Method for intelligently identifying high-cut low-loop, flat-cut flat-loop carpet pile positioning superposition

Abstract

The application discloses a method for intelligently identifying and positioning and superimposing a high-cut low-loop and flat-cut flat-loop carpet pile surface, and belongs to the technical field of digital printing. First, under the irradiation of a high-intensity light source, the carpet surface image is collected by a high-speed camera by utilizing the significant difference in the reflection characteristics of the cut-pile and loop-pile areas. Second, the cut-pile area (A) and the loop-pile area (B) are accurately identified and distinguished based on an image segmentation technology, and specifically, a threshold segmentation based on color features or a semantic segmentation model based on deep learning can be adopted. Third, a mapping relationship between the areas and the printed patterns is established, a first pattern is printed in the A area, and a second pattern is printed in the B area. Finally, compensation is performed on the area boundary by a boundary pixel expansion algorithm, and accurate positioning and superimposition are realized. The application completely solves the industry problems of misplacement and ghosting of three-dimensional pile carpet printing, realizes perfect fusion or accurate comparison of patterns and underlines, and greatly improves the design freedom and product artistic value.

Description

Technical Field

[0001] This invention relates to the field of digital printing and textile fabric pattern processing technology, and in particular to a method for intelligent recognition and zoned positioning overprinting based on the characteristics of pile structure for high-cut low-loop and flat-cut flat-loop tufted carpets with three-dimensional pile structure. Background Technology

[0002] Tufted carpets, with their rich texture and excellent durability, have become a mainstream flooring material. Among them, high-cut low-loop and flat-cut flat-loop carpets, which create unique three-dimensional patterns through their staggered pile structure, are highly popular in the market. Carpet printing technology has gone through several stages, from traditional screen printing and pigment printing to high-definition digital inkjet printing.

[0003] However, existing technologies have the following technical shortcomings when processing carpet printing with significant three-dimensional structures, such as high-cut low-circle and flat-cut flat-circle patterns: Poor structural adaptability: Traditional screen printing is only suitable for flat raw carpets and cannot form clear and accurate patterns on uneven carpet surfaces.

[0004] Pattern and background misalignment: Most current digital inkjet printing is full-coverage printing, which ignores the three-dimensional pattern (background) formed by the weaving of the raw carpet and directly prints the design pattern as a whole. Because the raw carpet will inevitably be stretched and deformed during the weaving, gluing, and backing processes, the printed pattern cannot be accurately aligned with the original woven background, resulting in ghosting, misalignment, or blank areas, and poor visual effect.

[0005] Design limitations: To avoid the aforementioned misalignment issues, designers often have to use floral patterns with long cycles and no emphasis on alignment to cover up the defects, making it impossible to achieve precise integration or deliberate contrast between the pattern and the background, thus limiting the product's artistic expression.

[0006] Therefore, how to achieve precise, controllable, and artistic positioning and overprinting of carpets with three-dimensional pile structures is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0007] This invention addresses the shortcomings of existing technologies by providing a method for intelligently identifying and overprinting high-cut and low-circle, flat-cut and flat-circle carpet pile surfaces. It aims to solve the problems of inaccurate registration of printed patterns and backgrounds, pattern misalignment, and design limitations caused by the three-dimensional structure of carpet surfaces.

[0008] Therefore, in its first aspect, this application provides a method for intelligently identifying and positioning overprinted carpet pile surfaces with high-cut and low-circle, and flat-cut and flat-circle designs, comprising the following steps: Step S1, Image Acquisition Step: At the front end of the printing area of ​​the digital printing machine, the carpet surface of the carpet blank to be printed (high-cut low-loop or flat-cut flat-loop) is illuminated by a light source, and the original image of the carpet surface containing the cut pile area and the loop pile area is captured by a camera device. Step S2, Region Segmentation Step: Obtain the original image of the carpet surface. Utilize the difference in color reflection characteristics between the cut pile area and the loop pile area under the illumination of the light source to segment and label the original image of the carpet surface into a first type of region and a second type of region. The first type of region corresponds to a high cut pile or flat cut pile area, and the second type of region corresponds to a low loop pile or flat loop pile area. Step S3, Pattern Mapping Step: Establish a printing coordinate mapping relationship based on the region segmentation results, assign a first set of digital patterns to the first type of region, and assign a second set of digital patterns to the second type of region. The first set of digital patterns and the second set of digital patterns are independent of each other. Step S4, Edge Compensation Step: Perform morphological processing on the region segmentation results, identify the boundary between the first type of region and the second type of region, and extend the boundary of the first type of region outward by a predetermined number of pixels along the boundary to form a compensation region; Step S5, Printing Step: Control the digital printhead to print the first set of digital patterns in the first type of area after edge compensation, and print the second set of digital patterns in the second type of area, according to the printing coordinate mapping relationship, so as to realize the positioning and overprinting of the carpet pile structure and the printed pattern.

[0009] By adopting the above technical solution: This solution utilizes the three-dimensional pile structure naturally formed during the weaving process of high-cut low-loop or flat-cut flat-loop carpet blanks (the yarns in the cut pile area are loose and upright, while the yarns in the loop pile area are closed loops). Under active light source illumination, the two types of areas exhibit fundamentally different light reflection characteristics. This difference is captured by a camera device and converted into a digital image. Then, through image segmentation, coordinate mapping, and zone printing, a closed-loop control of the entire chain of "identification-positioning-printing" of the carpet surface's three-dimensional structure is achieved.

