Classification of shape aggregate using activation function based multilayer perceptron (MLP)

Mechanical sifting and hand grading have traditionally been employed to evaluate the quality of aggregates. However, such assessments require various mechanical, chemical, and physical tests, which are usually carried out manually and are slow, subjective, and time-consuming. This study seeks to develop an image-based classification system for categorizing aggregates. An artificial neural network was utilized to process the captured images and classify their shapes. The aggregate images are captured and used as the input parameter for prediction before the threshold process occurs. The Logsig activation function, based on a Multilayer Perceptron (MLP) network, has the lower mean square error (MSE) and the higher regression, compared to the Pureline activation functions. The Logsig-based network has an MSE of 1.7473 and a regression capacity of 0.9521.