Article
HYBRID DEEP LEARNING ALGORITHMS FOR DOG BREED IDENTIFICATION
Dog breed identification is a challenging fine-grained visual classification task due to high inter-breed similarity and substantial intra-breed variation in appearance, pose, and lighting. Accurate automated identification is valuable for pet management, veterinary care, lost pet recovery, and animal research. This paper proposes a Hybrid Deep Learning approach that combines the complementary strengths of ResNet-101, InceptionV3, and Xception architectures through ensemble feature fusion and weighted majority voting. The hybrid system leverages transfer learning from ImageNet pre-trained weights and applies extensive data augmentation to improve generalization on the Stanford Dogs dataset (120 breeds, 20,580 images). A comparative analysis evaluates individual and hybrid models. Experimental results demonstrate that the proposed hybrid approach achieves 91.4% top-1 accuracy, outperforming individual models by 3-7%, offering a robust solution for automated dog breed recognition.
Full Text Attachment
