An image processing method for the cladding effect on the surface of a substrate

Abstract

The present invention provides a method for processing the image of the substrate surface cladding effect, belonging to the field of image processing. For the input image of the substrate surface cladding effect, it is processed by a frequency domain decomposition and adaptive fusion strategy and a high-order dynamic feature extraction strategy based on phase space evolution and recurrence quantification spectrum. The enhanced image generation module for the substrate surface cladding effect is used to generate and output the enhanced image of the substrate surface cladding effect. This method has high feature extraction efficiency, high clarity of the enhanced image, and rich detail expression, is applicable to the real-time processing ability of complex industrial scenarios, and can provide reliable technical support for the automatic detection and evaluation of the substrate surface cladding quality.

Description

Technical Field

[0001] The present invention belongs to the field of image processing, and particularly relates to an image processing method for the cladding effect on the surface of a substrate. Background Art

[0002] The surface cladding technology of the substrate is a commonly used surface treatment method. A specific material is cladded on the surface of the substrate by using a laser to improve the wear resistance, corrosion resistance and other properties of the workpiece. It has been widely used in the fields of aerospace, automobile manufacturing, metallurgy and machining. However, due to the complex cladding process, which involves multiple links such as high-temperature heating, material melting, and cooling, there are often some defects on the surface of the cladded substrate, such as pores, cracks, uneven coating, cladding thickness deviation, etc. These defects directly affect the quality and service life of the cladding layer, and may even cause the workpiece to fail. Therefore, the detection and quality control of the cladded surface are particularly important.

[0003] At present, the detection of the cladding effect on the surface of the substrate mainly relies on traditional non-destructive testing methods, such as ultrasonic testing, X-ray testing and optical microscope testing. Although these methods can provide a certain degree of quality assessment, the detection efficiency is relatively low and the information obtained is limited. With the rapid development of artificial intelligence and image processing technology, the combination of multi-modal information (such as RGB images and depth images) has shown significant advantages in the field of industrial inspection. Especially in the detection of the cladding effect on the surface of the substrate, the RGB image can provide surface texture information, while the depth image can capture tiny three-dimensional morphological changes. By combining the two, the substrate surface image can be processed more accurately. This method not only solves the problems of data scarcity and low detection efficiency, but also can achieve high-precision defect detection in a complex industrial environment, thus providing strong support for improving the reliability and stability of the substrate cladding process. Summary of the Invention

[0004] The present invention provides an image processing method for the cladding effect on the surface of a substrate, aiming to generate an enhanced image of the cladding effect on the surface of the substrate by fusing RGB images and depth modal features, so as to improve the evaluation accuracy and reliability of the cladding quality on the surface of the substrate, including the following steps.

[0005] S1. Collect the images of the cladding effect on the surface of the substrate containing RGB and depth information, and make a dataset of the images of the cladding effect on the surface of the substrate.

[0006] S2. Construct a frequency-domain decomposition and adaptive fusion strategy, including: decomposing the images of the cladding effect on the surface of the substrate into low-frequency and multi-directional high-frequency sub-bands by using discrete wavelet transform, introducing learnable parameters to adaptively fuse RGB and depth information in each frequency band, and reconstructing the fused feature map of the cladding effect on the surface of the substrate by using inverse wavelet transform.

[0007] S3. Construct a high-order dynamic feature extraction strategy based on phase space evolution and recursive quantization spectrum to extract features from the fused feature map of the substrate surface cladding effect, including: converting the fused feature map of the substrate surface cladding effect into grayscale and expanding it into a one-dimensional sequence, constructing a phase space trajectory using the Takens embedding theory, and then generating corresponding recursive quantization spectrum indicators based on a preset threshold, thereby constructing a global feature vector of the substrate surface cladding effect.

[0008] S4. Construct a module for generating an enhanced image of the substrate surface cladding effect, including: calculating the difference image of the fused feature map of the substrate surface cladding effect, designing an adaptive enhancement intensity adjustment factor, and using the adaptive enhancement intensity adjustment factor and the global feature vector of the substrate surface cladding effect to adjust the enhancement intensity of the local variation amount to generate an enhanced image of the substrate surface cladding effect.

