Moisture Content Measurement of Yarn based on Deep Multi-task Learning

The moisture content detection of fabrics is essential in production and processing of textiles and microwave moisture content measurement has the advantages of online, no contact and real time. Because of data driven principle, current regression models for moisture content detection are only applicable to a range of identical subjects under test. To overcome this limitation, this paper investigates the use of deep multi-task model to achieve accurate prediction of moisture content for a series of similar subjects under test that have the same structure but differ in material and size. The model consists of deep belief network at the bottom and support vector regression at the top. The experimental results show that the proposed deep multi-task model is suitable for moisture content detection of cotton yarn and polyester yarn, and improves prediction accuracy compared with single-task learning.