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.

[0057] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0058] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0059] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0060] The "first\second" mentioned in the embodiments is only to distinguish similar objects, and does not represent a specific order for the objects. It is understandable that the "first\second" can be interchanged with the specific order or sequence where permitted. It should be understood that the objects distinguished by "first\second" can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0061] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0062] The invention discloses a laser paint removal detection method, device, equipment and readable storage medium, aiming to solve the problem of insufficient detection capability of laser paint removal on sample surface in the prior art.

[0063] See also Figure 2 The first embodiment of the present invention discloses a laser paint removal detection method, which can be performed by a laser paint removal detection device (hereinafter referred to as the detection device), and in particular, is performed by one or more processors in the detection device to at least implement the following steps:

[0064] S101, acquiring acoustic wave data of the sample during the laser paint removal process collected by the acoustic wave collector, and visual images of the sample during the laser paint removal process collected by the image collector;

[0065] In this embodiment, the detection device can be a terminal with data processing and analysis capabilities such as a desktop computer, a laptop computer, a server, a workstation, etc., wherein the corresponding operating system and application software can be installed in the detection device, and the functions required by this embodiment can be realized through the combination of the operating system and the application software.

[0066] In this embodiment, the sound wave collector can be a microphone, and the image collector can be a visual camera. Specifically, in this embodiment, the sampling frequency of the sound wave collector can be 48000 Hz, each segment of sound wave data collected is a one-dimensional data point, and the collected visual image is two-dimensional data.

[0067] S102, performing dimensionality reduction processing on the acoustic wave data to generate reduced-dimensionality data, and converting the reduced-dimensionality data into an acoustic wave two-dimensional image aligned with the visual image through a GAF Gram angle field;

[0068] It should be noted that since the data dimensions of the sound wave data and the visual image are not interconnected, in this embodiment, by introducing PAA dimensionality reduction, the amount of data is effectively reduced without destroying the characteristics of the sound wave data. The data after dimensionality reduction is converted into a two-dimensional sound wave image by GAF conversion, and the temporal nature of the information is retained, which can be aligned with the visual image one by one to extract the structure such as Figure 3 shown.

[0069] Specifically, in this embodiment:

[0070] Performing segmented aggregation approximation on the acoustic wave data to generate dimension-reduced data;

[0071] The time series of the dimension reduction data in the Cartesian coordinate system is normalized and scaled to [-1, 1], and the scaled time series is converted to a polar coordinate system;

[0072] Calling the arccosine function to convert the scaled time series into an angle in a polar coordinate system, and generating a Gram matrix according to the angle;

[0073] Each segment of sound wave data is converted according to the Gram matrix to generate a two-dimensional image of the sound wave aligned with the visual image.

[0074] More specifically: First, each segment of the sound wave data is segmented and aggregated for approximation. Segmented aggregation approximation is an effective data dimensionality reduction method that plays a great role in reducing the amount of data and improving computational efficiency. The principle is to assume that the sound wave data is X = {x1, x2, ..., x n} has a length of n, and its length is reduced to d dimensions, expressed as Its expression is shown in (1):

[0075]

[0076] like Figure 4 The original sound wave data shown in point 1 is after PAA dimension reduction, which reduces the time complexity and replaces each segment with its average value, smoothing the noise to varying degrees. While controlling the number of dimensions reduced, the characteristics of the data can be retained and the amount of data can be reduced.

[0077] The reduced-dimensional sound wave data is converted into a two-dimensional image through the GAF operation so that deep learning can be used for feature extraction. GAF (Gramian Angular Field) can convert one-dimensional time series data into two-dimensional images, and scale the reduced-dimensional one-dimensional sound wave data to [-1,1] in the Cartesian coordinate system sequence through normalization.

[0078] The scaled time series is converted to the polar coordinate system, and the scaled time series is converted to the angle in polar coordinates by the arc cosine function. Sequence timestamp t i / N is converted to radius r, where the angle value and radius variables are as follows:

[0079]

[0080] After the sequence conversion, a series of angles and radii are obtained, and then a similar inner product operation in the Gram angular field is defined, as shown in formula (3).