[0010] For the first time, the three-dimensional structural features of the carpet surface itself are used as the benchmark for printing positioning, fundamentally avoiding the problems of pattern misalignment and ghosting caused by the deformation of the raw carpet in traditional full-coverage printing; it achieves precise registration of the printed pattern and the carpet surface texture, greatly improving the product yield; it lays the technical foundation for subsequent independent printing of different patterns in different areas, and greatly expands the design space of carpet patterns.

[0011] Preferably, the light source is a high-intensity incandescent light source with a color temperature of 5000K-6000K, a color rendering index ≥90, and an angle of 20-40 degrees between the light source irradiation direction and the normal of the carpet surface.

[0012] By employing the above technical solution: the high-intensity incandescent light source, with its full-spectrum and high color rendering index, can produce uniform and stable illumination on the carpet surface. The color temperature of 5000K-6000K, close to midday sunlight, ensures accurate reproduction of the color and luster of both cut pile and loop pile areas. An illumination angle of 20-40 degrees causes multiple diffuse reflections between the loose fibers in the cut pile area, forming bright, discrete light spots; while the dense, ring-shaped surface of the loop pile area produces relatively soft diffuse reflections, maximizing the contrast between the two. A color rendering index ≥90 ensures that the color information of the yarn itself is not distorted, facilitating subsequent image processing.

[0013] The difference in reflective properties between cut pile and loop pile is maximized, greatly improving the accuracy of image segmentation; the lighting is stable and the color temperature is accurate, avoiding recognition fluctuations caused by changes in ambient light; the light source parameters have been optimized to be applicable to a variety of common carpet yarn materials such as nylon, polyester, and wool.

[0014] Preferably, the difference in color reflectivity is specifically manifested as follows: the cut pile area, due to the yarn tips being cut open and in a loose, upright state, reflects light strongly, has high brightness, and dispersed light spots under light source illumination; the loop pile area, due to the yarn forming a closed loop with a dense surface, reflects light softly, has low brightness, and uniform light spots under light source illumination.

[0015] By employing the above technical solution: In the cut pile area, the yarn tips are cut, exposing the fiber end faces and causing them to spread outwards, forming a microstructure similar to a "brush." ​​When light shines at a certain angle, the fiber end faces and sidewalls together produce strong diffuse reflection, and due to the different fiber orientations, the reflected light spots are discretely distributed. In the loop pile area, the yarns are closed loops with a continuous and dense surface. When light shines on them, they mainly undergo uniform diffuse reflection, with lower reflection intensity and a smooth distribution. This physical optical difference is inherent to the carpet weaving structure and is unrelated to dye color, therefore it has universal applicability.

[0016] The identification method, based on physical structure rather than dye color, is not affected by the carpet background color and can accurately distinguish between cut pile and loop pile using yarn of the same color. It provides a stable and robust feature basis for image segmentation algorithms and is applicable to high-cut and low-loop pile carpets and flat-cut and flat-loop pile carpets of various colors, without the need for retraining or parameter adjustment for different patterns.

[0017] Preferably, in the region segmentation step, the color reflection characteristics of each pixel in the original carpet image are simulated and compared with a preset standard color library to calculate the approximation degree of each pixel belonging to cut pile or loop pile, thereby achieving segmentation.

[0018] By employing the above technical solution: A large number of carpet sample images of known categories (cut pile / loop pile) are pre-collected, and the HSV or LAB color space features of each pixel are extracted under standard lighting conditions to construct a standard color library. During actual production, for each pixel in the image to be identified, the Euclidean or Mahalanobis distance between its color feature vector and the centers of cut pile and loop pile samples in the standard color library is calculated. Based on the distance, a probability value is assigned to the pixel as either cut pile or loop pile. The final category is then determined through threshold judgment or maximum likelihood method. This color library-based simulation comparison method is essentially a nearest neighbor classification or Bayesian classification method in supervised learning. It is simple to implement, has a fast computation speed, and is suitable for real-time processing on high-paced production lines. The standard color library can be quickly updated and expanded according to different batches and carpet materials, exhibiting good adaptability. Compared to traditional human judgment or single-threshold segmentation, this method can effectively handle transition pixels at the boundary between cut pile and loop pile, resulting in smoother segmentation boundaries.

[0019] Preferably, the region segmentation step employs a deep learning-based semantic segmentation model for end-to-end pixel-level classification.

[0020] By employing the aforementioned technical solutions, deep learning semantic segmentation models (such as U-Net, DeepLab, and SegFormer) automatically learn multi-level features of cut pile and loop pile regions in carpet images, ranging from low-level (edges, textures) to high-level (semantic concepts), through multi-layer convolution and pooling operations. After training on a large dataset with pixel-level annotations, the model can directly input the original image and output a class probability map for each pixel, achieving end-to-end pixel-level classification. Its core lies in the fact that the convolutional kernel can capture the subtle differences between the unique scattered fiber texture of the cut pile region and the uniform texture of the loop pile region—differences that are difficult to quantify in traditional color spaces. The recognition accuracy is significantly higher than traditional methods, and the segmentation accuracy is greatly improved in challenging scenarios such as complex textures, uneven lighting, and similar yarn colors. No manual feature rule design is required; the model can automatically adapt to carpets of different materials and colors, exhibiting strong transferability. The end-to-end processing streamlines the system architecture, resulting in fast inference speed (a single image can be processed within 15-20 milliseconds), meeting the real-time requirements of high-speed production lines.

