발간년도 : [2022]
| 논문정보 |
|
| 논문명(한글) |
[Vol.17, No.4] XAI-based Dog Breed Classification |
|
| 논문투고자 |
Min-Kyu Park, Sang-Jun Park, Jun-Ho Shin, Min-Ji Lee, Ik-Su Kim |
|
| 논문내용 |
Recently, interest in companion dogs is getting hotter day by day. In korea, the era of petconomy has arrived as the population who owns dogs has rapidly increased. As of 2020, around 5.1 million households have companion dogs. When adopting a companion dog, people choose a dog breed based on the appearance and personality. However, cheating with more expensive breeds in pet shops has become a social problem. Accordingly, research on breed distinction using deep learning has been actively conducted. However, it is difficult for buyers to understand the breed identification method using deep learning. To solve this problem, in this paper, we build a CNN-based deep learning model that distinguishes dog breeds and visualize the classification basis as an image using LIME, one of the XAI techniques. For the data set, we selected the dog breeds that are raised the most in Korea, and then added the dog breeds that are difficult to distinguish from each other. Then through experiments, we found the optimal hyperparameters and extracted the images that best show the basis for classification. Based on this, a person who adopts a pet can trust the breed of the pet they want to adopt, and a pet shop can also build trust by presenting the evidence of the breed. |
|
| 첨부논문 |
|
|
|
|
|
자료실 > 논문지 
