A Novel Aggregate Classification Technique Using Moment Invariants and Cascaded Multilayered Perceptron Network

Occupying more than 70% of the concrete’s volume, aggregates play a vital role as the raw feed for construction materials; particularly in the production of concrete and concrete products. Often, the characteristics such as shape, size and surface texture of aggregates significantly affect the quality of the construction materials produced. This article discusses a novel method for automatic classification of aggregate shapes using moment invariants and artificial neural networks. In the processing stage, Hu, Zernike and Affine moments are used to extract features from binary boundary and area images. In the features selection stage, discriminant analysis is employed to select the optimum features for the aggregate shape classification. In the classification stage, a cascaded multilayered perceptron (c-MLP) network is proposed to categorize the aggregate into six shapes. The c-MLP network consists of three MLPs which are arranged in a serial combination and trained with the same learning algorithm. The proposed method has been tested and compared with twelve machine learning algorithms namely Levenberg-Marquardt (LM), Broyden-Fletcher-Goldfarb-Shanno quasi-newton (BFG), Resilient back propagation (RP), Scaled conjugate gradient (SCG), Conjugate gradient with Powell-Beale restarts (CGB), Conjugate gradient with Fletcher-Reeves updates (CGF), Conjugate gradient with Polak-Ribiere updates (CGP), One step secant (OSS), Bayesian regularization (BR), Gradient descent (GD), Gradient descent with momentum and adaptive learning rate (GDX) and Gradient descent with momentum (GDM) algorithms. Also, the classification performance of the c-MLP network is compared with those of the hybrid multilayered AC C EP TE D M AN U SC R IP T ACCEPTED MANUSCRIPT 2 perceptron (HMLP), the radial basis function (RBF) as well as discriminant analysis classifiers. Concerning the cascaded MLP, 3 stage c-MLP gives the best accuracy compared to the 2 stage c-MLP and the standard MLP. Compared to other learning algorithms, LM algorithm achieved the best result. As far as the overall conclusion is concerned, c-MLP gives better classification performance than that of the HMLP, RBF and discriminant analysis. Index Terms Aggregate classification; Pattern classification; Moment invariants; Image processing; Discriminant analysis, Cascaded multilayered perceptron (c-MLP) network; Artificial neural network.