[0009] S5. Construct an image processing model for the enhanced substrate surface cladding effect and train it using the substrate surface cladding effect image dataset.

[0010] Preferably, in S1, an industrial depth camera is used to take pictures of the substrate surface with laser cladding to obtain the RGB image of the substrate surface cladding effect and the depth image of the substrate surface cladding effect , and the obtained RGB image of the substrate surface cladding effect and the depth image of the substrate surface cladding effect are labeled and made into a substrate surface cladding effect image dataset.

[0011] Preferably, in S2, the input images include the RGB image of the substrate surface cladding effect and the depth image of the substrate surface cladding effect , and the specific process includes:

[0012] S21. Perform discrete wavelet transform on the input images respectively. For each channel of , perform discrete wavelet transform. The specific process is as follows:

[0013] ,

[0014] where represents the low-frequency subband of on channel c, , and respectively represent the high-frequency subbands of in the horizontal, vertical, and diagonal directions on channel c, represents discrete wavelet transform, represents the image matrix of

[0015] For Perform a discrete wavelet transform. The specific process is as follows:

[0016] ,

[0017] where represents the low-frequency sub-band of , and respectively represent the high-frequency sub-bands of

[0018] S22. Introduce learnable parameters to adjust the fusion ratio of RGB and depth information in each frequency band. The specific process is as follows:

[0019] ,

[0020] where is the learnable parameter of the low-frequency sub-band, , which is used to control the contribution ratio of the low-frequency information of each channel of the RGB image, is the fused low-frequency sub-band,

[0021] ,

[0022] where is the learnable parameter of the high-frequency sub-band in the horizontal direction, , which is used to control the contribution ratio of each channel of the RGB image in the horizontal high-frequency component, is the fused high-frequency sub-band in the horizontal direction,

[0023] ,

[0024] where is the learnable parameter of the high-frequency sub-band in the vertical direction, , which is used to control the contribution ratio of each channel of the RGB image in the vertical high-frequency component, is the fused high-frequency sub-band in the vertical direction,

[0025] ,

[0026] where is the learnable parameter of the high-frequency sub-band in the diagonal direction, , which is used to control the contribution ratio of each channel of the RGB image in the diagonal high-frequency component, is the fused high-frequency sub-band in the diagonal direction;

[0027] S23. Reconstruct the fused feature map of the substrate surface cladding effect by performing an inverse wavelet transform on the fused frequency bands . The specific process is as follows:

[0028] ,

[0029] wherein represents the inverse wavelet transform.

[0030] Preferably, in the step S2, based on the strategy of frequency domain decomposition and adaptive fusion, the information of the RGB and depth images of the substrate surface cladding effect is processed separately in different frequency bands, and dynamic fusion is adaptively achieved by introducing learnable parameters, which can make more full use of the complementary advantages of the two modalities in global structure and local details. After the DWT decomposition, not only the low-frequency overall information of the image can be separated, but also the high-frequency details can be finely captured, which is crucial for the detection of tiny defects in the substrate surface cladding effect image. At the same time, the RGB and depth modality information is more carefully fused, providing a richer and more accurate feature basis for subsequent image enhancement and defect detection.

[0031] Preferably, in the step S3, the fused feature map of the substrate surface cladding effect is converted into a grayscale image of the substrate surface cladding effect , and the specific process is as follows:

[0032] ,

[0033] wherein , and are the values of the three RGB channels of the fused feature map of the substrate surface cladding effect at respectively;

[0034] The grayscale image of the substrate surface cladding effect is unfolded into a one-dimensional sequence in row order, and the specific process is as follows:

[0035] ,

[0036] where H and W are the height and width of the grayscale image of the substrate surface cladding effect respectively;

[0037] According to the Takens theory, the embedding dimension m and the time delay r are selected to construct the reconstructed phase space vector , and the construction of the phase space vector corresponding to the th moment is specifically as follows:

[0038] ,

[0039] wherein represents matrix transpose, , and respectively represent the , and values in the one-dimensional sequences , represents the dimension of the reconstructed phase space vector ;

[0040] Construct a recurrence matrix using the reconstructed phase space vector , and the specific process is as follows:

[0041] ,

[0042] where represents the Euclidean distance, represents the distance threshold, and respectively represent the -th and -th phase space vectors corresponding to the -th and represents the Heaviside step function,

[0043] ,

[0044] Extract recurrence quantification spectrum indices from the recurrence matrix , including the recurrence quantity , determinism , entropy and the average diagonal length , and the specific process is as follows:

[0045] ,

[0046] where represents the dimension of the recurrence matrix ,

[0047] ,

[0048] where represents the length of the diagonal in the recurrence matrix , represents the number of strips with a diagonal length of , represents the minimum length threshold of the diagonal,

[0049] ,

[0050] where represents the log function,

[0051] ,

[0052] Then, the extracted recursive quantization spectrum indexes are combined to construct a global feature vector of the substrate surface cladding effect. , and the specific process is as follows:

[0053] .

[0054] Preferably, in S3, the high-order dynamic feature extraction strategy based on phase space evolution and recursive quantization spectrum reveals the internal repetitive patterns and complexity of the substrate surface cladding effect image from the perspective of system dynamic evolution, provides an intuitive and physically meaningful feature description. Through phase space reconstruction and the construction of a recursive matrix, it captures both the local detail repeatability of the substrate surface cladding effect image and reflects the dynamic changes in the global structure. This strategy relies on strictly mathematical nonlinear dynamics and recursive quantization analysis methods. The recursive quantization indexes have high robustness to noise and local perturbations, can accurately extract image features in complex backgrounds, and are applicable to the detection of complex substrate surface cladding effect images.

[0055] Preferably, in S4, calculate the difference image of the substrate surface cladding effect fusion feature map , , which reflects the detailed information of the local changes in the substrate surface cladding effect fusion feature map. The specific process is as follows:

[0056] ,

[0057] where , and respectively represent the pixel values at the , and positions on the c-th channel of the substrate surface cladding effect fusion feature map;

[0058] Construct an adaptive enhancement intensity adjustment factor , and the specific process is as follows:

[0059] ,

[0060] where , , and respectively represent the recurrence quantity, determinism, entropy, and average diagonal length extracted from the recursive matrix , represents the adjustment coefficient of the global enhancement intensity, controlling the overall intensity of enhancement, with a value range of [0, 10], represents the adjustment coefficient of local details, controlling the influence of the difference image on the enhancement intensity, with a value range of [0, 2], It ensures that in the area with small local changes, the enhancement effect will not be too strong;

[0061] The enhancement intensity of the local change amount is adjusted by using the adaptive enhancement intensity adjustment factor and the global feature vector of the substrate surface cladding effect to generate the enhanced image of the substrate surface cladding effect. , and the specific process is as follows:

[0062] ,

[0063] where represents the global feature vector of the substrate surface cladding effect.

[0064] Preferably, in S4, by calculating the difference image of the fused feature map of the substrate surface cladding effect and combining it with the global feature vector of the substrate surface cladding effect, the gain of the image is adaptively adjusted, enhancing the key details of the substrate surface, while controlling the enhancement intensity to avoid over-enhancing background noise or unnecessary areas. This can effectively highlight surface defects and details and ensure that the overall structure of the image is not damaged. The enhancement effect is natural and has high visual and analysis accuracy, thus improving the quality of the enhanced image of the substrate surface cladding effect.

[0065] Preferably, in S5, an image processing model for enhancing the substrate surface cladding effect is constructed. For the input image of the substrate surface cladding effect, it is processed through frequency domain decomposition and adaptive fusion strategy, and a high-order dynamic feature extraction strategy based on phase space evolution and recursive quantization spectrum. The enhanced image generation module for the substrate surface cladding effect is used to generate the enhanced image of the substrate surface cladding effect and output it.