[0081]

[0082] That and Respectively represent the angles of x1 and x2 in polar coordinates, which are used to better explain the connection information from two perspectives, resulting in a Gram matrix similar to the following formula (4):

[0083]

[0084] like Figure 5 As shown, Figure 5 (a) is a two-dimensional image obtained after converting each segment of one-dimensional sound wave data. Figure 5 (b) is a visual image that corresponds one-to-one with the sound wave picture.

[0085] S103, calling and introducing the DPN network and the CBAM network optimized by the CBAM attention mechanism to extract features of the acoustic wave two-dimensional image and the visual image respectively, and generating acoustic wave features and visual features respectively;

[0086] It should be noted that although the deep network can extract deeper feature images, it will also bring more calculations, and the shallow network will often face the problem of overfitting. The model is not stable enough and affects the accuracy of fault diagnosis. This embodiment improves the DPN92 and DenseNet121 networks, and adds the CBAM attention mechanism on the original basis, so that the model can obtain richer feature maps, extract more effective features, and reduce redundant calculations.

[0087] In this embodiment, the CBAM network optimized by the CBAM attention mechanism is called to extract features of the visual image and generate visual features, specifically:

[0088] The visual image is subjected to large-scale convolution and pooling and then input into four blocks to extract features of the visual image, wherein each of the blocks includes a plurality of bottle necks for connecting with the CBAM attention mechanism, and each of the blocks has a transition layer;

[0089] Features of the visual image are visualized to generate visual features.

[0090] More specifically:

[0091] Visual image features are complex and difficult to extract. In this embodiment, the CBAM attention mechanism is embedded in the DPN92 network to form a CBAM-DPN92 network for visual image feature extraction. Figure 6 As shown in the figure, after embedding CBAM into each convolutional block, the ability to extract visual image features is further improved.

[0092] The structure of CBAM-DPN92 is optimized to embed CBAM into each Bottleneck convolution block, and after obtaining the shared weights, it enhances the useful features in the feature map according to the different importance of channels and spaces, suppresses irrelevant features, and enhances the feature extraction capability. Figure 6As shown in the figure, the feature matrix is ​​passed through the Bottleneck convolution block to obtain a new feature mapping matrix, and then enters the channel attention mapping calculation. First, the input feature map size is w×h×c, and average pooling and maximum pooling are performed to obtain a result of 1×1×c, and then two layers of fully connected layers are connected. FC1 is 1×1×c / r, and the parameter r is to reduce the number of channels and reduce the amount of calculation. In this model, r is 16. It is increased to 1×1×c by FC2, and the average pooling is the same. After the Sigmoid function, it is multiplied with the original input, which acts as a different weight for each channel. In the spatial attention mechanism, global channel pooling is first performed to generate two w×h×1 feature maps, and the two feature maps are merged by channels. Then, a 7×7 convolution is performed to restore the dimension to w×h×1, and finally, the Sigmoid function is passed, and the result is multiplied with the input after channel recalibration. A new feature mapping matrix is ​​obtained by extracting dual-dimensional features of space and channels, thereby effectively highlighting the features.

[0093] The steps of CBAM-DPN92 feature extraction for visual images are as follows: Figure 7 As shown in the figure, the visual image passes through four blocks after 7×7 large-scale convolution and pooling, and each block contains 3, 4, 20, and 3 connections between Bottleneck and CBAM respectively. Each block has a transition layer, which can reduce network parameters and data redundancy. After embedding the CBAM attention mechanism after each Bottleneck to form a new Block, the extracted visual features are visualized. Figure 8 As shown in the figure, after the introduction of the CBAM attention mechanism, the model highlights important features, and the features cover more unclean parts of the sample, improving the efficiency of feature extraction.

