A method and system for detecting defects of regenerated environment-friendly composite polyester fabric

Abstract

This invention discloses a method and system for detecting defects in recycled environmentally friendly composite polyester fabric, relating to the field of textile fabric quality inspection technology. The method includes: performing image enhancement processing on an acquired image of the recycled environmentally friendly composite polyester fabric to obtain an enhanced image; dynamically constructing an adaptive notch filter based on the spectral characteristics of the enhanced image, and performing frequency domain filtering on the enhanced image to suppress background texture, obtaining a residual image; performing multi-scale analysis on the residual image, extracting and fusing various texture feature maps to generate a saliency map characterizing the probability of defects, and obtaining a preliminary detection result based on this saliency map; fusing the preliminary detection result with an auxiliary detection result obtained through another detection path based on the enhanced image, and outputting the final detection result. This method solves the problem of difficulty in decoupling texture background and defect signals in the frequency and spatial domains due to the poor fiber uniformity of recycled environmentally friendly composite polyester fabric.

Description

Technical Field

[0001] This application relates to the field of textile fabric quality testing technology, and in particular to a method and system for detecting defects in recycled environmentally friendly composite polyester fabric. Background Technology

[0002] Because the fiber source of recycled PET material is recycled, recycled environmentally friendly composite polyester fabrics generally suffer from poor fiber fineness and strength uniformity during production. This results in the fabric surface being prone to hidden defects such as "stripes" and "cloud weaves" that are highly coupled with the background texture and have low contrast. These defects have very little visual contrast and their spatial frequency characteristics are intertwined with the normal texture, which is a long-standing problem in the field of textile fabric quality inspection.

[0003] In existing technologies, methods for detecting defects in textiles are mainly divided into two categories, but neither is suitable for detecting latent defects in recycled composite polyester fabrics:

[0004] Methods based on spatial domain statistics and filtering include gray-level thresholding, edge detection, local binary patterns and their variants, and gray-level co-occurrence matrices. These methods are sensitive to global or local gray-level changes in images, but they struggle to distinguish between the large-scale, gradually changing gray-level fluctuations caused by fiber unevenness in recycled fabrics and real defects, resulting in extremely high false alarm rates. Furthermore, they are highly sensitive to uneven illumination and exhibit poor stability.

[0005] Methods based on frequency domain or time-frequency analysis, such as Fourier transform, Gabor filter banks, and wavelet transform, separate texture from defects by analyzing the frequency characteristics of an image. However, the fundamental frequency of texture in recycled environmentally friendly composite polyester fabrics drifts and broadens due to fiber unevenness. Using filters with fixed center frequencies and bandwidths or fixed threshold spectrum filters is difficult to accurately adapt, easily leading to over-filtering of texture information or loss of defect signals, resulting in unstable detection results.

[0006] Therefore, there is an urgent need for a dedicated detection method that can adapt to the inherent fluctuations in the texture of recycled and environmentally friendly composite polyester fabrics and has high robustness and high precision for low-contrast hidden defects. Summary of the Invention

[0007] To address the technical problems of the prior art, this application provides a method and system for detecting defects in recycled environmentally friendly composite polyester fabrics. This method combines adaptive frequency domain suppression based on frequency domain perception with spatial domain analysis based on multi-dimensional feature decision-level fusion. By utilizing the deep synergy between the two, it achieves accurate, stable, and adaptive detection of latent defects in recycled environmentally friendly composite polyester fabrics. This aims to solve the problem that the texture background and defect signals are difficult to decouple in the frequency and spatial domains due to the poor fiber uniformity of recycled environmentally friendly composite polyester fabrics.

[0008] This application provides a method for detecting defects in recycled environmentally friendly composite polyester fabric, including: Step S10: Perform homomorphic filtering on the real-time acquired image of the recycled environmentally friendly composite polyester fabric to correct uneven lighting in the image and improve image contrast, thereby obtaining an enhanced image of the recycled environmentally friendly composite polyester fabric. Step S20: Perform frequency domain transformation on the enhanced image of the recycled environmentally friendly composite polyester fabric to obtain its spectrum, identify the main frequency components in the spectrum and calculate the energy diffusion of each frequency component, dynamically construct an adaptive notch filter bank based on the energy diffusion, use the filter bank to filter the spectrum and perform inverse transformation to obtain the residual image of the recycled environmentally friendly composite polyester fabric with suppressed background texture. Step S30: Perform multi-scale and multi-directional transformation on the residual image of the recycled environmentally friendly composite polyester fabric to obtain sub-band coefficients, calculate and fuse the local energy feature map, local statistical outlier feature map and local orientation consistency feature map of each sub-band, generate a comprehensive saliency map for image segmentation, and obtain preliminary defect location results. Step S40: Based on the enhanced image of the recycled environmentally friendly composite polyester fabric, the auxiliary detection results are obtained through parallel auxiliary detection paths. The preliminary defect location results and the auxiliary detection results are fused with confidence-driven decision, and the fusion results are post-processed to output the final defect location image.

