Intelligent animal fibre classification with artificial neural networks

Artificial neural networks (ANN) are increasingly used to solve many problems related to pattern recognition and object classification. In this paper, we report on a study using artificial neural networks to classify two kinds of animal fibers: merino and mohair. We have developed two different models, one extracting nine scale parameters with image processing, and the other using an unsupervised artificial neural network to extract features automatically, which are determined in accordance with the complexity of the scale structure and the accuracy of the model. Although the first model can achieve higher accuracy, it requires more effort for image processing and more prior knowledge, since the accuracy of the ANN largely depends on the parameters selected. The second model is more robust than the first, since only raw images are used. Because only ordinary optical images taken with a microscope are employed, we can use the approach for many textile applications without expensive equipment such as scanning electron microscopy. ters. These classifiers still work with parametric discriminant functions but are able to implement a larger class of discriminant shapes (eventually any shapes, which makes them universal approximators) [9]. Supervised artificial neural networks are one of the most exciting semi parametric classifiers. Recently, artificial neural network models and learning algorithms for pattern recognition and classification have been developed. In the current work, we develop an intelligent animal fiber classifier by integrating image processing and artificial neural networks to classify two kinds of animal fibers: mohair and merino. Scale Patterns of Animal Fibers Merino and mohair fibers, like other natural animal fibers, consist of three morphological components: the cuticle on the surface, the cortex, and the medulla. The cuticle is composed of flat, plate-like cells called scales. They are laid down in an overlapping pattem with the free ends pointing towards the fiber tip. Scale patterns are associated with aspects of fleece quality and have a great bearing on the characteristics of products made from the fibers. They also provide very important information about the identities of animal fibers in classification. However, there are considerable variations in the shape and contour of the scale cells and their arrangement within the cuticle. This happens even within the same animal and along the same fiber because of the nature of growth [I, 2]. Furthermore, there are very subtle differences between different animal fiber types, but there are still some common Accurate classification of animal fibers used in the wool industry is very difficult, although a number of techniques have been developed. Some techniques distinguish these fibers from the patterns of their cuticular scales and others from their physical and chemical properties. However, the characteristic features of these scales are still the most useful evidence for a skilled microscopist to distinguish animal fibers such as merino, mohair, and cashmere [6, 13]. From this point of view, classification of animal fibers is actually a typical task of pattern recognition and classification. Recently, to develop an objective method to identify and subsequently classify animal fibers, Robson used an imaging processing technique to extract characteristic information from scale patterns and linear demarcation functions to classify cashmere and merino fibers based on a linear discriminant statistical analysis [11]. This method is called a parametric classifier because its discriminant functions are based on statistical models and have a well-defined mathematical functional form (normally Gaussian) that depends on a set of parameters uch as mean and variance. However, some nonparametric classifiers have no assumed functional form for the discriininants. Nonparametric classification is solely driven by the data. It is free from assumptions about the shapes of discriminant functions or data distributions that may be erroneous. but this method requires a large number of data in order to perform acceptably [10]. There is another, more versatile method called a semiparametric classifier. which is an excellent compromise between versatility and the number of trainable parameTextile Res. J. 72(). 594-600 (2002) 0040-5175/$1500