[0021] Preferably, in the pile treatment of the flat-cut and flat-circle carpet, the flat-cut area is equivalent to the first type of area, and the flat-circle area is equivalent to the second type of area.

[0022] By adopting the above technical solution: in flat-cut, flat-loop carpets, the yarns in the flat-cut area are also cut flat, exposing the fiber ends, and have similar optical reflection characteristics (strong reflection, high brightness) to the high-cut pile area; the yarns in the flat-loop area remain closed in a loop, consistent with the optical characteristics of the low-loop pile area (soft reflection, low brightness). Therefore, in the image segmentation algorithm, there is no need to distinguish between "high-cut" and "flat-cut," or "low-loop" and "flat-loop," only a unified identification of "cut type" and "loop type" is required. This equivalent processing is based on the homology of the two types of carpets in terms of microstructure and optical properties. The same set of image segmentation and printing control algorithms can be directly applied to both high-cut, low-loop, and flat-cut, flat-loop carpet types without separate development; it reduces the system's adaptation cost for different products and improves the equipment's versatility; when switching product types on the production line, there is no need to change the software configuration, making operation simple.

[0023] Preferably, the first set of digital patterns and the second set of digital patterns are each a continuous pattern arranged in a preset cycle unit, and the two are the same or different in terms of pattern content, color and style.

[0024] By adopting the above technical solution—designing the two sets of patterns as a continuous whole and arranged in cyclical units—it is to adapt to the continuous production characteristics of carpets. The carpet blank moves continuously during the printing process, therefore the printed patterns must be able to be infinitely repeated and spliced. The first and second sets of patterns are independent of each other; they can be different layers of the same style (such as foreground and background), or completely different styles (such as geometric shapes and floral patterns), or even one set can be patterned while the other is solid color. Both are designed, stored, and mapped independently, dynamically switching during printing based on real-time segmentation results. Designers can fully unleash their creativity, utilizing the independence and freedom of the two sets of patterns to achieve a three-dimensional, multi-layered visual effect of "one layer of background texture, two layers of pattern." It is possible to create differentiated products that traditional full-coverage printing cannot achieve, such as "one pattern in the cut pile area and another pattern in the loop pile area." The cycle period and starting position of the two sets of patterns can be set independently, further enriching the possibilities of carpet patterns.

[0025] Preferably, the predetermined number of pixels is 2-3 pixels, more preferably 2 pixels.

[0026] By adopting the above technical solution: In actual production, the yarns at the edges of the cut pile area often have a certain degree of flattening, and the annular structure at the edges of the loop pile area may also produce tiny gaps, resulting in a "blank" or "overlapping" area of ​​about 1-2 pixels wide at the boundary between the two types of areas in the segmentation mask. If printed directly according to the original segmentation mask, white areas or color overflow may appear at the boundary. Through morphological dilation, the boundary of the cut pile area (the first type of area) is extended outward by 2-3 pixels, which is equivalent to physically covering the uncertain area at the boundary with a narrow band. This extension width has been optimized experimentally: less than 2 pixels is insufficient to completely cover the defect area, while more than 3 pixels may cause the cut pile pattern to excessively penetrate into the loop pile area, affecting the visual effect. 2 pixels is the optimal value that balances integrity and clarity. It completely solves the printing seam problem at the junction of cut pile and loop pile, so that the two patterns can be smoothly transitioned and seamlessly spliced ​​at the junction; the algorithm is simple, the amount of calculation is minimal, and it does not increase the system latency; the verified 2-pixel extension value has good universality for different materials (nylon, polyester, wool) and different carpet densities.

[0027] A second aspect of this application provides a system for intelligently identifying and positioning overlapping prints on carpet pile surfaces with high-cut and low-loop, and flat-cut and flat-loop patterns, comprising: Image acquisition module: includes a camera device installed above the front end of the printing area of ​​the digital printing machine, and a light source installed on the same side or adjacent to the camera device, for capturing original images of the carpet surface containing cut pile area and loop pile area under the illumination of the light source; Image processing and segmentation module: used to receive the original image of the carpet surface, and use the difference in color reflection characteristics of the cut pile area and the loop pile area under the illumination of the light source to segment and mark the original image of the carpet surface into a first type of area and a second type of area; Pattern mapping and edge compensation module: It is used to establish printing coordinate mapping relationship based on the region segmentation result, assign the first set of digital patterns to the first type of region, assign the second set of digital patterns to the second type of region, and perform morphological processing on the region segmentation result to extend the boundary of the first type of region outward by a predetermined number of pixels to form a compensation region. Printing execution module: It is used to receive the processed area segmentation results and pattern data, and control the digital printhead to print the first set of digital patterns in the first type of area and the second set of digital patterns in the second type of area according to the coordinate mapping relationship during the movement of the carpet blank.

[0028] By adopting the above technical solution, this system integrates the light source, camera, image processing, pattern mapping, edge compensation, and printing execution into a closed-loop control unit. The image acquisition module uses an active light source to enhance the reflective difference between cut pile and loop pile, acquiring high-quality original images of the carpet surface; the image processing and segmentation module uses color matching or deep learning algorithms to accurately segment the carpet surface into cut pile and loop pile areas; the pattern mapping and edge compensation module dynamically generates printing coordinate mappings for the two patterns based on the segmentation results and performs pixel-level expansion compensation at the boundaries; the printing execution module controls the printhead to spray the corresponding ink in the corresponding areas in real time according to the mapping table. The four modules work together to form a complete technical chain of "perception-analysis-decision-execution". It achieves full automation from carpet surface structure recognition to zone printing without manual intervention; the modular design facilitates system maintenance and upgrades, such as replacing with a higher resolution camera or upgrading the deep learning model without affecting other modules; it can be easily integrated into existing digital printing production lines with low modification costs; compared with traditional full-coverage printing systems, this system can produce differentiated carpet products with higher added value.