[0066] Compared with the prior art, the present invention has the following technical effects: By adaptively fusing RGB and depth information in the frequency domain, it not only retains rich textures and colors but also can capture geometric changes; Using phase space evolution and recursive quantization spectrum to extract high-order dynamic features greatly improves the recognition ability for unevenness and defects in the cladding layer; Combining the difference image with the global feature realizes adaptive enhancement of key areas and suppression of background noise; Introducing an adaptive adjustment factor to balance the local and overall structures, the enhancement effect is natural; The overall processing flow has strong robustness and versatility, facilitating industrial inspection and subsequent analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 is a flowchart of an image processing method for the substrate surface cladding effect provided by the present invention.

[0068] Figure 2 is a process diagram of an image processing for the substrate surface cladding effect provided by the present invention.

[0069] Figure 3 is a process diagram of the enhanced image generation module for the substrate surface cladding effect provided by the present invention.

[0070] Figure 4 This is a comparison diagram of the image processing of the substrate surface cladding effect in an embodiment provided by the present invention. Detailed implementation manners

[0071] Next, in combination with the drawings of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0072] Please refer to Figures 1 to 4 , the present invention provides a method for processing an image of the substrate surface cladding effect, aiming to generate an enhanced image of the substrate surface cladding effect by fusing RGB images and depth modal features, thereby improving the evaluation accuracy and reliability of the substrate surface cladding quality, including the following steps.

[0073] S1. Collect images of the substrate surface cladding effect containing RGB and depth information, and make a dataset of images of the substrate surface cladding effect.

[0074] Further, in S1, an industrial depth camera is used to take pictures of the substrate surface with laser cladding, and an RGB image of the substrate surface cladding effect and a depth image of the substrate surface cladding effect are obtained. A total of 3,000 images are obtained. The obtained RGB image of the substrate surface cladding effect and the depth image of the substrate surface cladding effect are labeled using Labelimg and made into a dataset of images of the substrate surface cladding effect, which is divided into a training set and a validation set according to 7:3.

[0075] S2. Construct a frequency-domain decomposition and adaptive fusion strategy, including: using discrete wavelet transform to decompose the image of the substrate surface cladding effect into low-frequency and multi-directional high-frequency sub-bands, introducing learnable parameters to adaptively fuse RGB and depth information in each frequency band, and using inverse wavelet transform to reconstruct the fused feature map of the substrate surface cladding effect.

[0076] Further, in S2, the input images include an RGB image of the substrate surface cladding effect and a depth image of the substrate surface cladding effect , and the specific process includes:

[0077] S21. Perform discrete wavelet transform on the input images respectively. For each channel of perform discrete wavelet transform. The specific process is as follows:

[0078] ,

[0079] Among them represents the low-frequency subband on channel c, , and respectively represent the high-frequency subbands in the horizontal, vertical, and diagonal directions on channel c, represents the discrete wavelet transform, represents the image matrix on channel c,

[0080] For perform the discrete wavelet transform, and the specific process is as follows:

[0081] ,

[0082] Among them represents the low-frequency subband of , and respectively represent the high-frequency subbands in the horizontal, vertical, and diagonal directions;

[0083] S22. Introduce learnable parameters to adjust the fusion ratio of RGB and depth information in each frequency band. The specific process is as follows:

[0084] ,

[0085] Among them is the learnable parameter of the low-frequency subband, , used to control the contribution ratio of the low-frequency information of each channel of the RGB image, is the fused low-frequency subband,

[0086] ,

[0087] Among them is the learnable parameter of the high-frequency subband in the horizontal direction, , used to control the contribution ratio of each channel of the RGB image in the horizontal high-frequency component, is the fused high-frequency subband in the horizontal direction,

[0088] ,

[0089] Among them is the learnable parameter of the high-frequency subband in the vertical direction, , used to control the contribution ratio of each channel of the RGB image in the vertical high-frequency component, is the fused high-frequency subband in the vertical direction,

[0090] ,

[0091] where is the learnable parameter of the diagonal high-frequency sub-band, , which is used to control the contribution ratio of each channel of the RGB image in the diagonal high-frequency component. is the high-frequency sub-band in the diagonal direction after fusion. In this embodiment, , , and are all set to 0.5 initially. Gradient update is performed through backpropagation combined with the loss function, and the values of , , and are restricted to the range [0, 1] by the Sigmoid function;

[0092] S23. Reconstruct the fused feature map of the substrate surface cladding effect by inverse wavelet transform for each frequency band . The specific process is as follows:

[0093] ,

[0094] where represents the inverse wavelet transform.

