CN-115496717-B - Digital analysis method for textile fabric

CN115496717BCN 115496717 BCN115496717 BCN 115496717BCN-115496717-B

Abstract

The invention discloses a textile fabric digital analysis method which comprises the steps of collecting high-definition images on the front side and the back side of a textile fabric, carrying out gray level treatment, extracting an ROI initial selected area of fabric textures based on the gray level images, obtaining an ROI initial selected area replication body, determining offset of the replication body relative to the initial ROI area through a template matching method, further obtaining an ROI correction selected area as a textile fabric digital analysis basic unit, utilizing a trained deep learning yarn structure image identification model to identify a yarn structure to obtain a mark image and determine the type of the textile fabric, mapping the yarn structure on the mark image onto an upper weaving pattern template corresponding to the type of the fabric to obtain an upper weaving schematic drawing of the fabric.

Inventors

LI NANA
GUO DAN
ZHANG XIAODONG

Assignees

天津工业大学

Dates

Publication Date: 20260512
Application Date: 20220915

Claims (5)

1. The digital analysis method of the textile fabric is characterized by comprising the following steps of: s1, collecting a front high-definition image and a back high-definition image of a textile fabric in a plane unfolding state, wherein the high-definition images at least comprise three complete pattern circulation units, and the resolution of fabrics in the high-definition images is at least 0.1mm; S2, carrying out graying treatment on the high-definition image acquired in the step S1 to obtain a gray image, and extracting to obtain an ROI initial selected region of the fabric texture in the image based on a traditional image processing algorithm; S3, obtaining an initial region of the ROI in the gray level image to obtain a duplicate, superposing the duplicate on the gray level image obtained in the step S2, and performing translation, scaling or rotation on the duplicate by a template matching method to determine the offset of the duplicate relative to the initial ROI; s4, correcting the initial ROI selection area in the image obtained in the step S2 according to the offset in the length direction and the offset in the width direction obtained in the step S3 to obtain an ROI correction selection area; S5, inputting the textile fabric digital analysis basic unit into a trained deep learning yarn structure image recognition model to perform image recognition on the yarn structure of the textile fabric, obtaining mark images with different yarn structures selected in a frame mode and the names of the yarn structures displayed on the frame selection frame, and obtaining the type of the textile fabric; S6, based on the type of the textile fabric obtained in the step S5, calling an upper weaving pattern template corresponding to the type of the textile fabric, scaling the upper weaving pattern template and the marked image obtained in the step S5 to the same size, and mapping the yarn structure knots identified in the marked image onto the upper weaving pattern template in a one-to-one correspondence manner, so that an upper weaving schematic diagram of the fabric is obtained.
2. The method for digitally analyzing textile fabric according to claim 1, wherein in step S1, the high-definition image is acquired by an image acquisition system comprising two cameras (1) equipped with high-definition industrial lenses, a fabric clamping frame (2) for clamping fabric, a light source (3) and a camera bracket (5), The fabric clamping frame (2) is composed of a horizontal moving mechanism, a lifting moving mechanism and a fabric clamping frame which are sequentially arranged from bottom to top, the bottom of the horizontal moving mechanism is fixed on the ground, the lifting moving mechanism is fixed on a sliding block of the horizontal moving mechanism, the fabric clamping frame is vertically fixed on the top of the lifting moving mechanism, the fabric clamping frame is enabled to reciprocate along the length direction of the fabric clamping frame through the horizontal moving mechanism and reciprocate along the height direction of the fabric clamping frame through the lifting moving mechanism, two cameras (1) provided with high-definition industrial lenses are symmetrically arranged on two sides of the fabric clamping frame (2) through camera supports (5) fixed on the ground, the lenses of each camera (1) are horizontally arranged towards a textile fabric (4) so as to realize that high-definition images acquired by the two cameras are respectively a front image and a back image of the same area of the textile fabric, and two light sources (3) are respectively symmetrically arranged on two sides of the fabric clamping frame (2), and each light source faces the textile fabric on the fabric clamping frame (2) and is obliquely downwards arranged in a 45-degree mode.
3. The method for digitally analyzing the textile fabric according to claim 2, wherein the textile fabric clamping frame is a rectangular frame which is divided into two parts in a split way, so that the textile fabric (4) can be clamped between the two parts divided into the rectangular frame and fixed into a whole through a clamp, and the textile fabric (4) can realize image acquisition in a state of being unfolded in a plane and freely drooping.
4. The method according to claim 1, wherein in step S3, the specific step of determining the offset of the replica with respect to the initial ROI area is: S301, copying an ROI initial selected region part in the image with the ROI initial selected region marking frame obtained in the step S2 to obtain an ROI selected region copying body; S302, superposing the replication body on an ROI initial region selection marking frame of the image obtained in the step S2, and finding a position which corresponds to and coincides with the texture of the replication body at an adjacent position in the gray level image by translating the replication body leftwards or rightwards along the length direction of the initial ROI region selection direction, wherein the offset of the replication body in the length direction of the initial ROI region selection direction is the optimal length of a basic unit; s303, upwards translating or downwards translating the replication body along the width direction of the initial ROI selection direction, and finding out the position which can correspond to and coincide with the texture of the replication body at the adjacent position in the gray level image, wherein at the moment, the offset of the replication body in the width direction of the initial ROI selection direction is the optimal width of the basic unit.
5. The method for digitally analyzing textile fabric according to claim 1, wherein the step S5 is specifically performed as follows: s501, constructing a deep learning yarn structure image recognition model and training, (1) The deep learning yarn structure image recognition model is based on a deep learning network and is a model trained by a large number of different types of yarn structure samples, wherein the deep learning network is selected from a Faster R-CNN network, a LeNet network, a AlexNet network, a ZFNet network, a VGGNet network, a GoogLeNet network, a ResNet network or a YOLOV5 network; (2) The training method of the deep learning yarn structure image recognition model comprises the following steps: The method comprises the steps of constructing a yarn structure training sample, namely respectively acquiring two-dimensional yarn structure images of a left open coil, a right open coil, a left closed coil, a complete coil, a tuck coil, a left loop transferring coil, a right loop transferring coil, a non-woven coil, a warp intersection point and a weft intersection point as input images, respectively drawing rectangular frames outside each two-dimensional yarn structure image and displaying yarn structure names on the rectangular frames to form a mark image as an output result image, wherein each two-dimensional yarn structure image is not less than 2000, and further acquiring the yarn structure training sample containing not less than 200000 mark images; Model training, namely taking a two-dimensional yarn structure image as input information, setting a rectangular frame outside the yarn structure image and displaying a marking image of a yarn structure name on the rectangular frame as an output result, sequentially inputting a yarn structure training sample into a designated deep learning network to complete training and optimization of the deep learning network and obtain a deep learning yarn structure image recognition model; s502, inputting the textile fabric digital analysis basic unit obtained in the step S4 into a deep learning yarn structure image recognition model to obtain a mark image, wherein a rectangular frame is sequentially arranged on the outer side of each yarn structure on the image, and the name of the yarn structure is displayed on the rectangular frame; S503, determining the type of the textile fabric according to the yarn structure names marked on the marked image, wherein the textile fabric comprises warp knitting fabric, weft knitting fabric and woven fabric.