[0029] Preferably, the pattern mapping and edge compensation module extends the boundary of the first type of region outward by a predetermined number of pixels, which is 2-3 pixels, preferably 2 pixels.

[0030] By adopting the above technical solution: the pattern mapping and edge compensation module maintains a mask matrix of the same size as the carpet image. The initial mask is generated from the image segmentation results (cut-pile areas are 1, loop-pile areas are 0). The edge compensation submodule performs morphological dilation on this mask, using a circular or square kernel with a radius of 2 pixels as the structuring element. In the dilated mask, the boundary of the original cut-pile area is pushed outward by 2 pixels. These newly added pixels, which originally belonged to the loop-pile area, are now re-marked as cut-pile areas. During actual printing, the printing execution module performs ink jetting according to the dilated mask, thus forming a "transition band" of about 2 pixels wide at the boundary. The pattern within this transition band remains consistent with the cut-pile area. The hardware (module) implementation method is consistent with the effect of the aforementioned method's optimized features, ensuring the integrity and coordination of the system; the edge compensation function is solidified within the module, reducing the computational burden on the main controller; the module's output can directly drive the printhead, resulting in a fast response speed that meets the requirements of high-speed printing; the optimized value of 2 pixels has been verified through extensive experiments and represents the best balance between cost and effect.

[0031] Compared with the prior art, the present invention has the following outstanding substantive features and significant progress: Achieving Decoupling and Precise Overprinting of Structure and Pattern: This invention, for the first time, uses the "difference in reflective physical properties between cut pile and loop pile under a specific light source" as the identification criterion, and combines it with high-speed vision and digital image segmentation technology to fundamentally solve the problem that existing technologies cannot distinguish the three-dimensional structure of carpet surfaces. By partitioning, positioning, and independently printing different patterns, a perfect fusion or precise comparison between the printed pattern and the original three-dimensional texture of the carpet is achieved, eliminating ghosting and misalignment defects. This is a key technological breakthrough in this field.

[0032] Innovative pixel-level edge compensation ensures printing integrity: Addressing the minor gaps or overlaps (typically 1-2 pixels) caused by the flattening of cut pile yarns and looped pile structures at their boundaries in actual production, a compensation algorithm is creatively proposed to "extend the boundary of the cut pile area outward by 2-3 pixels." This simple yet efficient step effectively solves the printing seam problem caused by physical structure, ensuring the continuity and integrity of complex patterns at the boundaries of the three-dimensional pile surface, and significantly improving the artistic quality of the final product.

[0033] Enhancing design freedom and product added value: This method allows designers to independently design two or more sets of patterns for the "high / flat cut" and "low / flat loop" sections of the carpet surface, achieving a three-dimensional, multi-layered visual effect of "one layer of background texture and two layers of pattern." It transforms passive design of "covering up misalignment" into active design of "utilizing structure," greatly expanding the design space for tufted carpet patterns.

[0034] Robust to various yarn materials and deformations: The identification method of this invention is based on physical reflective properties and can work effectively on yarn materials with different ink absorption and gloss levels, such as nylon, modified polyester, pure polyester, and wool. It can also automatically adapt to the stretching and deformation of the raw blanket during the production process. Attached Figure Description

[0035] The present application will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0036] Figure 1 This is a photograph of the actual high-cut, low-loop tufted carpet blank to be processed in an embodiment of the present invention (illustrating the three-dimensional pile structure).

[0037] Figure 2 This is a schematic diagram of the area captured by a high-speed camera and segmented by software in an embodiment of the present invention. The dark area is the first type of area (cut pile area A), and the light area is the second type of area (loop pile area B).

[0038] Figure 3 This is a schematic diagram of partitioned printing in an embodiment of the present invention, showing the effect of printing a foreground pattern in a first type of region (A) and a background pattern in a second type of region (B).

[0039] Figure 4 This is a physical image of a high-cut, low-circle carpet after the final positioning and overprinting of an embodiment of the present invention, demonstrating the visual effect of precise composite of pattern and background. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0041] Example 1 (Basic Implementation): This embodiment provides a method for intelligently identifying and positioning overlapping prints on high-cut, low-loop carpet pile surfaces. The specific steps are as follows: I. Equipment Layout and Image Acquisition like Figure 1 As shown, a high-speed area array CCD camera (resolution no less than 5 megapixels, frame rate ≥100fps) is installed above the feed section of a wide-format digital printing machine. Two sets of high-power LED incandescent light sources (color temperature 5500K, color rendering index ≥95) are symmetrically installed on both sides of the camera, with the light source direction forming a 30-degree angle with the carpet surface normal. The nylon high-cut, low-loop carpet blank to be printed moves at a constant speed of 0.5m / s. When the blank carpet passes the shooting station, the light source is triggered to illuminate at maximum power, and simultaneously the camera captures an RGB image containing the complete carpet surface texture.