[0009] Furthermore, the images of recycled environmentally friendly composite polyester fabrics acquired by an industrial area scan camera under standard light sources are converted into grayscale images. Homomorphic filtering is then used to process the grayscale images, including the following detailed steps: A logarithmic transformation is performed on the grayscale image of the recycled environmentally friendly composite polyester fabric to transform the multiplicative model of the illuminance and reflectance components into an additive model. Subsequently, a high-frequency enhancement and low-frequency suppression filter function is used in the frequency domain for processing. Finally, an exponential transformation is performed to obtain the enhanced image.

[0010] In the enhanced image of recycled composite polyester fabric, the areas of light and dark gradient caused by the position of the light source and slight undulations of the fabric are significantly smoothed, and the gray distribution of the entire image is more uniform. The gray difference between the yarn texture, weaving structure, and hidden defects such as stripes and cloud patterns of the fabric itself and the surrounding normal areas is nonlinearly amplified, making the defects more prominent in terms of visual and statistical characteristics.

[0011] Furthermore, addressing the unstable fundamental frequency and spectral broadening of textures caused by uneven fiber distribution in recycled fabrics, step S20 intelligently and adaptively extracts the dominant, periodic, or quasi-periodic background texture from the enhanced image, thereby highlighting defect areas that disrupt texture regularity: Based on the spectral characteristics of the enhanced image from recycled and environmentally friendly composite polyester fabric, a filter is dynamically constructed to achieve adaptive suppression of background texture, including the following detailed steps: Step S21: Perform a two-dimensional fast Fourier transform on the enhanced image of the recycled environmentally friendly composite polyester fabric to obtain the complex spectrum and calculate the logarithmic amplitude spectrum. Step S22: In the logarithmic amplitude spectrum, automatically detect the main frequency component set of the fundamental frequency and harmonics of the regular texture of the recycled environmentally friendly composite polyester fabric, and record the corresponding coordinates. Step S23: For each main frequency point in the main frequency component set, calculate the information entropy of the neighborhood centered on the main frequency point, as the energy diffusion degree corresponding to the main frequency component. Step S24: For each main frequency point, a two-dimensional Gaussian notch bandstop filter is generated. The filter parameters and the energy spread obtained in step S23 are dynamically correlated through a preset monotonically increasing function. Step S25: Multiply the transfer functions of all two-dimensional Gaussian notch bandstop filters to obtain a composite filter. Apply the composite filter to filter the logarithmic amplitude spectrum. Combine the filtered amplitude spectrum with the original phase spectrum and perform inverse Fourier transform and exponential adjustment to finally obtain a residual image with significantly suppressed background texture.

[0012] In the residual image of recycled composite polyester fabric, the regular warp and weft interlacing patterns become very faint or almost disappear. Hidden defects such as stripes and cloud patterns that disrupt the periodicity and uniformity of the texture are largely preserved during the filtering process because their frequency components are different from the background dominant frequency. Therefore, they become relatively clear and prominent in the image, reducing the difficulty of feature extraction and segmentation in subsequent spatial domain analysis.

[0013] Furthermore, step S30 involves a detailed analysis of the residual image of the recycled environmentally friendly composite polyester fabric in the spatial domain. By fusing multi-dimensional features, a saliency map sensitive to defects is constructed, including the following detailed steps: Step S31: Perform non-subsampled shear wave transform on the residual image of the recycled environmentally friendly composite polyester fabric, and obtain a set of subband coefficients after multi-layer decomposition. Step S32: For each directional sub-band coefficient, calculate the local energy feature map, local statistical outlier feature map, and local directional consistency feature map within a sliding window in the spatial domain; Step S33: Within the same scale, average pooling is performed on the feature maps of the same type in each direction to obtain the comprehensive feature map of that scale. Weighted fusion is performed on the comprehensive feature maps of different scales to generate a global feature map. Step S34: The global feature maps of local energy, local statistical outlier and local orientation consistency are weighted and fused to generate the final comprehensive saliency map; Step S35: Apply the adaptive threshold segmentation algorithm to the comprehensive saliency map obtained in step S34 to perform image segmentation, and obtain a preliminary binary defect mask as the preliminary defect localization result.