Description

Digital analysis method for textile fabric Technical Field The invention relates to the technical field of textile fabric analysis and processing, in particular to a textile fabric sample analysis method. Background Textile fabrics are important components in daily life and industrial application, and comprise three major categories of woven fabrics, knitted fabrics and non-woven fabrics. The weave structure of the fabric is the inherent property of the fabric formed by knitting yarns, and different weave structures endow the fabric with different styles and appearances. In the current textile industry, the fabric weave structure is usually analyzed by manually disassembling single yarns, recording the threading or sinking and floating rule among the yarns, determining the knitting action of the yarns at the position according to the bending morphology of the yarns, and finally recording the analysis result. The analysis process generally needs 10-60 minutes according to different complexity degrees of the fabric, has the problems of low working efficiency, serious influence by working experience and subjective state of workers and the like, and is not suitable for the development demands of enterprises with high quality and high speed. The image processing technology has the characteristics of high processing speed, high accuracy and the like, and is widely applied to the fields of medicine, engineering, construction, other manufacturing industries and the like along with continuous progress and development of modern technology. The Chinese patent application CN201511003862.0 provides a woven fabric tissue map identification method capable of identifying the three-primary structure of woven fabric and the simple woven fabric tissue map with a changed tissue structure, and the Chinese patent application CN201811277402.0 relates to a woven fabric tissue structure analysis method and system, which have a good identification effect on a single tissue region in a fabric image and still have difficulty in identifying other types of fabrics and composite tissues. Therefore, development of a digital analysis method capable of rapidly acquiring woven patterns of various textile fabrics is urgently needed. Disclosure of Invention The invention aims to provide a textile fabric digital analysis method capable of realizing digital analysis on different types of textile fabrics and obtaining corresponding on-machine knitting patterns. For this purpose, the technical scheme of the invention is as follows: A digital analysis method for textile fabric comprises the following specific implementation steps: s1, collecting a front high-definition image and a back high-definition image of a textile fabric in a plane unfolding state, wherein the high-definition images at least comprise three complete pattern circulation units, and the resolution of fabrics in the high-definition images is at least 0.1mm; S2, carrying out graying treatment on the high-definition image acquired in the step S1 to obtain a gray image, and extracting to obtain an ROI initial selected region of the fabric texture in the image based on a traditional image processing algorithm; S3, obtaining an initial region of the ROI in the gray level image to obtain a duplicate, superposing the duplicate on the gray level image obtained in the step S2, and performing translation, scaling or rotation on the duplicate by a template matching method to determine the offset of the duplicate relative to the initial ROI; s4, correcting the initial ROI selection area in the image obtained in the step S2 according to the offset in the length direction and the offset in the width direction obtained in the step S3 to obtain an ROI correction selection area; S5, inputting the textile fabric digital analysis basic unit into a trained deep learning yarn structure image recognition model to perform image recognition on the yarn structure of the textile fabric, obtaining mark images with different yarn structures selected in a frame mode and the names of the yarn structures displayed on the frame selection frame, and obtaining the type of the textile fabric; S6, based on the type of the textile fabric obtained in the step S5, calling an upper weaving pattern template corresponding to the type of the textile fabric, scaling the upper weaving pattern template and the marked image obtained in the step S5 to the same size, and mapping the yarn structure knots identified in the marked image onto the upper weaving pattern template in a one-to-one correspondence manner, so that an upper weaving schematic diagram of the fabric is obtained. Wherein, the pattern plate is a pattern plate for upper knitting of weft knitting fabric shown in fig. 3 (a), the pattern plate is a pattern plate for upper knitting of warp knitting fabric shown in fig. 3 (b), and the pattern plate is a pattern plate for upper knitting of woven fabric shown in fig. 3 (c). Further, in step S1, the high-definition image