[0095] S3. Construct a high-order dynamic feature extraction strategy based on phase space evolution and recursive quantization spectrum to extract features from the fused feature map of the substrate surface cladding effect, including: converting the fused feature map of the substrate surface cladding effect to grayscale and unfolding it into a one-dimensional sequence, constructing a phase space trajectory using the Takens embedding theory, and then generating corresponding recursive quantization spectrum indices based on a preset threshold, thereby constructing a global feature vector of the substrate surface cladding effect.

[0096] Furthermore, in S3, convert the fused feature map of the substrate surface cladding effect to the grayscale image of the substrate surface cladding effect . The specific process is as follows:

[0097] ,

[0098] where , and are the values of the RGB three channels of the fused feature map of the substrate surface cladding effect at respectively;

[0099] Unfold the grayscale image of the substrate surface cladding effect into a one-dimensional sequence in row order , the specific process is as follows:

[0100] ,

[0101] where H and W are respectively the height and width of the grayscale image of the cladding effect on the substrate surface .

[0102] According to the Takens theory, select the embedding dimension m and time delay r to construct the reconstructed phase space vector . The selection of the embedding dimension m is to ensure that the reconstructed phase space can fully capture the dynamic information of the system. The selection range of m is [5, 10], and it is selected by the FNN method. The selection of the time delay r determines the temporal relationship between points in the phase space. The selection range of r is [5, 15], and it is selected by the autocorrelation function. It is determined by calculating the autocorrelation function and selecting its first zero crossing point. In this embodiment, the initial value of m is set to 8, and the initial value of r is set to 10, which can effectively capture the texture and structural characteristics of the cladding effect image on the substrate surface. The phase space vector corresponding to the th moment

[0103] ,

[0104] where represents matrix transpose, , and respectively represent the values in the , and th one-dimensional sequences , represents the size of the reconstructed phase space vector .

[0105] Use the reconstructed phase space vector to construct the recurrence matrix , the specific process is as follows:

[0106] ,

[0107] where represents the Euclidean distance, represents the distance threshold, which controls the image similarity judgment. In order to capture the subtle changes in the cladding effect image on the substrate surface, in this embodiment is set to 0.1, and respectively represent the phase space vectors corresponding to the th and th moments, and , denotes the Heaviside step function,

[0108] ,

[0109] Extract the recurrence quantification spectrum index from the recurrence matrix , including recurrence quantity , determinism , entropy and average diagonal length , and the specific process is as follows:

[0110] ,

[0111] where represents the dimension of the recurrence matrix ,

[0112] ,

[0113] where denotes the length of the diagonal in the recurrence matrix , represents the number of strips with a diagonal length of , represents the minimum length threshold of the diagonal,

[0114] ,

[0115] where represents the log function,

[0116] ,

[0117] Then combine the extracted recurrence quantification spectrum indexes to construct a global feature vector of the substrate surface cladding effect , and the specific process is as follows:

[0118] .

[0119] S4. Construct an enhanced image generation module for the substrate surface cladding effect, including: calculating the difference image of the fused feature map of the substrate surface cladding effect, designing an adaptive enhancement intensity adjustment factor, and using the adaptive enhancement intensity adjustment factor and the global feature vector of the substrate surface cladding effect to adjust the enhancement intensity of the local variation amount to generate an enhanced image of the substrate surface cladding effect.