[0014] Furthermore, to improve the decision-making reliability of the detection system under complex conditions, step S40 introduces a parallel auxiliary detection path for verification, including the following detailed steps: Step S41: Extract the multi-scale complete local binary mode features and the contrast, correlation and energy features of the gray-level co-occurrence matrix of the enhanced image of the recycled environmentally friendly composite polyester fabric obtained in step S10. Step S42: Input the various features obtained in step S41 into the pre-trained classifier to classify each pixel and obtain the auxiliary path confidence and the auxiliary detection binary mask. Step S43: For each connected region of the preliminary binarized mask in the preliminary defect localization result, calculate the average saliency value of the region in the comprehensive saliency map, and use it as the main path confidence of the region. Step S44: Based on the confidence of the main path and the confidence of the auxiliary path, the preliminary defect localization result and the auxiliary detection result are weighted and fused to obtain the fused binary mask; Step S45: Perform a series of morphological operations on the fused binary mask to finally output the optimized defect localization image.

[0015] This application also provides a defect detection system for recycled environmentally friendly composite polyester fabrics, including: Image acquisition and preprocessing module: used to perform homomorphic filtering on the real-time acquired images of recycled environmentally friendly composite polyester fabric, correct uneven lighting in the image and improve image contrast, to obtain an enhanced image of the recycled environmentally friendly composite polyester fabric. Adaptive frequency domain suppression module: used to perform frequency domain transformation on the enhanced image of recycled environmentally friendly composite polyester fabric to obtain its spectrum, identify the main frequency components in the spectrum and calculate the energy spread of each frequency component, dynamically construct an adaptive notch filter bank based on the energy spread, use the filter bank to filter the spectrum and perform inverse transformation to obtain the residual image of recycled environmentally friendly composite polyester fabric with suppressed background texture. The spatial domain multidimensional analysis module is used to perform multi-scale and multi-directional transformations on the residual image of recycled environmentally friendly composite polyester fabric to obtain sub-band coefficients, calculate and fuse the local energy feature map, local statistical outlier feature map and local orientation consistency feature map of each sub-band, generate a comprehensive saliency map for image segmentation, and obtain preliminary defect location results. Decision Fusion and Optimization Module: This module is used to enhance images based on recycled environmentally friendly composite polyester fabrics. It acquires auxiliary detection results through parallel auxiliary detection paths, performs confidence-driven decision fusion between the preliminary defect location results and the auxiliary detection results, and performs post-processing on the fusion results to output the final defect location image.

[0016] This application discloses the following technical effects: This application provides a method and system for detecting defects in recycled environmentally friendly composite polyester fabrics. Compared with existing frequency domain filtering methods with fixed parameters, this invention dynamically adjusts the filter width by quantifying the energy diffusion of the main frequency, and automatically associates the filter design parameters with the spectral physical characteristics of the fabric image itself that reflect the uniformity of the fibers. This fundamentally solves the problem of poor adaptability of fixed parameter filters to the texture fluctuations of recycled fabrics.

[0017] In spatial domain analysis, in addition to traditional energy and outlier features, this invention proposes a local orientation consistency feature, which can effectively capture the disruption of texture regularity and provide a deep description of the essential attributes of latent defects. By fusing three features that characterize defects from different angles at the decision level, the discriminative power and robustness of defects are significantly improved, which is an effect that cannot be achieved by a single feature or simply connecting multiple features.

[0018] Furthermore, this invention is not a simple serial processing, but rather a parallel mechanism based on the main path (frequency domain and spatial domain collaboration) and the auxiliary path (texture classification). It uses confidence to perform decision-level fusion, allowing the detection system to refer to the results of another independent judgment path when its own uncertainty is high, thereby improving the overall stability and accuracy of the system when facing complex and ambiguous situations. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for detecting defects in recycled environmentally friendly composite polyester fabrics, provided in an embodiment of this application.

[0020] Figure 2 This is a schematic diagram of the structure of a defect detection system for recycled environmentally friendly composite polyester fabric provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] Example 1: This application provides a method for producing recycled and environmentally friendly composite polyester fabric, such as... Figure 1 As shown, the method includes: Step S10: Perform homomorphic filtering on the real-time acquired image of the recycled environmentally friendly composite polyester fabric to correct uneven lighting in the image and improve image contrast, thereby obtaining an enhanced image of the recycled environmentally friendly composite polyester fabric.

