International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 11 | Nov 2025
p-ISSN: 2395-0072
www.irjet.net
A Fusion-Based VGG19 and Vision Transformer Framework for Automated Dog Breed Recognition with Lifespan and Trait Profiling Prof.Shobha S. Biradar1, Naveen 2 1Professor, Master of Computer Application VTU’s CPGS, Kalaburagi, India 2Student, Master of Computer Application VTU’s CPGS, Kalaburagi, India
---------------------------------------------------------------------***--------------------------------------------------------------------effective in the laboratory, those methods are challenged ABSTRACT
by varying sources of illumination, varied poses, and occlusion of parts of the animal body, as well as the high degree of similarity between some breeds, such as Golden Retrievers and Labrador Retrievers.
Due to the extensive range in morphology, behaviors, and lifespan across breeds, the significance of dog breed identification to veterinary science, genetics and increased care of companion animals cannot be understated. Identification of canine breeds through conventional methodologies is reliant on veterinarians making independent decisions, or on using handcrafted features. Generalized breed recognition methods are hindered by combinatorial variability within breeds and similarities to other breeds, thus we developed a fusion, action oriented approach to dog breed recognition that utilizes VGG19 and Vision Transformers to automate breed recognition and trait profiling of lifetime and breed attributes. VGG19, with fine-grained local discriminating powers, permits identification of breeds, while Vision Transformer models global dependencies to produce robust and complementary feature representations. Our framework had greater accuracy in breed recognition and generalization and robustness relative to conventional methods of model predictions, based on findings from experiments utilizing annotated datasets. The framework can be used for dog recognition, and identify actionable information for preventive health screening, targeted breeding, and best practices for pet care. Keywords: Dog breed recognition, VGG19, Vision Transformer, deep learning, trait profiling, lifespan prediction, companion animal management.
The advent of deep learning methods has initiated a new era for fine-grained recognition research, where our best success has come from CNNs models including VGGNet and ResNet. However, a limitation of CNNs is their inability to model long-range relationships that exist in image data. In contrast, Vision Transformers (ViT) can capture global context through the use of self-attention mechanisms, but most ViTs require expansive datasets in order to be effective. As a result, hybrid CNN-Transformer systems have emerged as a promising trend in the literature. Here we propose a dog-breed recognition model that fuses a VGG19 with a ViT model for predicting breed, life expectancy, and trait profiling.
2. PROBLEM STATEMENT Identifying dog breeds reliably is a difficult challenge because of broad variation in physical traits, overlapping physical traits across breeds, and changes due to age, health, or environment. And while this issue is often addressed by dog breed experts manually classifying breeds, their subjective manual methods might lead to mistakes and take a considerable amount of time and effort--especially with mixed or rare breeds. Automated approaches have been developed, but many are only mediocre in their accuracy and don't provide additional features like life expectancy or behavioral traits along with those predictions. Such limitations emphasize the need for accurate, robust, and informative recognition paradigm that supports veterinary professionals and responsible pet owners alike.
1. INTRODUCTION Dogs exhibit incredible diversity as one of the most varied domesticated animals on the planet, with hundreds of breeds that have diversity in physical attributes, behavior, and life expectancy. Given their varied roles as companions, working animals, and bespoke research models, accurately identifying breeds has important implications for veterinary medicine, behavioral science, genetics, and increasingly for intelligent pet care systems. Correct breed recognition leads to informed healthcare decision-making about pets, and can aid in life expectancy and behavioral inclusion when personalizing pet care.
3. OBJECTIVES The primary goal of this work is to create a sophisticated dog breed identification system that employs deep learning methods for precise and useful predictions. This strategy utilizes a hybrid model of VGG19 and Vision Transformer (ViT) to exploit both detailed local texture information and holistic global interactions in order to enhance classification accuracy. The model is
Historically, breed identification has relied upon visual assessment by experts or through calibration of features derived from handcrafted computer-vision. While it is
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 732