Identification of plastics using wavelets and quaternion numbers

In recent years the recycling sector has become a vigorous economic activity. The more important it becomes, the more automatism is required in order to increase the profit. This work describes the construction of an automatic classification system for plastic containers in domestic wastes. The system takes advantage of the different near infrared spectra from the plastics of economic interest. These spectra can be obtained very quickly by means of the acousto-optic tunable filter technology. Spectral data are processed on line by means of an algorithm based on wavelets to achieve a positive identification. Wavelets are chosen because of their noise immunity and their easy implementation. Once identified, the containers can be taken from the conveyor belt by using a pneumatic system. The algorithm uses the wavelet transform in order to obtain a coefficient set that characterizes the original signal with few data. Even with these reduced data the signal keeps enough information to be able to classify each type of plastic material. The coefficient set is used to construct a quaternion number. Using a calibration method, a standard quaternion is obtained for each type of plastic. A container can be classified by calculating the Euclidean distances among these standard quaternions and the measured quaternion of the container.