[0094] Specifically, in this embodiment, the DPN network optimized by the CBAM attention mechanism is called to extract features of the two-dimensional sound wave image to generate sound wave features, specifically:

[0095] The acoustic wave two-dimensional image is subjected to large-scale convolution and pooling and then input into four Dense Blocks to extract features of the acoustic wave two-dimensional image, wherein the Dense Blocks are connected by a Transition layer, each Dense Block includes multiple Bottlenecks, and each Bottleneck is embedded with a CBAM attention mechanism;

[0096] The features of the acoustic wave two-dimensional image are visualized to generate an acoustic wave signature.

[0097] It should be noted that the feature extraction of acoustic wave images is also relatively complex. It is difficult for shallow feature extraction to accurately grasp the information of acoustic wave images. By embedding the CBAM attention mechanism into DenseNet121, a CBAM-DenseNet121 network is formed. The structural optimization of CBAM-DenseNet is similar to that of CBAM-DPN. After embedding CBAM into each Bottleneck, useful features are enhanced according to the different importance of space and channels to enhance the ability to express features. Its single structural block is as follows: Fig. 9 shown.

[0098] The steps of CBAM-DenseNet to extract features from acoustic wave images are as follows: Fig.10 As shown in the figure, the sound wave image passes through four Dense Blocks after 7×7 large-scale convolution and pooling. Every two Dense Blocks are connected by a Transition layer to compress the model and reduce the size of the feature map. Each Dense Block contains 6, 12, 24, and 16 Bottlenecks respectively. After embedding the CBAM attention mechanism after each Bottleneck, the extracted visual features are visualized. Fig.11 As shown in the figure, the extracted sound wave image features are visualized. After the introduction of the CBAM attention mechanism, the features cover more parts of the sample than before, which is conducive to improving the accuracy of distinguishing samples.

[0099] S104, concatenating and fusing the acoustic wave features and the visual features to generate a fusion result, and generating a classification result of the sample according to the fusion result.

[0100] Specifically:

[0101] Performing feature splicing and pooling on the acoustic wave features and visual features to generate a fusion result;

[0102] The fusion result is input into the fully connected layer and the Softmax function for classification prediction to generate the classification result of the sample.

[0103] More specifically: feature fusion is to output different feature vectors into a new vector through mapping, such as Fig.12 As shown in the figure, this paper extracts the acoustic wave features f from CBAM-DPN92 and CBAM-DenseNet121. i and visual image features f j Perform feature concatenation and pooling, that is, F = concate(f i +f j ), the fused feature F contains the complete feature information of the two data, and is classified and predicted by the fully connected layer and the Softmax function, so that the network can obtain more accurate classification results.

[0104] The above embodiment has at least the following beneficial effects:

[0105] (1) Starting from the dimension of data alignment, this embodiment combines PAA segmented aggregation approximate dimensionality reduction and GAF ​​Gram angular field to convert the acoustic wave data into a two-dimensional image, retaining the data features and reducing the size of the data without destroying the time series. The acoustic wave time series is converted into a relatively sparse two-dimensional image, which reduces the difficulty of extracting features from one-dimensional acoustic wave information and is conducive to the fusion of modalities.

[0106] (2) By introducing the attention mechanism, we construct deep learning networks CBAM-DPN92 and CBAM-DenseNet121 to extract features of the two types of data, improving the feature extraction capability of the data.

[0107] (3) By fusing and splicing the extracted modal features and constructing a fusion network, the accuracy and stability of sample classification during laser paint removal are greatly improved.