[0042] Principle Explanation: A high-intensity incandescent light source (color temperature 5000K-6000K, color rendering index ≥90) is used with an illumination angle of 20-40 degrees to maximize the difference in reflective characteristics between the cut pile and loop pile areas. In the cut pile area, the yarn tips are cut open and stand loosely upright. Under oblique illumination, the fiber end faces and sidewalls produce strong diffuse reflection, forming bright, discrete light spots. In the loop pile area, the yarn is closed and dense, producing relatively soft diffuse reflection with low and uniform brightness. The contrast between the two is maximized, facilitating subsequent image segmentation.

[0043] Effects: ① Maximizes the difference in reflective properties between cut pile and loop pile, improving image segmentation accuracy to over 95%; ② Stable lighting and accurate color temperature avoid recognition fluctuations caused by changes in ambient light; ③ Optimized light source parameters make it suitable for various common carpet yarn materials such as nylon, polyester, and wool.

[0044] II. Intelligent segmentation of velvet areas (based on color and texture features) The captured images are transmitted to an embedded image processing industrial control computer. The computer is pre-installed with segmentation algorithms based on color and texture features. Specifically, a large number of histograms of the HSV color space distribution of cut and looped textured regions under standard lighting are pre-calculated. For each pixel of the input image, its hue (H), saturation (S), and brightness (V) values ​​are calculated and compared with the pre-stored cut and looped texture feature models using Bayesian classification. Since the V value (brightness) of the cut textured region is significantly higher than that of the looped textured region, the classifier can accurately determine the classification of each pixel. Finally, a binary mask image of the same size as the original image is generated, where white (pixel value 255) represents the cut textured region (A), and black (pixel value 0) represents the looped textured region (B). The recognition results are as follows: Figure 2 As shown, Figure 2 The dark-colored block 2 is the identified cut-pile area, and the light-colored background 3 is the looped-pile area.

[0045] Principle Explanation: Due to the loose fibers, the cut pile area reflects light strongly, has high brightness, and dispersed light spots under a light source; the loop pile area, due to its dense surface, reflects light softly, has low brightness, and uniform light spots. Based on this difference in color reflection characteristics, by comparing with a standard color library simulation (i.e., a pre-stored feature model), the assignment of each pixel can be accurately calculated, the segmentation boundaries are smooth, and the processing effect on transition pixels is good.

[0046] Results: ① Simple to implement and fast to calculate, suitable for real-time processing on high-speed production lines; ② The standard color library can be quickly updated and expanded according to different batches and different materials of carpet; ③ Compared with traditional human eye judgment or single threshold segmentation, this method can effectively handle the transition pixels at the junction of cut pile and loop pile, and the segmentation boundary is smoother.

[0047] III. Partition Pattern Mapping and Edge Compensation The designers pre-designed two sets of patterns: the foreground background pattern X is a continuous, geometrically aesthetic diamond grid pattern (such as...). Figure 3 (As shown in Mark 4); the background floral pattern Y is a continuous, light-colored vine pattern (such as...) Figure 3 (As shown in label 5). Each of the two sets of patterns is a continuous whole and arranged in a preset cycle unit, and the content, color and style can be designed independently.

[0048] The printing control system reads the binary mask image and establishes a printing coordinate mapping: for coordinate points with a mask value of 255, the corresponding pixel of pattern X is printed; for coordinate points with a mask value of 0, the corresponding pixel of pattern Y is printed.

[0049] To avoid white areas appearing at the boundary between cut pile and loop pile (typically 1-2 yarn diameters wide) due to yarn collapse or loop structures, the system performs a morphological dilation operation on the mask image. Specifically, a circular structuring element with a radius of 2 pixels is used to dilate the white area (cut pile area A), extending its boundary outward by 2 pixels. The dilated mask is then used for final printing control.

[0050] Explanation of the principle: The first and second patterns are independent of each other, corresponding to the cut pile area and the loop pile area respectively. This allows designers to fully unleash their creativity and achieve a three-dimensional, multi-layered visual effect of "one layer of background texture and two layers of pattern." Regarding edge compensation, in actual production, the falling of yarns at the edges of the cut pile area and the gaps in the ring structure of the loop pile area can cause approximately 1-2 pixels of blank space or overlap at the junction of the segmented masks. By extending the boundary of the cut pile area outward by 2-3 pixels (preferably 2 pixels), the uncertain area at the junction can be physically covered. 2 pixels is the optimal value: less than 2 pixels is insufficient to cover defects, and more than 3 pixels will cause excessive pattern intrusion.

[0051] Effects Description: ① Designers can independently design two sets of patterns, creating differentiated products that traditional all-over printing cannot achieve; ② It completely solves the printing seam problem at the junction of cut pile and loop pile, allowing the two sets of patterns to transition smoothly and seamlessly at the junction; ③ The algorithm is simple, with minimal computation and does not increase system latency; ④ The verified 2-pixel extension value has good universality for different materials and different carpet densities.

[0052] IV. Positioning and Overprinting The processed mask and two sets of pattern data are sent to the printing machine. The printing machine is equipped with a high-precision encoder that tracks the carpet displacement in real time, ensuring the print head is at every physical coordinate and accurately sprays the corresponding ink based on the mask value at the current coordinate. Ultimately, on the finished carpet, we see a clear geometric grid pattern in the diamond-shaped cut pile area, while the surrounding loop pile area presents a continuous vine pattern. The two are perfectly joined at the junction, without misalignment or gaps, creating a strong sense of three-dimensionality and artistry (e.g., Figure 4 (As shown).

