Development and Evaluation of a Dog Breed and Identity Recognition Model Based on Convolutional Neural Networks

This study uses CNN to improve dog management by recognizing breeds and identities from images. It starts with dog and face detection using YOLOv2, creating the YOLODog and YOLODogFace models. InceptionV3 extracts features for breed classification, which are classified by XGBoost into the Inception $\mathrm{XG}_{\text {Dog }}$ and Inception $\mathrm{XG}_{\text {DogFace }}$ models and combined in the Inception $\mathrm{XG}_{\text {Fusion }}$ model. For identity recognition, YOLO ${ }_{\text {DogFace }}$ detects the face, InceptionV3 extracts features and the Inception ${ }_{\text {DogID }}$ model uses similarity measures for identification. Testing shows accurate detection except for incomplete images. Breed classification is effective but struggles with similar or low-resolution images. Combining features improves results, and identity recognition is less effective for similar dogs of the same breed.