[0108] See also Fig.13 The second embodiment of the present invention provides a laser paint removal detection device, comprising:

[0109] The data acquisition unit 201 is used to acquire the acoustic wave data of the sample during the laser paint removal process collected by the acoustic wave collector, and the visual image of the sample during the laser paint removal process collected by the image collector;

[0110] A data dimension reduction unit 202 is used to perform dimension reduction processing on the acoustic wave data to generate dimension reduction data, and convert the dimension reduction data into an acoustic wave two-dimensional image aligned with the visual image through a GAF Gram angle field;

[0111] The feature generation unit 203 calls and introduces the DPN network and the CBAM network optimized by the CBAM attention mechanism to extract features of the acoustic wave two-dimensional image and the visual image, and generates acoustic wave features and visual features respectively;

[0112] The classification result generating unit 204 is used to combine and fuse the acoustic wave features and the visual features to generate a fusion result, and generate a classification result of the sample according to the fusion result.

[0113] Preferably, the data dimension reduction unit is specifically used for:

[0114] Performing segmented aggregation approximation on the acoustic wave data to generate dimension-reduced data;

[0115] The time series of the dimension reduction data in the Cartesian coordinate system is normalized and scaled to [-1, 1], and the scaled time series is converted to a polar coordinate system;

[0116] Calling the arccosine function to convert the scaled time series into an angle in a polar coordinate system, and generating a Gram matrix according to the angle;

[0117] Each segment of sound wave data is converted according to the Gram matrix to generate a two-dimensional image of the sound wave aligned with the visual image.

[0118] Preferably, the classification result generating unit is specifically used for:

[0119] Performing feature splicing and pooling on the acoustic wave features and visual features to generate a fusion result;

[0120] The fusion result is input into the fully connected layer and the Softmax function for classification prediction to generate the classification result of the sample.

[0121] A third embodiment of the present invention provides a laser paint removal detection device, including a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement a laser paint removal detection method as described in any one of the above.

[0122] A fourth embodiment of the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program can be executed by a processor of a device where the computer-readable storage medium is located to implement a laser paint removal detection method as described in any one of the above.

[0123] A laser paint removal detection method, device, equipment and readable storage medium provided by the present invention collects the acoustic wave data and visual images of the sample during the laser paint removal process through an acoustic wave collector and an image collector, then performs dimensionality reduction processing on the acoustic wave data to generate reduced dimensionality data, and converts the reduced dimensionality data into an acoustic wave two-dimensional image aligned with the visual image through a GAF Gram angle field, calls and introduces a DPN network optimized by a CBAM attention mechanism to extract features from the acoustic wave two-dimensional image, calls and introduces a CBAM network optimized by a CBAM attention mechanism to extract features from the visual image, and generates acoustic wave features and visual features respectively; splices and fuses the acoustic wave features and the visual features to generate a fusion result, and generates a classification result of the sample according to the fusion result, thereby solving the problem of insufficient detection capability of the prior art during laser paint removal on the sample surface.

[0124] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for detecting laser paint removal, characterized in that: include: Acquiring acoustic wave data of the sample during the laser paint removal process collected by the acoustic wave collector, and visual images of the sample during the laser paint removal process collected by the image collector; Performing dimensionality reduction processing on the acoustic wave data to generate reduced-dimensionality data, and converting the reduced-dimensionality data into an acoustic wave two-dimensional image aligned with the visual image through a GAF Gram angle field; The DPN network and CBAM network optimized by the CBAM attention mechanism are called and introduced to extract features of the acoustic two-dimensional image and the visual image, and to generate acoustic wave features and visual features respectively; specifically: the acoustic two-dimensional image is subjected to large-scale convolution and pooling and then input into four Dense Blocks to extract features of the acoustic two-dimensional image, wherein the Dense Blocks are connected by a Transition layer, and each Dense Block includes a plurality of Bottlenecks, wherein each Bottleneck is embedded with a CBAM attention mechanism; the features of the acoustic two-dimensional image are visualized to generate acoustic wave features; the visual image is subjected to large-scale convolution and pooling and then input into four Blocks to extract features of the visual image, wherein each Block includes a plurality of Bottlenecks for connecting with the CBAM attention mechanism, and each Block has a Transition layer; the features of the visual image are visualized to generate visual features; The acoustic wave features and the visual features are spliced ​​and fused to generate a fusion result, and a classification result of the sample is generated according to the fusion result.

