High Dynamic Range Edge Filtered LeNet based Gemstone Classification

The precise detection of subtle characteristics and textures in gemstone photographs is necessary for the hard process of effective gemstone categorization. This research provides a unique method of classifying gemstones using convolutional neural networks (CNNs) based on LeNet and edge-filtered images. By highlighting the edges and contrasts in gemstone images, edge filtering improves the visibility of important characteristics including cuts, facets, and defects. The HDR function reveals subtle color and texture changes that are essential for differentiating between different types of gemstones by capturing a wider range of light intensities. This paper proposes High Dynamic Range Edge Filtered LeNet (HDEF-LeNet) that categorizes the Gemstone types with high accuracy. The Gemstone Dataset that includes 800 images was used for this execution. The HDR edge filtered images are extracted from the original images. After that, a customized HDEF-LeNet architecture designed to take into account the special qualities of gemstone images that is fed with this preprocessed data. Our strategy seeks to outperform conventional approaches based on conventional imaging techniques in terms of gemstone classification and accuracy. The findings of the experiments show that combining edge filtering and HDR with LeNet improves performance in classification tasks for gemstones, providing a viable approach to more accurate and consistent gemstone identification. The experiment demonstrates that the suggested HDEF-LeNet model works better in the categorization of gemstone types, with a high accuracy of $\mathbf{99.68 \%}$.