[0053] Summary of Results: This embodiment completely solves the industry problem of misalignment and ghosting in high-cut, low-circle carpet printing by using a technical chain of "active light source to enhance reflective differences - standard color library segmentation - independent mapping of zones - pixel-level edge compensation - real-time printing". It achieves precise composite of patterns and backgrounds, and increases the product yield from about 75% in traditional methods to over 96%.

[0054] Example 2 (Implementation of Semantic Segmentation Based on Deep Learning): This embodiment is basically the same as Embodiment 1, except that the region segmentation in step S2 uses a semantic segmentation model based on deep learning to perform end-to-end pixel-level classification, so as to further improve the recognition accuracy and robustness to complex textures.

[0055] II. Intelligent Segmentation of Textured Regions (Based on Deep Learning Semantic Segmentation Model) The original carpet image (512×512 pixel RGB image) obtained in step S1 is input into a pre-trained deep learning semantic segmentation network. This embodiment uses a U-Net network structure, which includes an encoder path and a decoder path. The encoder path consists of four downsampling blocks, each containing two 3×3 convolutional layers (each followed by a ReLU activation function and a batch normalization layer) and a 2×2 max pooling layer. After four downsampling passes, the feature map size gradually decreases from 512×512 to 32×32, and the number of channels gradually increases from 64 to 512.

[0056] The decoder path consists of four upsampling blocks. Each upsampling block contains a 2×2 upsampling convolutional layer (doubling the feature map size), a feature concatenation layer (concatenating the upsampled feature map with the cropped feature map from the corresponding encoder layer along the channel dimension), and two 3×3 convolutional layers (each followed by a ReLU activation function and a batch normalization layer). After four upsampling passes, the feature map is restored to a size of 512×512.

[0057] Output layer: The last 1×1 convolutional layer maps the 64-channel feature map to a 2-channel output (corresponding to the classification probabilities of cut-textured and looped textured regions, respectively). After passing through the Softmax activation function, the probability distribution of each pixel belonging to the cut-textured or looped textured region is obtained. Finally, the category with the highest probability is taken as the segmentation result of the pixel, generating a binary mask image (cut-textured regions are marked as 1, and looped textured regions are marked as 0).

[0058] Model training process: Collect training dataset: Collect 1000 images of carpet surfaces with different lighting conditions, different patterns, and different yarn materials (nylon, polyester, wool, etc.) with high cut and low loop, and flat cut and flat loop. Each image is manually annotated at the pixel level to accurately mark the boundaries of the cut pile area and the loop pile area.

[0059] The loss function used is the cross-entropy loss function, as shown in the following formula: ; in, Total number of pixels Category (1 is the cut pile area, 2 is the loop pile area). The true label (one-hot encoded) for the i-th pixel. Let be the probability predicted by the model that the i-th pixel belongs to category c.

[0060] The optimizer used was the Adam optimizer, with an initial learning rate of 0.001, a decay factor of 0.9, a batch size of 8, and 50 training epochs. After training, the model achieved a pixel accuracy of over 98.5% and a mIoU of 95.2% on the validation set.

[0061] Segmented execution process: During real-time production, after the camera captures the original image of the carpet surface, the image is scaled to 512×512 pixels and input into the trained U-Net model. The model performs inference on a GPU (such as an NVIDIA Tesla T4), with an inference time of approximately 15-20 milliseconds per image, which meets the real-time processing requirements at a carpet movement speed of 0.5 m / s. The binary mask image output by the model is directly used for subsequent pattern mapping and edge compensation steps.

[0062] Explanation of the principle: Deep learning models can automatically learn the differences between cut-pile and loop-pile regions in terms of multi-level features such as texture, edges, and local structure, rather than relying solely on a single brightness threshold. The subtle differences between the scattered fiber texture of the cut-pile region and the uniform texture of the loop-pile region are difficult to quantify in traditional color spaces, but convolutional neural networks can effectively capture them.

[0063] Results Description: Compared with the segmentation method based on a standard color library in Example 1, this example has the following advantages: significantly higher recognition accuracy, with a segmentation accuracy of over 98% even in challenging scenarios such as complex textures, uneven lighting, and similar yarn colors; no need for manual design of feature rules, the model can automatically adapt to carpets of different materials and colors, and has strong transferability; the end-to-end processing flow simplifies the system architecture, and the inference speed is fast, meeting the real-time requirements of high-speed production lines.

[0064] Example 3 (Application of flat-cut, flat-circle carpet): This embodiment is basically the same as Embodiment 1, except that the type of carpet to be printed is a flat-cut, flat-loop carpet. In the pile processing of a flat-cut, flat-loop carpet, the flat-cut area is equivalent to the first type of area (A), and the flat-loop area is equivalent to the second type of area (B). The specific processing procedure of this embodiment is as follows: The yarn tips in the flat-cut area are cut flat and stand loosely upright, reflecting light strongly and brightly under high-intensity light source illumination. The yarn in the flat loop area is in a closed loop shape with a dense surface. Under high-intensity light source illumination, it reflects light softly, has low brightness, and uniform light spots.

[0065] Based on the aforementioned differences in reflective properties, the image segmentation module can accurately distinguish between the flat cut area and the flat circle area. Then, it prints the first set of digital patterns in the flat cut area and the second set of digital patterns in the flat circle area, thereby achieving the positioning and overprinting of the flat cut and flat circle carpet.