[0023] In this embodiment, the original grayscale image of the recycled environmentally friendly composite polyester fabric acquired in real time is used. The imaging process is modeled as an illuminance component. With reflection component The product of:

[0024] in, The low-frequency illumination field that corresponds to slow changes is the source of interference; Corresponding to the reflective properties of the fabric itself, it includes high-frequency information about fabric texture and defects; and These represent the x-coordinate and y-coordinate in the image coordinate system, respectively. First, the original grayscale image of the recycled environmentally friendly composite polyester fabric is subjected to a natural logarithmic transformation to convert the multiplicative model into an additive model, which facilitates separation processing.

[0025] in, This represents the transformed image. This represents the logarithmic operation; for Perform a two-dimensional discrete Fourier transform to obtain the frequency domain representation. , and These represent the spatial frequencies in the horizontal and vertical directions, respectively. Secondly, this embodiment uses a Gaussian homomorphic filter to... The transfer function of the filter is processed. This can be expressed as a formula:

[0026] in, Representing frequency point Distance to the center of the frequency plane This is the cutoff frequency of the homomorphic filter. To control the steepness of the slope of the homomorphic filter, and These represent the high-frequency gain coefficient and the low-frequency gain coefficient, respectively. Furthermore, the parameters of the homomorphic filter were specifically adjusted to suit the image characteristics of recycled composite polyester fabrics: Set within the range [2.0, 2.2] to enhance the edges of defects and fine textures; Set to 0.5 for uneven lighting in fabric production workshops; The size of the defects in the selected fabric image is relevant; for defects with a certain width, such as stripes and cloud patterns, Set to 0.05 to 0.15 times the short side size of the image (normalized frequency). By suppressing components below this frequency, large-scale shadows can be effectively eliminated while preserving defect information. The filtering operation is performed using the transfer function described above to obtain the filtering result. Perform an inverse Fourier transform to transform to the spatial domain, and obtain the spatial representation of the filtered result. ; Finally, Perform an exponential transform to cancel out the initial natural logarithmic transform, resulting in the final enhanced image. : ; Step S20: Perform frequency domain transformation on the enhanced image of the recycled environmentally friendly composite polyester fabric to obtain its spectrum, identify the main frequency components in the spectrum and calculate the energy diffusion of each frequency component, dynamically construct an adaptive notch filter bank based on the energy diffusion, use the filter bank to filter the spectrum and perform inverse transformation to obtain the residual image of the recycled environmentally friendly composite polyester fabric with suppressed background texture.

[0027] To address the unstable fundamental frequency and spectral broadening of textures in recycled fabrics caused by uneven fiber distribution, this step intelligently and adaptively extracts the dominant, periodic, or quasi-periodic background texture from the enhanced image, thereby highlighting defect areas that disrupt texture regularity. This includes the following detailed steps: Step S21: Enhance the image of the pre-processed recycled environmentally friendly composite polyester fabric. Perform a frequency domain transformation to obtain the complex frequency. To facilitate observation and analysis of the spectral energy distribution, its logarithmic amplitude spectrum is calculated. :

[0028] Logarithmic transformation can compress the spectral values ​​of high dynamic range, making the frequency components representing weaker texture harmonics clearly visible in the spectrum.

[0029] Step S22, in the calculated logarithmic amplitude spectrum In the text, the frequency components representing the regular background texture of the fabric (warp and weft yarn interweaving) will appear as bright spots (spectral peaks) with significantly higher energy than the surrounding background. These dominant frequency components can be automatically identified through the following steps: exist The local maximum detection algorithm is applied to a candidate point. If its amplitude value is the largest in a predetermined neighborhood, it is considered a local spectral peak. Calculate the energy values ​​of all detected local spectral peaks and sort them from highest to lowest energy. Since the fabric background texture is usually composed of the fundamental frequency and a few strong harmonics, select the top K highest energy spectral peaks as the main frequency component set. The K value is set empirically to 4 to 10, which is sufficient to cover the main texture frequencies.

[0030] Step S23, for each identified main frequency point Calculate its energy diffusivity This is used to quantify the stability of the frequency component in the frequency domain, in order to Define a radius centered at a point. circular neighborhood , Set to 3 to 7 pixels to cover the main energy distribution region of the spectral peak; in this neighborhood Within, calculate the standard deviation of the logarithmic magnitude values. As energy diffusivity Measurement:

[0031] in, The formula for calculating standard deviation is given. For recycled environmentally friendly composite polyester fabrics, due to poor fiber uniformity, the texture is not strictly periodic in the spatial domain, resulting in the corresponding spectral peak energy being unconcentrated in the frequency domain and spreading to surrounding frequencies. (Standard deviation is also mentioned.) The larger the value, the less sharp the spectral peak is, and the worse the periodicity and uniformity of the corresponding texture. This is the key perception basis for achieving adaptive behavior.

