Classification of animal species using efficient neuron attention stage-by-stage network

Abstract

Classifying animal species is crucial for understanding their evolutionary relationships and characteristics. This article introduces the Efficient Neuron Attention Stage-by-Stage Network (ENAS-Net) for animal species classification. The process involves obtaining animal images from a database, applying anisotropic diffusion for noise reduction, and using O-SegNet for segmentation. Image augmentation techniques like colour augmentation, flipping, and rotation are used to enhance performance. Features are extracted using Entropy-based Opponent Color Local Binary Pattern (E-OCLBP) and Convolutional Neural Network (CNN). The ENAS-Net is a combination of Neural Architecture Search Network (NASNet) and EfficientNet. Moreover, for dataset 1, the ENAS-Net recorded an accuracy of 91.405 %, a True Positive Rate (TPR) of 92.045 %, and a True Negative Rate (TNR) of 91.549 %. For dataset 2, the proposed approach demonstrated strong performance, achieving an accuracy rate of 92.172 %. Additionally, it maintained a TPR of 92.994 %, effectively identifying relevant instances, while sustaining a TNR of 92.452 %, minimizing false detections.