Training algorithm based MLP network prediction for shape aggregate

The assessnoent of aggregate quality is predicated on a combination of manual grading and mechanical filtering through established traditional methods. Aggregates are mandated to undergo a series of mechanical, physical, and chemical testing protocols to verify their compliance with predefined standards. The manual evaluation processes are intrinsically inefficient and subjective, resulting in significant temporal resource expenditure. This project to design an image processing system capable of categorizing aggregates into distinct classifications. The classification system leverages an artificial neural network (ANN) to perform image analysis for the identification of aggregate morphologies. The study systematically evaluates the performance of various training algorithms för the ANN, specifically juxtaposing Levenberg Marquardt (LM) against Bayesian Regularization (BR) as training paradigms. The findings demonstrate that BR training outperforms other methodologies, as evidenced by superior mean square error (MSE) metrics and improved regression results. The integration of the BR training algorithm with a multilayer perceptron (MLP) network achieves optimal performance in terms of regression accuracy and MSE assessment. Through the implementation of BR training, the network attained an MSE of 1.2042 and a regression coefficient of 0.9892, thereby validating its capability to classify aggregates through image analysis effectively. This novel approach provides researchers with a robust and objective solution that supersedes traditional manual classification methodologies.