Morphological Dilated Preprocessing with Depthwise Separable Visual Geometry Group Model for Classification of Minerals

Accurate identification of minerals may have considerable influence on research and industrial applications, which makes it a crucial task in geology and materials science. Image classification challenges have shown significant results for traditional Convolutional Neural Networks (CNNs), such as VGG model. Unfortunately, because of their depth and variety of parameters, these CNN models frequently need large amounts of computational power. For enhancing the performance of mineral classification while preserving high accuracy, this work introduces a unique method that incorporates depth wise separable convolutions into the VGG architecture. mineral images to the current CNN models. By splitting the convolution process into depthwise and pointwise convolutions, depthwise separable convolutions are a viable substitute for constrained computing settings due to their ability to reduce both computational complexity and model size. So the traditional VGG model was intergrated with depthwise separable convolutions at the end to propose Depthwise Separable VGG16 (DS-VGG16) that uses Mineral Identification Dataset from KAGGLE for execution. The efficiency of the proposed DS-VGG model was done in by applying the morphological reconstructed Dilated mineral image throug preprocessing. The DS-VGG16 classifies six mineral types by comparing with existing CNN models on evaluating the mineral identification dataset. This paper makes a significant addition to the fields of CNN by demonstrating the potential of depthwise separable convolutions in strengthening deep learning models for mineral classification tasks. The execution shows that the proposed DS-VGG16 model performs well in classifying mineral types, with a high accuracy of 99.63%.