<![CDATA[Distinguishing animals that appear visually similar but belong to different species or taxonomic groups, such as Eurasian and house sparrows, koi and common carp, or leopard cat and domestic cat, remains challenging and hinders biodiversity education. This study develops a Convolutional Neural Network (CNN)-based multi-output, multi-class taxonomy classification system capable of identifying seven animal species across five taxonomic levels (class, order, family, genus, species), producing 35 possible outputs. The dataset comprised 6,998 images from public sources. Among various configurations, the best-performing model (D3-M2), trained using the High Dataset with 256×256 input size, 0.2 dropout, and four hidden layers, achieved 90.15% average accuracy, the highest F1-score at the family level (98.11%), and 95.99% at the species level. Slightly lower species-level performance was due to high visual similarity among particular species. Edge AI deployment offered faster inference (0.17s) and offline capability, making it ideal for field use. Real-world testing under bright and low light at 30, 60, and 100 cm showed higher accuracy (64.8%) than low light (57.1%), with the most stable performance at 60 cm. However, limitations include an imbalanced dataset and limited environmental variation affecting species-level accuracy. Future work will focus on expanding dataset diversity and employing advanced architectures to improve fine-grained classification. This system offers a practical tool for biodiversity education and species identification, particularly in field environments where rapid, offline, and accurate classification is essential.]]>
