Research On Image Retrieval Algorithm And Application Of Colored Spun Knitted Fabric

The textile industry is a traditional pillar industry of China’s national economy and an important livelihood industry,and it is also an industry with obvious advantages in international competition.In recent years,the color-spun knitted fabric,which is processed by mixing two or more dyed fibers of different colors according to a certain proportion and spinning process,is popular among consumers and formed a unique color spinning industry because of its delicate and soft appearance and unique hazy three-dimensional effect and texture.With the continuous expansion of the scale of the color spinning industry,the production process of color spinning knitted fabrics is becoming more and more complicated,and its types and quantities are gradually increasing.It is difficult to meet the development needs of the fast-paced consumer market by relying on traditional manual retrieval methods.At the same time,the special process of "dyeing,then spinning,then weaving" of the color knitted fabric leads to randomness and uncertainty in the appearance of the fabric,making it difficult to characterize the image anisotropy of the color knitted fabric due to the different spinning and weaving processes,as the image features are easily disturbed by the image background when extracted based on convolutional neural networks.In view of the above problems,this paper focuses on an extensive and in-depth study on the following.(1)Aiming at the problem that a single feature does not accurately describe the complexity and anisotropy of color and texture in color textile images,an image retrieval of colored knitted fabrics combined with attention mechanisms is proposed.Firstly,the Res Net50 network is fused with an attention mechanism module to improve the network model’s ability to express the semantic features of color-spun knitted fabric images.At the same time,deep hash coding is used to calculate the similarity of semantic features.On this basis,step-by-step retrieval is achieved by combining shallow visual features of images,thus improving retrieval efficiency.The experimental results show that the recall rate and the average accuracy rate of the whole category are 98.26% and 88.83%,respectively,for the retrieval of 14 classes of color-spun knitted fabric images with different stylistic semantic features.Meanwhile,compared with the Res Net50 single network model with fused attention mechanism and the fused shallow visual features approach,the recall rate and the average accuracy rate of the whole category improved by 12.62%,7.53% and 18.95%,11.6%,respectively.(2)Aiming at the problem that image feature spectral overlap between dyed fibers and tissue structure in color spinning knitted fabric images,a color spinning knitted fabric image retrieval algorithm based on image decoupling features is proposed.Firstly,we decouple the color spinning knitted fabric image by the relative total variance model to obtain its style feature map and organization structure feature map.At the same time,The VGG16 network is used as a feature extractor for semantic feature extraction of the two decoupled feature maps separately.The experimental results show that the recall rate and the average accuracy rate of the whole category of the method Top-10 are 89.21% and 82.56%,respectively..(3)Due to the variability in the contribution of decoupling features to the retrieval of images of different types of color-spun knitwear,a color spinning knitted fabric image retrieval algorithm incorporating variable weight features is proposed.Firstly,a channel feature cascade is performed on the style features and the organization structure features.Then,after global averaging pooling is processed and input to the fully connected layer,the dynamic weights of the two decoupled features are obtained.The experimental results show that the recall rate and the average accuracy rate of the whole category are 90.47% and83.49%,respectively.Compared with the fixed-weight feature,its recall rate and the average accuracy rate of the whole category improved by 1.26% and 0.93%,respectively..This paper is a study on the image retrieval of colored spun knitted fabrics,which can provide a theoretical basis for the retrieval of images of color textile products as well as digital management and the reuse of process parameters,thus enabling companies to meet the market demand for small batches,multiple varieties and fast delivery.