Logsig activation function based multilayer perceptron network for aggregate classification

Mechanical filtration and manual sorting have long been standard methods for evaluating aggregate quality. While producing high-quality aggregates necessitates a variety of mechanical, chemical and physical assessments, these tests are often conducted manually, leading lo inefficiencies. subjectivity, and significant labour-demands. This research aims to develop an innovative image-based classification system to categorize aggregates more effectively. An artificial neural network (ANN) has heen employed for the classification of the images captured in this process. ln contrast to the Purelin activation function, the Logsig activation function shows improved performance, indicated by
a decrease in mean square error (MSE) and better regression outcomes. Notably the BR training algorithm utilizing a multilayer perceptron (MLP) network, aimed at reducing the MSE, provides the most effective regression results and the lowest MSE. The MSE achieved by the network trained with BR was 1.4235, accompanied by a regression coefficient of 0.9760. These fìndings that implementing advanced computational techniques can signiticantly enhance the quality control processes in aggregate production, therey by promising improvements in efficiency and material performance standards .