CN-122023362-A - Intelligent fabric shearing property prediction method and device for clothing modeling

CN122023362ACN 122023362 ACN122023362 ACN 122023362ACN-122023362-A

Abstract

The invention discloses a fabric shearing intelligent prediction method and device for clothing modeling, which are used for acquiring a corresponding image of a fabric to be detected, preprocessing the image through the fabric shearing intelligent prediction method, inputting the image into a target segmentation model for target segmentation, carrying out binarization in the target segmentation model according to a local brightness dynamic adjustment threshold of the image, improving the accuracy of the target segmentation model, extracting embedded features from a segmentation result through a feature embedding model, carrying out dimension reduction on the embedded features through principal component analysis, fusing the dimension-reduced embedded features with the acquired three-dimensional features, inputting the fusion result into an error correction model, obtaining an error correction value, and obtaining a final shear stiffness prediction value according to the error correction value and a shear stiffness theoretical value obtained through a shear stiffness mechanism model.

Inventors

LIU CHENGXIA
SUN YUE
ZHOU LIJIN
WU YUCHEN
WANG LIGUANG
FANG XIN
LI CHUNJING

Assignees

浙江理工大学

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. A clothing modeling-oriented fabric shearing intelligent prediction method is characterized by comprising the following steps: Cutting the fabric to be tested into a test sample, and hanging the test sample in a mode of clamping one corner of the test sample; The method comprises the steps of constructing a target segmentation model, dividing a fabric and a background in a fabric image by using the target segmentation model to obtain a binary image, wherein the target segmentation model comprises an encoder, a decoder, a bottleneck layer and an output layer, the encoder comprises a plurality of sub-coding modules, the decoder comprises a plurality of sub-decoding modules, the sub-coding modules and the sub-decoding modules are the same in number and correspond to each other one by one, the corresponding sub-coding modules and the sub-decoding modules are connected in a jumping manner, the output layer processes an output feature image of the last sub-decoding module through an activation function and a dual-threshold calibration module which are connected in series to obtain an output result of the target segmentation model, dividing the input feature image into a plurality of local blocks in the dual-threshold calibration module, calculating an average probability value of each local block, setting a pixel threshold according to the average probability value, and taking a part of the local block, the probability of which is greater than or equal to the pixel threshold, as a target, and the rest part as the background; constructing a feature embedding model, extracting embedded features from the binary image by using the feature embedding model, and fusing the embedded features and the three-dimensional features to obtain fused feature vectors; constructing a fabric shear stiffness mechanism model, and acquiring theoretical shear stiffness according to the fabric shear stiffness mechanism model; and constructing an error correction model, inputting the fusion feature vector into the error correction model to obtain an error correction value, and obtaining a final shear stiffness predicted value according to the theoretical shear stiffness and the error correction value.
2. The intelligent prediction method for fabric shearing property for clothing modeling according to claim 1, wherein in the encoder, the other sub-coding modules except the last sub-coding module all process the respective output characteristic map through a maximum pooling layer and then input the processed output characteristic map to the next sub-coding module; In the decoder, the input of the other sub-decoding modules except the first sub-decoding module is the splicing result of the processing result of the output characteristic diagram of the last sub-decoding module processed by the deconvolution enhancement module and the output characteristic diagram of the corresponding sub-coding module, and the input of the first sub-decoding module is the splicing result of the processing result of the output characteristic diagram of the bottleneck layer processed by the deconvolution enhancement module and the output characteristic diagram of the corresponding sub-coding module.
3. The intelligent prediction method for fabric shearing performance of clothing modeling according to claim 2, wherein the deconvolution enhancement module comprises a deconvolution layer and an edge alignment module connected in series, wherein the deconvolution layer is used for performing up-sampling operation on an input characteristic map, and the edge alignment module is used for shifting an output characteristic map of the deconvolution layer according to the shift amount of a maximum pooling layer.
4. The intelligent fabric shearing property prediction method for clothing modeling according to claim 1, wherein the sub-coding module and the sub-decoding module comprise a first convolution block, a channel attention module and a second convolution block which are sequentially connected, the channel attention module adopts an SE module, and the first convolution block and the second convolution block comprise a convolution layer, a batch normalization layer and an activation function which are sequentially connected.
5. The intelligent prediction method for fabric shearing property for clothing modeling according to claim 1, wherein the feature embedding model comprises a multi-layer convolutional neural network, and each layer of convolutional neural network comprises a plurality of convolutional layers, a plurality of maximum pooling layers and a global average pooling layer which are sequentially connected.
6. The intelligent fabric shearing property prediction method for clothing modeling according to claim 1, wherein the embedded features after feature screening and the three-dimensional features are fused to obtain a fused feature vector, and the feature screening mode is principal component analysis.
7. The intelligent fabric shearing property predicting method for clothing modeling according to claim 1, wherein the three-dimensional characteristics comprise overlapping length, longitudinal height difference, intersection point-lowest point length, front included angle and side included angle, wherein the overlapping length is the length of a sagging overlapping part of two ends of the fabric, the longitudinal height difference is the vertical distance between the intersection point and the lowest point of the fabric in a side view, and the intersection point-lowest point length is the horizontal distance between the intersection point and the lowest point of the fabric.
8. The intelligent prediction method for fabric shearing property for clothing modeling according to claim 1, wherein said error correction model optimizes back propagation neural network ‌ by genetic algorithm.
9. The intelligent fabric shearing performance predicting method for clothing modeling as set forth in claim 8, wherein the error correcting model uses unknown parameters in the shearing rigidity model as optimizing targets and the least mean square error between the fabric shearing rigidity mechanism model predicted value and the measured value as fitness function.
10. The intelligent fabric shearing property prediction device for the garment modeling is characterized by being used for executing the intelligent fabric shearing property prediction method for the garment modeling according to claim 1, and comprises a base (1), a support (2), a reference object (3), a fixing clamp (4), a data acquisition module and a shearing property prediction module, wherein the support (2) comprises a longitudinal section which is vertically fixed on the base (1) and a transverse section which is fixed at the top end of the longitudinal section, the fixing clamp (4) is arranged on the transverse section of the support (2) and used for clamping a sample (5) to be tested, the reference object (3) is used for calibrating the image size, the data acquisition module comprises an image data acquisition module and a point cloud data acquisition module, the point cloud data acquisition module is used for acquiring point cloud data, and the shearing property prediction module is used for predicting the shearing property of the fabric to be tested according to the data acquired by the data acquisition module.