[0032] Step S24: Based on the calculated energy spread, dynamically construct a two-dimensional Gaussian notch bandstop filter for each dominant frequency point. The transfer function of each filter Represented as:

[0033] Wherein, the function is at the center frequency The gain at a certain point is 0 (complete suppression), and the gain gradually approaches 1 (no effect). The degree of suppression is determined by the stopband width parameter. control, The larger the value, the wider the inhibition range; Filter stopband width parameter Compared with the calculated energy diffusivity Proportional relationship:

[0034] in, As a scaling factor, it is set to [1.5, 3] in this embodiment. The base stopband width is set to [0.5, 1] ​​in this embodiment; for energy concentration ( Small, regular texture) main frequency, using a narrower stopband ( (Small) precise suppression is performed to avoid damaging defect information; for energy diffusion ( If the main frequency is large and the texture is uneven, then a wider stopband ( (Large) blur suppression is performed to ensure that unstable background textures are effectively filtered out.

[0035] Step S25: Combine all the two-dimensional Gaussian notch bandstop filters targeting a single dominant frequency point into a composite filter. :

[0036] Apply the synthesized filter to the complex spectrum Selective attenuation is performed on each dominant frequency representing the background texture and its neighborhood to obtain the filtered spectrum. Perform an inverse Fourier transform and take the real part to obtain the complex result in the spatial domain. Then, adjust the result exponentially to obtain the final residual image. :

[0037] in, Indicates the inverse Fourier transform. This indicates the operation of taking the real part.

[0038] Step S30: Perform multi-scale and multi-directional transformation on the residual image of the recycled environmentally friendly composite polyester fabric to obtain sub-band coefficients. Calculate and fuse the local energy feature map, local statistical outlier feature map, and local directional consistency feature map of each sub-band to generate a comprehensive saliency map for image segmentation and obtain preliminary defect location results.

[0039] This step involves multi-scale, multi-directional analysis of the residual image where background texture is suppressed, and the fusion of three features with complementary discriminative power for texture anomalies to construct a comprehensive saliency map that can accurately highlight defect areas, thereby achieving preliminary localization. The detailed steps include: Step S31: Perform non-subsampled shear wave transform (NSST) on the residual image obtained in step S20 to capture texture details at different scales and directions in the image. The decomposition process includes: Perform L-level multiscale decomposition, at the Lth level scale( =1,2,...,L, where L=4 in this example), decompose the image into... Different directions ( The subbands (which increase with scale) are decomposed using NSST to obtain a set of subband coefficients. , For direction index, These are the spatial coordinates.

[0040] Step S32: For each sub-band coefficient at each spatial location, compute three types of feature maps in parallel within its local neighborhood, including: Local energy feature map : through The size of the center is It is calculated from the sum of squares of the subband coefficients within the window and is used to characterize The intensity of texture activity at a location:

[0041] in, and These represent the horizontal and vertical directions, respectively, relative to the current center pixel. The offset; defective areas usually mean that the texture rules have been broken, resulting in abnormal changes in local energy.

[0042] Local statistical outlier feature map : Used to measure the degree of anomaly of the central sub-band coefficient value relative to the statistical characteristics of its local neighborhood, and to calculate Standardized absolute deviation of position sub-band coefficients:

[0043]

[0044]

[0045] in, Point The set of subband coefficients within a local neighborhood. and Let represent the mean and standard deviation of the sub-band coefficient set, respectively. It is a very small constant to prevent division by zero; It quantifies whether a pixel is an outlier relative to its surroundings, and is sensitive to point-like or fine strip-like defects.

[0046] Local orientation consistency feature map For the same spatial location and the same scale, collect all data at that scale. The absolute values ​​of the subband coefficients in each direction subband are used to form a vector. Normalize the probability distribution of this vector. Make the sum of its components equal to 1; calculate the information entropy of this probability distribution, which serves as the local directional consistency feature at this location and scale:

[0047] in, Indicates the location Location, scale Down, direction The normalized energy probability corresponding to the subband coefficient; In areas with normal texture, energy is concentrated in the warp or weft direction, resulting in a concentrated distribution and low entropy. However, in defect areas such as stripes or cloud-like patterns, the texture directionality is disrupted or blurred, and energy is dispersed in multiple directions, leading to… The distribution is more uniform, and the entropy value is significantly increased.