[0066] Explanation of the principle: In flat-cut, flat-loop carpets, the flat-cut area and the high-cut pile area have similar optical reflection characteristics (strong reflection, high brightness); the flat-loop area and the low-loop pile area have the same optical characteristics (soft reflection, low brightness). Therefore, in the image segmentation algorithm, there is no need to distinguish between "high-cut" and "flat-cut," or "low-loop" and "flat-loop," only to uniformly identify the "cut type" and "loop type." This equivalent processing is based on the homology of the two types of carpets in terms of microstructure and optical properties.

[0067] Results: The same image segmentation and printing control algorithm can be directly applied to two mainstream carpet types, high-cut low-circle and flat-cut flat-circle, without the need for separate development; it reduces the system's adaptation cost to different products and improves the equipment's versatility; there is no need to change the software configuration when switching product types on the production line, making operation simple.

[0068] Example 4 (Adaptability to different yarn materials): This embodiment verifies the adaptability of the method of the present invention to different yarn materials. High-cut, low-loop carpet blanks of four materials—nylon, modified polyester, pure polyester, and wool—were used, and positioning and overprinting were performed according to the method of Example 1. Experimental results show that: Under 5500K incandescent light, the cut pile and loop pile areas of the four materials all showed significant differences in reflective properties, greatly improving the recognition accuracy. Since different materials have different ink absorption properties, in the edge compensation step, for wool materials with strong ink absorption, adjusting the number of boundary expansion pixels to 3 pixels can achieve a better overprinting effect. All materials achieved precise composite of printed patterns and backgrounds, with no ghosting or white gaps, verifying the material universality of the method of this invention.

[0069] Principle Explanation: The recognition method of this invention is based on physical reflective properties (the different light reflection behaviors of the yarn end face and the annular surface), rather than relying on the color or chemical properties of a specific dye. Therefore, it has inherent adaptability to yarns of different materials (nylon, polyester, wool, etc.). The number of edge compensation pixels can be finely adjusted according to the ink absorption of the material to optimize the effect at the junction.

[0070] Effect description: This invention can work effectively on yarn materials with different ink absorption and gloss levels, such as nylon, modified polyester, pure polyester, and wool, and can automatically adapt to the stretching and deformation of the raw blanket during the production process, with excellent material versatility and deformation robustness.

[0071] Comparison (traditional full-coverage printing method): Using a traditional high-definition digital inkjet printer, the same batch of high-cut, low-ring blank blankets can be printed directly as a whole. Figure 3 The foreground pattern X. The results showed that, due to the stretching deformation of the raw carpet itself, the printed diamond grid could not completely overlap with the diamond pattern formed by the cut pile loops of the raw carpet, resulting in numerous instances of grid line misalignment and double contours, creating a visually chaotic appearance. This indicates that existing technologies cannot solve the technical problem addressed by this invention.

[0072] The above embodiments can all be configured with a system for intelligently identifying high-cut and low-loop, flat-cut and flat-loop carpet pile positioning and overprinting, including: Image acquisition module: includes a camera device installed above the front end of the printing area of ​​the digital printing machine, and a light source installed on the same side or adjacent to the camera device, for capturing original images of the carpet surface containing cut pile area and loop pile area under the illumination of the light source; Image processing and segmentation module: used to receive the original image of the carpet surface, and use the difference in color reflection characteristics of the cut pile area and the loop pile area under the illumination of the light source to segment and mark the original image of the carpet surface into a first type of area and a second type of area; Pattern mapping and edge compensation module: It is used to establish printing coordinate mapping relationship based on the region segmentation result, assign the first set of digital patterns to the first type of region, assign the second set of digital patterns to the second type of region, and perform morphological processing on the region segmentation result to extend the boundary of the first type of region outward by a predetermined number of pixels to form a compensation region. Printing execution module: It is used to receive the processed area segmentation results and pattern data, and control the digital printhead to print the first set of digital patterns in the first type of area and the second set of digital patterns in the second type of area according to the coordinate mapping relationship during the movement of the carpet blank.

[0073] This system integrates light source, camera, image processing, pattern mapping, edge compensation, and printing execution into a closed-loop control unit. The image acquisition module utilizes an active light source to enhance the reflective difference between cut and loop pile, acquiring high-quality original images of the carpet surface. The image processing and segmentation module employs color matching or deep learning algorithms to precisely segment the carpet surface into cut and loop pile areas. The pattern mapping and edge compensation module dynamically generates printing coordinate mappings for the two patterns based on the segmentation results and performs pixel-level expansion compensation at the boundaries. The printing execution module controls the printhead to spray the corresponding ink in the corresponding areas in real time according to the mapping table. These four modules work collaboratively, forming a complete technology chain of "perception—analysis—decision—execution." This achieves full automation from carpet surface structure recognition to zoned printing, requiring no manual intervention. The modular design facilitates system maintenance and upgrades; for example, replacing the camera with a higher resolution one or upgrading the deep learning model does not affect other modules. It can be easily integrated into existing digital printing production lines with low modification costs. Compared to traditional full-coverage printing systems, this system can produce differentiated carpet products with higher added value.

[0074] The pattern mapping and edge compensation module extends the boundary of the first type of region outward by a predetermined number of pixels, which is 2-3 pixels, preferably 2 pixels.