Description

Intelligent fabric shearing property prediction method and device for clothing modeling Technical Field The invention belongs to the technical field of clothing testing, and particularly relates to an intelligent fabric shearing property prediction method and device for clothing modeling. Background The human body surface is composed of a plurality of complex three-dimensional curved surfaces, and in order to enable the fabric to be attached to the change of the concave-convex, radian and inclination of the human body curved surface, curved surface modeling techniques such as provincial way, pleat, parting line and return-pull are generally needed, and the implementation of the techniques cannot be separated from the good shearing performance of the fabric. The fabric with poor shearing deformability is easy to generate oblique arching phenomenon under the action of external force, thereby affecting the wearing attractiveness of the garment. Therefore, the shearing property of the fabric plays a vital role in the clothing modeling, and since the 60 th century, a plurality of domestic and foreign scholars have studied the fabric widely and developed a plurality of testing methods, such as a KES-F testing method, an off-axis tensile testing method, a photo frame shearing testing method and the like. Although there are many methods of testing fabric shear performance, most are costly or the testing process is complex. The traditional fabric shearing property test mainly depends on professional equipment such as a KES-FB1-A style instrument, and has the technical defects that the KES instrument is high in unit price, complex in operation, difficult to popularize in small and medium-sized clothing enterprises and fabric detection mechanisms due to the need of professional personnel maintenance, capable of obtaining shearing rigidity by combining complex data processing due to the fact that a special specification sample is required to be prepared for single test, incapable of intuitively presenting the fabric shearing form, and time-consuming and labor-consuming. Meanwhile, the current shearing property research only aims at the warp and weft direction expansion of the fabric, in practice, the mechanical property of the fabric has obvious anisotropy, the shearing property difference in different directions is obvious, and in practical wearing, the fabric receives shearing force from all directions, so that only the shearing property in the warp and weft direction of the fabric is concerned to be slightly insufficient. In order to solve the above problems, there is a need in the industry for a low-cost, multi-directional, visual fabric shear test technique, and at the same time, an efficient diagonal shear stiffness prediction model needs to be established, so that the workload of the full-angle test is reduced, and the fabric performance evaluation efficiency is improved. Disclosure of Invention The invention aims to overcome the defects of the prior art, and provides an intelligent fabric shearing property prediction method and device for clothing modeling, which realize low-cost multi-directional fabric shearing property detection and high-precision oblique shearing rigidity prediction. In a first aspect, the present invention provides a method for intelligent prediction of fabric shear properties for garment shaping, the method comprising: Cutting the fabric to be tested into a test sample, and hanging the test sample in a mode of clamping one corner of the test sample; The method comprises the steps of constructing a target segmentation model, segmenting fabrics and backgrounds in fabric images by using the target segmentation model to obtain a binary image, wherein the target segmentation model comprises an encoder, a decoder, a bottleneck layer and an output layer, the encoder comprises a plurality of sub-encoding modules, the decoder comprises a plurality of sub-decoding modules, the sub-encoding modules and the sub-decoding modules are the same in number and correspond to each other one by one, the corresponding sub-encoding modules and the sub-decoding modules are connected in a jumping manner, the output layer processes an output feature map of the last sub-decoding module through an activation function and a double-threshold calibration module which are connected in series to obtain an output result of the target segmentation model, the input feature map is divided into a plurality of local blocks in the double-threshold calibration module, an average probability value of each local block is calculated, a pixel threshold is set according to the average probability value, and the part of the local blocks, the probability of which is greater than or equal to the pixel threshold, is used as a target, and the rest part is used as the background; constructing a feature embedding model, extracting embedded features from the binary image by using the feature embedding model, and fusing the embedded features