책임운영기관 농촌진흥청 국립축산과학원 (로고)

An accurate and convenient individual identification method for dogs is required alongside the increasing number of dogs to handle the issue of abandoned and lost dogs or expand the pet dog industry with advanced technology. Specifically, some researchers have applied human biometric technology to dogs to meet these requirements. Relatedly, nose pattern image recognition is a representative method in dog biometric technology. However, since the dogs don't cooperate during imaging, many resulting images for nose pattern recognition are of low quality. Using these low-quality images results in a lower recognition rate for dog nose pattern. Therefore, this research proposes improving dog nose pattern recognition by classifying the image quality with a deep learning model. Fifty-five original images collected from 11 dogs were used to generate a dataset of 5,500 images by reelecting ten distortion factors that may occur during image collection. VGG16 was used as a deep learning classifier in this study. The deep learning classifier achieved an average quality classification accuracy of 86.65 % for the dog nose pattern images. The classification performance by image distortion type was in the order of brightness (91.52 %), sharpness (87.65 %), contrast (86.43 %), and noise (82.58 %). In conclusion, the proposed method is expected to improve the recognition rate and reduce the number of unnecessary recognition processing in dog nose pattern recognition.