[0075] The pattern mapping and edge compensation module internally maintains a mask matrix the same size as the carpet image. The initial mask is generated from the image segmentation results (cut-pile regions are 1, loop-pile regions are 0). The edge compensation submodule performs morphological dilation on this mask, using a circular or square kernel with a radius of 2 pixels as the structuring element. In the dilated mask, the boundary of the original cut-pile region is pushed outward by 2 pixels. These newly added pixels, originally belonging to the loop-pile region, are now re-marked as cut-pile regions. During actual printing, the printing execution module jets ink according to the dilated mask, thus forming a "transition band" approximately 2 pixels wide at the boundary. The pattern within this transition band remains consistent with the cut-pile region. The hardware (module) implementation is consistent with the effect of the aforementioned method for optimizing features, ensuring the integrity and coordination of the system; embedding the edge compensation function within the module reduces the computational burden on the main controller; the module's output can directly drive the printhead, resulting in a fast response speed that meets the requirements of high-speed printing; the optimal value of 2 pixels has been verified through extensive experiments and represents the best balance between cost and performance.

[0076] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligently identifying and positioning overlapping prints on the pile surface of carpets with high-cut low-loop and flat-cut flat-loop patterns, characterized in that... Includes the following steps: Step S1, Image Acquisition Step: At the front end of the printing area of ​​the digital printing machine, the carpet surface of the carpet blank to be printed (high-cut low-loop or flat-cut flat-loop) is illuminated by a light source, and the original image of the carpet surface containing the cut pile area and the loop pile area is captured by a camera device. Step S2, Region Segmentation Step: Obtain the original image of the carpet surface. Utilize the difference in color reflection characteristics between the cut pile area and the loop pile area under the illumination of the light source to segment and label the original image of the carpet surface into a first type of region and a second type of region. The first type of region corresponds to a high cut pile or flat cut pile area, and the second type of region corresponds to a low loop pile or flat loop pile area. Step S3, Pattern Mapping Step: Establish a printing coordinate mapping relationship based on the region segmentation results, assign a first set of digital patterns to the first type of region, and assign a second set of digital patterns to the second type of region. The first set of digital patterns and the second set of digital patterns are independent of each other. Step S4, Edge Compensation Step: Perform morphological processing on the region segmentation results, identify the boundary between the first type of region and the second type of region, and extend the boundary of the first type of region outward by a predetermined number of pixels along the boundary to form a compensation region; Step S5, Printing Step: Control the digital printhead to print the first set of digital patterns in the first type of area after edge compensation, and print the second set of digital patterns in the second type of area, according to the printing coordinate mapping relationship, so as to realize the positioning and overprinting of the carpet pile structure and the printed pattern.

2. The method according to claim 1, characterized in that, The light source is a high-intensity incandescent light source with a color temperature of 5000K-6000K, a color rendering index of ≥90, and an angle of 20-40 degrees between the direction of the light source and the normal of the carpet surface.

3. The method according to claim 1, characterized in that, The specific differences in color reflectivity are as follows: the cut pile area, due to the yarn tips being cut open and in a loose, upright state, reflects light strongly, has high brightness, and dispersed light spots under light source illumination; the loop pile area, due to the yarn forming a closed loop and having a dense surface, reflects light softly, has low brightness, and uniform light spots under light source illumination.

4. The method according to claim 1, characterized in that, In the region segmentation step, the color reflection characteristics of each pixel in the original carpet image are simulated and compared with a preset standard color library to calculate the approximation degree of each pixel belonging to cut pile or loop pile, thereby achieving segmentation.

5. The method according to claim 1, characterized in that, The region segmentation step employs a deep learning-based semantic segmentation model for end-to-end pixel-level classification.

6. The method according to claim 1, characterized in that, In the pile treatment of the flat-cut and flat-circle carpet, the flat-cut area is equivalent to the first type of area, and the flat-circle area is equivalent to the second type of area.

7. The method according to claim 1, characterized in that, The first set of digital patterns and the second set of digital patterns are each a continuous pattern arranged in a preset cyclic unit. The two patterns may be the same or different in terms of content, color, and style.

8. The method according to claim 1, characterized in that, The predetermined number of pixels is 2-3 pixels, preferably 2 pixels.

9. A system for intelligently identifying and positioning overlapping prints on carpet pile surfaces with high-cut and low-loop patterns, characterized in that, include: Image acquisition module: includes a camera device installed above the front end of the printing area of ​​the digital printing machine, and a light source installed on the same side or adjacent to the camera device, for capturing original images of the carpet surface containing cut pile area and loop pile area under the illumination of the light source; Image processing and segmentation module: used to receive the original image of the carpet surface, and use the difference in color reflection characteristics of the cut pile area and the loop pile area under the illumination of the light source to segment and mark the original image of the carpet surface into a first type of area and a second type of area; Pattern mapping and edge compensation module: It is used to establish printing coordinate mapping relationship based on the region segmentation result, assign the first set of digital patterns to the first type of region, assign the second set of digital patterns to the second type of region, and perform morphological processing on the region segmentation result to extend the boundary of the first type of region outward by a predetermined number of pixels to form a compensation region. Printing execution module: It is used to receive the processed area segmentation results and pattern data, and control the digital printhead to print the first set of digital patterns in the first type of area and the second set of digital patterns in the second type of area according to the coordinate mapping relationship during the movement of the carpet blank.

10. The system according to claim 9, characterized in that, The pattern mapping and edge compensation module extends the boundary of the first type of region outward by a predetermined number of pixels, which is 2-3 pixels, preferably 2 pixels.

Images