2. A laser paint removal detection method according to claim 1, characterized in that: The dimensionality reduction processing is performed on the acoustic wave data to generate reduced-dimensional data, and the reduced-dimensional data is converted into an acoustic wave two-dimensional image aligned with the visual image through the GAF Gram angle field, specifically: Performing segmented aggregation approximation on the acoustic wave data to generate dimension-reduced data; The time series of the dimension-reduced data in the Cartesian coordinate system is normalized and scaled to [-1, 1], and the scaled time series is converted to the polar coordinate system; Calling the arccosine function to convert the scaled time series into an angle in a polar coordinate system, and generating a Gram matrix according to the angle; Each segment of sound wave data is converted according to the Gram matrix to generate a two-dimensional image of the sound wave aligned with the visual image.

3. A laser paint removal detection method according to claim 1, characterized in that: The acoustic wave features and visual features are spliced ​​and fused to generate a fusion result, and a classification result of the sample is generated according to the fusion result, specifically: Performing feature splicing and pooling on the acoustic wave features and visual features to generate a fusion result; The fusion result is input into the fully connected layer and the Softmax function for classification prediction to generate the classification result of the sample.

4. A laser paint removal detection device, characterized in that: include: A data acquisition unit, used to acquire the acoustic wave data of the sample during the laser paint removal process collected by the acoustic wave collector, and the visual image of the sample during the laser paint removal process collected by the image collector; A data dimension reduction unit, used for performing dimension reduction processing on the acoustic wave data to generate dimension reduction data, and converting the dimension reduction data into an acoustic wave two-dimensional image aligned with the visual image through a GAF Gram angle field; The feature generation unit calls and introduces the DPN network and the CBAM network optimized by the CBAM attention mechanism to extract features of the acoustic wave two-dimensional image and the visual image respectively, and generates acoustic wave features and visual features respectively; wherein the feature generation unit is specifically used for: performing large-scale convolution and pooling on the acoustic wave two-dimensional image and inputting it into four Dense Blocks to extract features of the acoustic wave two-dimensional image, wherein the Dense Blocks are connected by a Transition layer, and each Dense Block includes a plurality of Bottlenecks, wherein each Bottleneck is embedded with a CBAM attention mechanism; visualizing the features of the acoustic wave two-dimensional image to generate acoustic wave features; performing large-scale convolution and pooling on the visual image and inputting it into four Blocks to extract features of the visual image, wherein each Block includes a plurality of Bottlenecks for connecting with the CBAM attention mechanism, and each Block has a Transition layer; visualizing the features of the visual image to generate visual features; The classification result generating unit is used to concatenate and fuse the acoustic wave features and the visual features to generate a fusion result, and to generate a classification result of the sample according to the fusion result.

5. A laser paint removal detection device according to claim 4, characterized in that: The data dimension reduction unit is specifically used for: Performing segmented aggregation approximation on the acoustic wave data to generate dimension-reduced data; The time series of the dimension-reduced data in the Cartesian coordinate system is normalized and scaled to [-1, 1], and the scaled time series is converted to the polar coordinate system; Calling the arccosine function to convert the scaled time series into an angle in a polar coordinate system, and generating a Gram matrix according to the angle; Each segment of sound wave data is converted according to the Gram matrix to generate a two-dimensional image of the sound wave aligned with the visual image.

6. A laser paint removal detection device according to claim 4, characterized in that: The classification result generating unit is specifically used for: Performing feature splicing and pooling on the acoustic wave features and visual features to generate a fusion result; The fusion result is input into the fully connected layer and the Softmax function for classification prediction to generate the classification result of the sample.

7. A laser paint removal detection device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement a laser paint removal detection method as described in any one of claims 1 to 3.

8. A computer-readable storage medium, characterized in that: A computer program is stored, and the computer program can be executed by a processor of the device where the computer-readable storage medium is located to implement a laser paint removal detection method as described in any one of claims 1 to 3.

Images