[0120] Furthermore, in the above S4, calculate the difference image of the fused feature map of the substrate surface cladding effect , which reflects the detailed information of the local variation of the fused feature map of the substrate surface cladding effect, and the specific process is as follows:

[0121] ,

[0122] wherein , and respectively represent the pixel values at the positions on the c-th channel of the fusion feature map of the substrate surface cladding effect; , and ;

[0123] Construct an adaptive enhancement intensity adjustment factor , and the specific process is as follows:

[0124] ,

[0125] wherein , , and respectively represent the recursion quantity, certainty, entropy, and average diagonal length extracted from the recursion matrix ; represents the adjustment coefficient of the global enhancement intensity, controlling the overall intensity of the enhancement, and the value range is [0, 10], represents the adjustment coefficient of the local details, controlling the influence of the difference image on the enhancement intensity, and the value range is [0, 2], ensures that in the area with small local changes, the enhancement effect will not be too strong. In this embodiment, is set to 1 for the initial value, is set to 0.5 for the initial value, and the values of and are updated through the loss function ; The specific formula of

[0126] is:

[0127] Use the adaptive enhancement intensity adjustment factor and the global feature vector of the substrate surface cladding effect to adjust the enhancement intensity of the local change amount, and generate the enhanced image of the substrate surface cladding effect , and the specific process is as follows:

[0128] ,

[0129] wherein represents the global feature vector of the substrate surface cladding effect.

[0130] S5. Construct an image processing model for the enhanced substrate surface cladding effect, and use the image data set of the substrate surface cladding effect for training.

[0131] Further, in S5, an image processing model for enhancing the cladding effect on the substrate surface is constructed. For the input image of the cladding effect on the substrate surface, it is processed through a frequency-domain decomposition and adaptive fusion strategy and a high-order dynamic feature extraction strategy based on phase-space evolution and recursive quantization spectrum. The image generation module for enhancing the cladding effect on the substrate surface is used to generate and output the enhanced image of the cladding effect on the substrate surface.

[0132] Further, the image processing model for enhancing the cladding effect on the substrate surface is coded using the Pycharm application and the Python language, and the Pytorch framework is used. The model is trained with an image of the cladding effect on the substrate surface with a resolution of 640×640×3. The model is trained for 300 epochs, with 50 epochs frozen.

[0133] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the inventive concept of the present invention, several modifications and improvements can be made, and these all belong to the protection scope of the present invention.

Claims

1. A method for processing an image of a substrate surface cladding effect, characterized in that: The following steps are involved: S1, collecting substrate surface cladding effect images containing RGB and depth information, and preparing a substrate surface cladding effect image dataset; S2. Constructing a frequency domain decomposition and adaptive fusion strategy, including: using discrete wavelet transform to decompose the substrate surface cladding effect image into low-frequency and multi-directional high-frequency sub-bands, introducing learnable parameters to adaptively fuse RGB and depth information in each frequency band, and using inverse wavelet transform to reconstruct the substrate surface cladding effect fusion feature map; S3. Constructing a high-order dynamic feature extraction strategy based on phase space evolution and recursive quantization spectrum to extract features from the fusion feature map of the substrate surface cladding effect, including: converting the fusion feature map of the substrate surface cladding effect into grayscale and expanding it into a one-dimensional sequence, constructing a phase space trajectory using Takens embedding theory, and then generating corresponding recursive quantization spectrum indicators based on a preset threshold, thereby constructing a global feature vector of the substrate surface cladding effect; S4, constructing a substrate surface cladding effect enhanced image generation module, including: calculating a substrate surface cladding effect fusion feature map difference image, designing an adaptive enhancement strength adjustment factor, using the adaptive enhancement strength adjustment factor and the substrate surface cladding effect global feature vector to adjust the enhancement strength of the local variation, and generating a substrate surface cladding effect enhanced image; S5. Construct an image processing model for enhancing the cladding effect on the surface of a substrate, and use the image data set of the cladding effect on the surface of the substrate for training.

