D-Resnet: deep residual neural network for exploration, identification, and classification of beach sand minerals.

The beach sand minerals are in great demand since they are the source of titanium and are widely used in atomic energy and many other industries. Despite that, the identification and classification of beach sand minerals are challenging because of the width of the deposit, grain size, mineral composition, and locality. The proposed work develops a beach sand identification and mineral classification methodology using a deep computer vision technique with the support of neural networks. This work develops the deep residual neural network (D-Resnet) model, which extracts the mineral image features with the support of a convolutional feature selection process and filters the extracted mineral features. The D-Resnet model minimises the reliance on high-resolution mineral images and provides a more improved and efficient model. Further, the residual neural networks with the depth of 18, 34, 50, and VGG with a depth of 16 models are built by embedding various pooling methods for classifying beach sand minerals. The performance of the proposed D-Resnet model has been systematically experimented with and evaluated with the confusion matrix, classification accuracy, sensitivity, and specificity scores. Indeed, the D-Resnet model resulted in 89% of accuracy for classifying six types of beach sand minerals. [ABSTRACT FROM AUTHOR]