[0048] Step S33, on the same scale Within this range, average pooling is performed on similar feature maps in each direction to obtain the local energy comprehensive feature map at that scale. Local statistical outlier comprehensive feature map , and directional consistency feature It is already a comprehensive feature map; weighted fusion of comprehensive feature maps at different scales generates three types of global feature maps; Step S34: Combine the global feature map of local energy, local statistical outlier, and local orientation consistency. , and Weighted fusion is performed to generate the final comprehensive saliency map. :

[0049] in, This means normalizing the feature map to the [0,1] interval. , and These are weighting coefficients, determined through grid search optimization. The higher the value, the greater the likelihood that the location is a defect.

[0050] Step S35: Apply an adaptive threshold segmentation algorithm to the comprehensive saliency map from step S34 to perform image segmentation, obtaining a preliminary binarized defect mask. This serves as the initial result for locating the defect.

[0051] Step S40: Based on the enhanced image of the recycled environmentally friendly composite polyester fabric, the auxiliary detection results are obtained through parallel auxiliary detection paths. The preliminary defect location results and the auxiliary detection results are fused with confidence-driven decision, and the fusion results are post-processed to output the final defect location image.

[0052] This step is independent of the main path analysis process in steps S10 to S30. By introducing an independent auxiliary detection path, it performs cross-validation and intelligent fusion with the results of the main path, effectively handling the uncertainties that the main path may have in complex areas, thereby outputting more stable and accurate final results, including: Step S41: Extract the multi-scale complete local binary mode features and the contrast, correlation, and energy features of the gray-level co-occurrence matrix from the enhanced image of the recycled environmentally friendly composite polyester fabric obtained in step S10. Multi-scale complete local binary pattern features are derived by computing improved LBP histograms at multiple scales. The gray-level co-occurrence matrix is ​​calculated in the 0°, 45°, 90° and 135° directions, and contrast, correlation and energy features are extracted from it to describe the macroscopic structural rules of the texture. The above features are concatenated to form a high-dimensional feature vector.

[0053] Step S42: Input the various features obtained in step S41 into the pre-trained classifier to classify each pixel and obtain the auxiliary path confidence and the auxiliary detection binary mask. The extracted feature vectors are input into a pre-trained support vector machine (SVM) classifier, which has been trained on a sample set consisting of normal fabric pieces and defective fabric pieces. For each region in the input image, the SVM outputs not only its classification label (defect, normal) but also a probability estimate. This estimate is normalized and used as the auxiliary path confidence for that region. Generate the corresponding auxiliary detection binary mask. .

[0054] Step S43: For each connected region of the preliminary binary mask in the preliminary defect localization result, calculate the average saliency value of the region in the comprehensive saliency map, which is used as the main path confidence of that region. Preliminary defect location results Each connected region in (i.e., a candidate defect), calculate its average comprehensive significance value over the entire region, and use it as the main path confidence for that region. This reflects the degree of certainty that the main path considers the area to be a defect. Step S44: Based on the confidence scores of the main path and the auxiliary path, the preliminary defect localization results and the auxiliary detection results are weighted and fused to obtain the fused binary mask. Preset two thresholds and The confidence space is divided into three decision intervals: High confidence interval ( The main path result is very certain, so we directly adopt the main path result. Retained in the fusion result; Low confidence interval ( The main path result has high uncertainty. In this case, the corresponding query area is detected by the binary mask in the auxiliary detection. If the auxiliary path also detects a defect, the result of the auxiliary path is adopted; otherwise, the area is considered a false alarm and is removed from the results. Fuzzy interval ( The results from the main path have a certain degree of reliability but are not entirely certain. Within this range, the results from the main and secondary paths are weighted and fused. The weighting coefficients are... and Proportional.

[0055] Step S45: Perform a series of morphological operations on the fused binary mask to finally output the optimized defect localization image. .

[0056] Example 2: The recycled environmentally friendly composite polyester fabric system provided in this embodiment of the invention can execute the recycled environmentally friendly composite polyester fabric method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method, such as... Figure 2 As shown, it includes the following modules: Image acquisition and preprocessing module: used to perform homomorphic filtering on the real-time acquired images of recycled environmentally friendly composite polyester fabric, correct uneven lighting in the image and improve image contrast, to obtain an enhanced image of the recycled environmentally friendly composite polyester fabric. Adaptive frequency domain suppression module: used to perform frequency domain transformation on the enhanced image of recycled environmentally friendly composite polyester fabric to obtain its spectrum, identify the main frequency components in the spectrum and calculate the energy spread of each frequency component, dynamically construct an adaptive notch filter bank based on the energy spread, use the filter bank to filter the spectrum and perform inverse transformation to obtain the residual image of recycled environmentally friendly composite polyester fabric with suppressed background texture. The spatial domain multidimensional analysis module is used to perform multi-scale and multi-directional transformations on the residual image of recycled environmentally friendly composite polyester fabric to obtain sub-band coefficients, calculate and fuse the local energy feature map, local statistical outlier feature map and local orientation consistency feature map of each sub-band, generate a comprehensive saliency map for image segmentation, and obtain preliminary defect location results. Decision Fusion and Optimization Module: This module is used to enhance images based on recycled environmentally friendly composite polyester fabrics. It acquires auxiliary detection results through parallel auxiliary detection paths, performs confidence-driven decision fusion between the preliminary defect location results and the auxiliary detection results, and performs post-processing on the fusion results to output the final defect location image.