2. According to claim 1, a substrate surface cladding effect image processing method is characterized in that: In S2, the input image includes an RGB image of the cladding effect on the substrate surface. And the depth image of the cladding effect on the substrate surface , the specific method is: S21, perform discrete wavelet transform on the input image, Each channel Perform discrete wavelet transform. The specific method is: ， in express The low frequency subband on channel c, , and Respectively High frequency sub-bands in horizontal, vertical and diagonal directions on channel c, represents discrete wavelet transform, express The image matrix on channel c, for Perform discrete wavelet transform. The specific method is: ， in express The low frequency subband of , and Respectively high frequency sub-bands in horizontal, vertical and diagonal directions; S22. Introduce learnable parameters to adjust the fusion ratio of RGB and depth information in each frequency band. The specific method includes: ， in is the learnable parameter of the low-frequency subband, , is the fused low-frequency subband, ， in is the learnable parameter of the high-frequency sub-band in the horizontal direction, , is the high-frequency sub-band in the horizontal direction after fusion, ， in is the learnable parameter of the high-frequency sub-band in the vertical direction, , is the high-frequency sub-band in the vertical direction after fusion, ， in is the learnable parameter of the high-frequency subband in the diagonal direction, , is the high-frequency sub-band in the diagonal direction after fusion; S23, reconstruct the fused characteristic map of the cladding effect on the substrate surface by inverse wavelet transforming the fused frequency bands , the specific method is: ， in Represents inverse wavelet transform.

3. A substrate surface cladding effect image processing method according to claim 2, characterized in that: In S3, the substrate surface cladding effect is integrated with the characteristic map Convert to grayscale image of cladding effect on substrate surface , the specific method is: ， in , and They are the fusion characteristic diagrams of the substrate surface cladding effect. exist The values ​​of the three RGB channels; Grayscale image of the cladding effect on the substrate surface Expand into a one-dimensional sequence in behavioral order , the specific method is: ， Where H and W are grayscale images of the cladding effect on the substrate surface. height and width; According to Takens theory, we select the embedding dimension m and the time delay r to construct the reconstructed phase space vector , No. The phase space vector corresponding to the moment The specific method of constructing is: ， in represents the matrix transpose, , and Respectively represent , and One-dimensional sequence The value in Represents the reconstructed phase space vector Size; Using the reconstructed phase space vector Constructing a recursive matrix , the specific method is: ， in represents the Euclidean distance, represents the distance threshold, and Respectively represent and The phase space vector corresponding to each moment is , represents the Heaviside step function, From the recursive matrix Extract recursive quantization spectrum indicators from , Certainty ,entropy and the average diagonal length , the specific method is: ， in Represents the recursive matrix The dimension of ， in Represents a recursive matrix The length of the diagonal line, The length of the diagonal is The number of strips, Indicates the minimum length threshold of the diagonal line, ， in represents the log function, ， Then the extracted recursive quantization spectrum indexes are combined to construct the global feature vector of the substrate surface cladding effect , the specific method is: 。 4. A substrate surface cladding effect image processing method according to claim 3, characterized in that: In S4, the cladding effect of the substrate surface is calculated by fusion of the feature map difference image , the specific method is: ， in , and Respectively represent the fusion feature map of the substrate surface cladding effect on the c channel , and The pixel value at the position; Constructing an adaptive enhancement intensity regulator , the specific method is: ， in , , and Respectively represent the recursive matrix Extract the recursion amount, certainty, entropy and average diagonal length of the recursion, represents the adjustment coefficient of the global enhancement strength, The adjustment coefficient representing the local details; The enhanced intensity of the local variation is adjusted by using the adaptive enhanced intensity adjustment factor and the global feature vector of the substrate surface cladding effect to generate an enhanced image of the substrate surface cladding effect. , the specific method is: ， in Represents the global feature vector of the cladding effect on the substrate surface.

5. A substrate surface cladding effect image processing method according to claim 4, characterized in that: In the S5, a substrate surface cladding effect enhanced image processing model is constructed, and the input substrate surface cladding effect image is processed through frequency domain decomposition and adaptive fusion strategy, and high-order dynamic feature extraction strategy based on phase space evolution and recursive quantization spectrum. The substrate surface cladding effect enhanced image generation module is used to generate and output the substrate surface cladding effect enhanced image.

Images