Neural network classification of aggregates by means of line laser based 3D acquisition

This study focuses on the development of a new module for a more accurate determination of geometrical properties of aggregates. A laser-based imaging system has been developed for the shape characterization of aggregates by using various digital image analysis techniques. By using this system it is possible to create a three-dimensional (3D) image form of aggregate particles. The system has been optimized to minimize the possible errors during image capturing and processing. The aggregates were classified according to their shape properties as; round, flat, elongated, angular, sphere, and irregular during test procedures. Geometrical properties of each aggregate group were analysed in 3D spatial domain. 3D shape reconstruction and characterization of the aggregates were realized by using digital image processing and analysis techniques based on the laser imaging system. MatLab R © Image Processing Toolbox and Neural Network Toolbox were used to extract typical features of the aggregates and classify them according to their geometrical properties. Among the classifier types, multi-layered perceptron that has two hidden layers revealed the best performance (95.83%). The selection and production of appropriate shaped aggregate for various construction purposes seems to be possible by this developed method.