[0057] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0058] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for detecting defects in recycled environmentally friendly composite polyester fabric, characterized in that, The method includes: Step S10: Perform homomorphic filtering on the real-time acquired image of the recycled environmentally friendly composite polyester fabric to correct uneven lighting in the image and improve image contrast, thereby obtaining an enhanced image of the recycled environmentally friendly composite polyester fabric. Step S20: Perform frequency domain transformation on the enhanced image of the recycled environmentally friendly composite polyester fabric to obtain its spectrum, identify the main frequency components in the spectrum and calculate the energy diffusion of each frequency component, dynamically construct an adaptive notch filter bank based on the energy diffusion, use the filter bank to filter the spectrum and perform inverse transformation to obtain the residual image of the recycled environmentally friendly composite polyester fabric with suppressed background texture. Step S30: Perform multi-scale and multi-directional transformation on the residual image of the recycled environmentally friendly composite polyester fabric to obtain sub-band coefficients, calculate and fuse the local energy feature map, local statistical outlier feature map and local orientation consistency feature map of each sub-band, generate a comprehensive saliency map for image segmentation, and obtain preliminary defect location results. Step S40: Based on the enhanced image of the recycled environmentally friendly composite polyester fabric, the auxiliary detection results are obtained through parallel auxiliary detection paths. The preliminary defect location results and the auxiliary detection results are fused with confidence-driven decision, and the fusion results are post-processed to output the final defect location image.

2. The method for detecting defects in recycled environmentally friendly composite polyester fabric as described in claim 1, characterized in that, Step S20 includes the following detailed steps: Step S21: Perform a two-dimensional fast Fourier transform on the enhanced image of the recycled environmentally friendly composite polyester fabric to obtain the complex spectrum and calculate the logarithmic amplitude spectrum. Step S22: In the logarithmic amplitude spectrum, automatically detect the main frequency component set of the fundamental frequency and harmonics of the regular texture of the recycled environmentally friendly composite polyester fabric, and record the corresponding coordinates. Step S23: For each main frequency point in the main frequency component set, calculate the information entropy of the neighborhood centered on the main frequency point, as the energy diffusion degree corresponding to the main frequency component. Step S24: For each main frequency point, a two-dimensional Gaussian notch bandstop filter is generated. The filter parameters and the energy spread obtained in step S23 are dynamically correlated through a preset monotonically increasing function. Step S25: Multiply the transfer functions of all two-dimensional Gaussian notch bandstop filters to obtain a composite filter. Apply the composite filter to filter the logarithmic amplitude spectrum. Combine the filtered amplitude spectrum with the original phase spectrum and perform inverse Fourier transform and exponential adjustment to finally obtain a residual image with significantly suppressed background texture.

3. The method for detecting defects in recycled environmentally friendly composite polyester fabric as described in claim 2, characterized in that, The transfer function of the two-dimensional Gaussian notch bandstop filter Represented as: in, Indicates the first The transfer function of each filter. and These represent the spatial frequencies in the horizontal and vertical directions, respectively. The function has a center frequency of... The gain at that point is 0, and the degree of suppression is determined by the stopband width parameter. control, The larger the size, the wider the inhibition range. Compared with the calculated energy diffusivity They are directly proportional.

4. The method for detecting defects in recycled environmentally friendly composite polyester fabric as described in claim 1, characterized in that, Step S30 includes the following detailed steps: Step S31: Perform non-subsampled shear wave transform on the residual image of the recycled environmentally friendly composite polyester fabric, and obtain a set of subband coefficients after multi-layer decomposition. Step S32: For each directional sub-band coefficient, calculate the local energy feature map, local statistical outlier feature map, and local directional consistency feature map within a sliding window in the spatial domain; Step S33: Within the same scale, average pooling is performed on the feature maps of the same type in each direction to obtain the comprehensive feature map of that scale. Weighted fusion is performed on the comprehensive feature maps of different scales to generate a global feature map. Step S34: The global feature maps of local energy, local statistical outlier and local orientation consistency are weighted and fused to generate the final comprehensive saliency map; Step S35: Apply the adaptive threshold segmentation algorithm to the comprehensive saliency map obtained in step S34 to perform image segmentation, and obtain a preliminary binary defect mask as the preliminary defect localization result.

