Transfer Learning-based Rich Feature Analysis on Leather Images for Species Prediction

Determining animal species in leather plays a significant role in biodiversity preservation and consumer protection. Unlike the traditional methods, the present work employs digital leather image analysis for automatic leather species prediction. However, it is challenging to derive species-distinct features from the complex leather image data. Hence, this paper proposes transfer learning-based rich feature analysis. In addition, it proposed to use Generative Adversarial Network (GAN) to pre-process the leather images with plausible pore pixels. The combination of adaptive learning in GAN and transfer learning led to the extraction of distinguishable features for efficient classification. Finally, it trains a support vector machine (SVM) for the classification of rich features extracted from AlexNet, VGG16, GoogLeNet, and ResNetl8, respectively. The comparative study affirms ResNetl8, with 99.58% prediction accuracy, as an appropriate choice for rich feature extraction. It elevated leather species prediction over the existing hand-crafted feature analysis. Thus, the proposed work adaptively learned the diverse behavior of leather images for accurate species prediction.