Content area
Full text
1. Introduction
Up to now, the disposal methods of waste textiles mainly focus on donation, incineration or landfill. The latter two have serious damage to the ecology and lead to low self-utilization. Due to the small amount of rewearable in waste textiles, it is no longer possible for wear parts to become the main object of recycling. Recycling waste textiles and then reconstructing raw material of fiber is the main green and resource-saving direction to deal with the waste textiles. Efficient and high-accuracy intelligent sorting systems for textile color and material are the main bottlenecks restricting the recycling of waste textiles, as shown in Figure 1.
More and more countries have paid attention to the area of textile fiber reconstruction based on the waste textiles such as Prato in Italy and Panipat in India (Wedin et al., 2017). Prato's industry is known for recycling wool and cashmere blends, while Panipat is known for recycling cotton, polyester and many other textiles. Recycling of waste textiles into new yarns by mechanical recycling (shredding or processing) enables fiber reconstruction, which greatly realizes resource regeneration. Some international brands, such as H&M, have also begun to do research on recycling of waste textiles (H&M Foundation, 2018).
The biggest difficulty in the comprehensive utilization of waste textiles is the establishment of recycling system and sorting process (Alkazam, 2013), in which color sorting and material sorting are the main sorting processes for recycling waste textiles, and the former is often the first step, refer to Figure 1. With regard to material sorting, some techniques, such as NIR (near infrared) or hyperspectral techniques (Englund et al., 2018), can be exploited very well, and the detected color information is beneficial for this process. At present, color sorting of waste textiles mainly relies on manual operation, and no mature intelligent technology can be applied directly. Due to long-term continuous manual works, the sorting efficiency is low, and the consistency of sorting quality is difficult to be guaranteed. Additionally, overlapping areas between different colors (such as the mixing of different proportions of red and yellow) often exist; many textile samples with uncertain colors need to be handled in the classification process, which further reduces the manual points.
Computer vision (Vedaldi and Fulkerson, 2010) is...