5. The method for detecting defects in recycled environmentally friendly composite polyester fabric as described in claim 4, characterized in that, The local energy feature map , through It is calculated from the sum of squares of the sub-band coefficients within the centered window and is used to characterize... The intensity of texture activity at a given location. and Represent the x-axis and y-axis respectively; The local statistical outlier feature map It is used to measure the degree of anomaly of the central sub-band coefficient value relative to the statistical characteristics of its local neighborhood, and is calculated. Standardized absolute deviation, For scale indexing, For direction index; The local directional consistency feature map The calculation process includes: For the same spatial location and the same scale, collect all data at that scale. The absolute values ​​of the subband coefficients in each direction subband are used to form a vector. Normalize the probability distribution of this vector. Make the sum of its components equal to 1; calculate the information entropy of the probability distribution, and use it as the local directional consistency feature at that location and scale.

6. The method for detecting defects in recycled environmentally friendly composite polyester fabric as described in claim 1, characterized in that, Step S40 is independent of the main path analysis process of steps S10 to S30. It introduces an independent auxiliary detection path to cross-validate and intelligently fuse the results with the main path results, including the following detailed steps: Step S41: Extract the multi-scale complete local binary mode features and the contrast, correlation and energy features of the gray-level co-occurrence matrix of the enhanced image of the recycled environmentally friendly composite polyester fabric obtained in step S10. Step S42: Input the various features obtained in step S41 into the pre-trained classifier to classify each pixel and obtain the auxiliary path confidence. and auxiliary detection binary mask ; Step S43: For each connected region of the preliminary binarized mask in the preliminary defect localization result... Calculate the average significance value of the region in the comprehensive significance map, as... Main path confidence ; Step S44: Based on the confidence of the main path and the confidence of the auxiliary path, the preliminary defect localization result and the auxiliary detection result are weighted and fused to obtain the fused binary mask; Step S45: Perform a series of morphological operations on the fused binary mask to finally output the optimized defect localization image.

7. The method for detecting defects in recycled environmentally friendly composite polyester fabric as described in claim 6, characterized in that, In step S44, two thresholds are preset. and The confidence space is divided into three decision intervals: High confidence interval : Adopt the main path result, that is Retained in the fusion result; Low confidence interval The main path results are uncertain; the corresponding query area is detected by the binary mask in the auxiliary detection. If the auxiliary path also detects a defect, the result of the auxiliary path is adopted; otherwise, the area is considered a false alarm and is removed from the results. Fuzzy interval Within this range, the results of the main and auxiliary paths are weighted and fused, with the weighting coefficients being... and Proportional.

8. A defect detection system for recycled environmentally friendly composite polyester fabric, characterized in that, The system is used to implement the defect detection method for recycled environmentally friendly composite polyester fabric according to any one of claims 1-7, and the system includes: Image acquisition and preprocessing module: used to perform homomorphic filtering on the real-time acquired images of recycled environmentally friendly composite polyester fabric, correct uneven lighting in the image and improve image contrast, to obtain an enhanced image of the recycled environmentally friendly composite polyester fabric. Adaptive frequency domain suppression module: used to perform frequency domain transformation on the enhanced image of recycled environmentally friendly composite polyester fabric to obtain its spectrum, identify the main frequency components in the spectrum and calculate the energy spread of each frequency component, dynamically construct an adaptive notch filter bank based on the energy spread, use the filter bank to filter the spectrum and perform inverse transformation to obtain the residual image of recycled environmentally friendly composite polyester fabric with suppressed background texture. The spatial domain multidimensional analysis module is used to perform multi-scale and multi-directional transformations on the residual image of recycled environmentally friendly composite polyester fabric to obtain sub-band coefficients, calculate and fuse the local energy feature map, local statistical outlier feature map and local orientation consistency feature map of each sub-band, generate a comprehensive saliency map for image segmentation, and obtain preliminary defect location results. Decision Fusion and Optimization Module: This module is used to enhance images based on recycled environmentally friendly composite polyester fabrics. It acquires auxiliary detection results through parallel auxiliary detection paths, performs confidence-driven decision fusion between the preliminary defect location results and the auxiliary detection results, and performs post-processing on the fusion results to output the final defect location image.

Images