Intelligent Threshold Prediction for Hybrid Mesh Segmentation Through Artificial Neural Network

Abstract

Accurate and reliable Area deviation factor (threshold) is one of the decisive factors in hybrid mesh segmentation. Inadequate threshold leads to under-segmentation or over-segmentation. Setting the optimal threshold is a difficult task for a layman. This proposed method, automatically predicts the threshold using artificial neural networks (ANN). ANN predicts the threshold by considering mesh quality of Computer-Aided Design (CAD) mesh model as input feature vectors. Extensive testing on benchmark test cases validates ANN prediction model, and based on Levenberg-Marquardt back propagation (LM-BP) improves the accuracy and stability of prediction. The efficacy of the approach is quantified by measuring coverage. The ANN predicts the threshold elegantly using LM-BP algorithm with coverage for hybrid mesh segmentation greater than 95%. The novelty of the proposed method lies in the “mesh quality”-based threshold prediction through ANN. The predicted threshold finds application in automatic feature recognition from CAD mesh model using hybrid mesh segmentation.