Research On Cloth Material Recognition Method Based On Deep Learning

With the continuous development of the textile industry,fabric material identification technology has been widely used in industrial fields such as fashion,decoration,and design.Therefore,many fabric material identification methods have been proposed.In recent years,methods based on artificial recognition,physics,chemistry,microscopy,near-infrared spectroscopy,and image processing have emerged to identify fabric materials.However,there are problems such as long recognition cycles,multiple human factors,high technical barriers,and destruction of textiles.Therefore,in response to the above issues,this article proposes two methods for material identification of moving fabrics in video: cloth material identification based on multi branch SSTS and cloth material identification based on improved Transformer,which can ensure high accuracy while reducing computational complexity.The specific work is as follows:(1)Aiming at the problems of large computational complexity,low recognition accuracy,and background interference when processing long videos using three-dimensional convolutional neural networks,this paper proposes a cloth material recognition method based on multi branch SSTS.This method inputs a cloth motion video and identifies the type of cloth material in the video through the appearance changes of the cloth motion.This method divides 32 RGB video frames and optical flow frames into 16 RGB video frames and optical flow frames,and inputs them into the SSTS network,improving the recognition accuracy;At the same time,using STS networks to divide channels into spatial groups,temporal groups,and spatiotemporal groups significantly reduces the amount of computation and parameters of the network;In addition,the combination of STS and attention mechanism makes the network pay more attention to important image features,reducing the impact of background,and improving the accuracy of fabric material recognition.(2)Aiming at the problem that convolutional neural networks can only capture local information,have small receptive fields,and cannot establish long-distance connections to global images,this paper proposes a cloth material recognition method based on improved Transformer.This method inputs a cloth motion video and identifies the type of cloth material in the video through the appearance changes of the cloth motion.Decomposing the self attention in the Transformer block into temporal self attention and spatial self attention reduces the amount of computation and network complexity;At the same time,residual space reduction blocks are used in the Transformer model to improve the accuracy of fabric material recognition.