Objective Evaluation Of Fabric Sewing Flatness Based On Convolutional Neural Network

As the first place of clothing,food,shelter,and clothing,clothing has played an important role in protection and beauty from ancient times to today.Today,the material culture has been greatly satisfied,and the appearance quality of clothing reflects personal temperament and taste to a certain extent..Therefore,the garment industry attaches more and more importance to the appearance quality inspection of garments.As an important indicator of garment appearance quality inspection,the flatness of fabric sewing has attracted more and more attention.At present,scholars mostly use two-dimensional image technology and three-dimensional laser scanning technology to evaluate the sewing flatness of fabrics.However,because the type of fabric and environmental factors have a greater impact on the extraction of image data,the experimental data of the existing methods are partially distorted,which leads to the evaluation results.Low accuracy.In view of this situation,this paper uses a convolutional neural network to objectively evaluate the flatness of fabric sewing,uses a large number of labeled fabric sewing images to train a convolutional neural network model,and tests the trained model.Finally,the feasibility and accuracy of the model are compared using 3D laser scanning technology.The model can effectively reduce the impact of the sample color and other characteristics on the classification results.The specific research content and results are as follows(1)Construct a sample library of 10 common clothing fabrics,including cotton,linen,silk,wool,polyester,etc.,and cut,iron,sew,and pleate these 10 fabrics to make sample seams.A Canon digital camera was used to collect two-dimensional image data of the samples,and a total of 1200 images were obtained.After cropping,it was divided into 1000 training samples and 200 test samples according to a 5:1 ratio.The subjective evaluation method is used to label the training samples.The labeled image data is input to train the convolutional neural network model,and the test samples are used to test the model classification results.The classification accuracy rate is 96.7%.(2)Use the 3D point cloud processing technology to objectively evaluate the sewing flatness of the fabric,use the REVscan handheld laser 3D scanner to obtain the 3D point cloud data of 200 samples,and use Geomagic Studio 12 reverse engineering software to reconstruct the 3D surface Preprocessing such as noise,cropping,etc.,and calculating the five characteristic values of the mean value,roughness,distortion,mean shift,and kurtosis of the point cloud data after preprocessing by MATLAB,using SPSS software to correlate and regression analysis with subjective scores The significant correlation between roughness and subjective score was obtained,with a correlation coefficient of 0.897,and the regression model Y=7.146X-3.042 is obtained,where Y is the subjective score and X is the roughness in the characteristic value of the 3D point cloud data.The accuracy of this regression equation was tested by selecting 200 fabric samples and the accuracy was 91%.(3)Compare and analyze the advantages and disadvantages of the two methods of convolutional neural network model and 3D point cloud data technology.The convolutional neural network model can be used to classify sewing flatness grades more efficiently,not only the classification accuracy is better than that of 3D point cloud.The data processing method is 6%higher,and its sample data collection is simpler and faster.The classification results are less affected by the type of fabric and the model has a wider application range.