Predictive modeling of geometric shapes of different objects using image processing and an artificial neural network

In this study, an artificial neural network model was developed to predict the geometric shapes of different objects using image processing. These objects with various sizes and shapes (circle, square, triangle, and rectangle) were used for the experimental process. In order to extract the features of these geometric shapes, morphological features, including the area, perimeter, compactness, elongation, rectangularity, and roundness, were applied. For the artificial neural network modeling, the standard back-propagation algorithm was found to be the optimum choice for training the model. In the building of the network structure, five different learning algorithms were used: the Levenberg-Marquardt, the quasi-Newton back propagation, the scaled conjugate gradient, the resilient back propagation, and the conjugate gradient back propagation. The best result was obtained by 6-5-1 network architectures with single hidden layers for the geometric shapes. After artificial neural network training, the correlation coefficients (R-2) of the geometric shape values for training and testing data were very close to 1. Similarly, the root-mean-square error and mean error percentage values for the training and testing data were less than 0.9% and 0.004%, respectively. These results demonstrated that the artificial neural network is an admissible model for the estimation of geometric shapes using image processing.