Application and research of clothing fabric identification based on cloud, edge and end intellisense

Spectrum is the fingerprint of substance. Using spectrum analysis technology to identify substance will open up a new dimension of information for artificial intelligence and perception technology, which has a huge application demand in many fields. The project team uses the wavelet transform near infrared spectrum (WTNIR) method to complete the analysis of clothing fabrics. Due to the good characteristics of wavelet transform, it is widely used in the field of graphics and image processing, and achieves better practical results than the original technology. By studying the basic theory of wavelet transform, the wavelet transform theory is applied to image processing and overcomes the defects of short-time Fourier transform in single resolution. It has the characteristics of multi-resolution analysis, and has the ability to represent local information of signals in both time and frequency domains. Using wavelet transform multi-scale edge detection image segmentation method can clearly distinguish the yarn density of clothing fabrics, so as to better identify the quality of the fabric. This project uses decision tree to build material classification model by the spectroscopic data which we collected, uses U-net to build a neural network to distinguish the yarn density, uses small sample transfer learning detection technology to complete clothing color distinction, and uses Edge AI small model training and the method in the transfer learning to improve the performance and efficiency of the neural network on the terminal target hardware. Through experiments, it is found that the material composition of clothing fabrics can be accurately analyzed at the spectral band of about 1500 nm, which provides a basis for the subsequent development of hand-held clothing fabric analyzer.