Pill Recognition Using Minimal Labeled Data
© 2017 IEEE. Inappropriate medication use such as wrong drug or wrong dose intake can be harmful to patients. In this work we present a method to automatically identify a pill from a single image using Convolutional Neural Network (CNN). We first localize the pill in the image by detecting the regio...
| Main Authors: | , , , , |
|---|---|
| Format: | Conference Paper |
| Published: |
2017
|
| Online Access: | http://hdl.handle.net/20.500.11937/70029 |
| _version_ | 1848762196690993152 |
|---|---|
| author | Wang, Y. Ribera, J. Liu, C. Yarlagadda, Sri Kalyan Zhu, Maggie |
| author_facet | Wang, Y. Ribera, J. Liu, C. Yarlagadda, Sri Kalyan Zhu, Maggie |
| author_sort | Wang, Y. |
| building | Curtin Institutional Repository |
| collection | Online Access |
| description | © 2017 IEEE. Inappropriate medication use such as wrong drug or wrong dose intake can be harmful to patients. In this work we present a method to automatically identify a pill from a single image using Convolutional Neural Network (CNN). We first localize the pill in the image by detecting the region with the highest concentration of edges. To overcome the challenge of minimal labeled training data and domain shift from the training images taken under the controlled lab environment to the consumer images taken under natural living conditions, several data augmentation techniques are applied on the Region of Interest to generate synthetic pill images for training the CNN. We adopted GoogLeNet Inception Network as our main classifier. Three GoogLeNet models with different specialties on color, shape and feature are trained on the augmented dataset. We evaluate our proposed method with a publicly available dataset provided by National Institute of Health that contains 1000 different pill classes. |
| first_indexed | 2025-11-14T10:43:44Z |
| format | Conference Paper |
| id | curtin-20.500.11937-70029 |
| institution | Curtin University Malaysia |
| institution_category | Local University |
| last_indexed | 2025-11-14T10:43:44Z |
| publishDate | 2017 |
| recordtype | eprints |
| repository_type | Digital Repository |
| spelling | curtin-20.500.11937-700292018-08-08T04:56:27Z Pill Recognition Using Minimal Labeled Data Wang, Y. Ribera, J. Liu, C. Yarlagadda, Sri Kalyan Zhu, Maggie © 2017 IEEE. Inappropriate medication use such as wrong drug or wrong dose intake can be harmful to patients. In this work we present a method to automatically identify a pill from a single image using Convolutional Neural Network (CNN). We first localize the pill in the image by detecting the region with the highest concentration of edges. To overcome the challenge of minimal labeled training data and domain shift from the training images taken under the controlled lab environment to the consumer images taken under natural living conditions, several data augmentation techniques are applied on the Region of Interest to generate synthetic pill images for training the CNN. We adopted GoogLeNet Inception Network as our main classifier. Three GoogLeNet models with different specialties on color, shape and feature are trained on the augmented dataset. We evaluate our proposed method with a publicly available dataset provided by National Institute of Health that contains 1000 different pill classes. 2017 Conference Paper http://hdl.handle.net/20.500.11937/70029 10.1109/BigMM.2017.61 restricted |
| spellingShingle | Wang, Y. Ribera, J. Liu, C. Yarlagadda, Sri Kalyan Zhu, Maggie Pill Recognition Using Minimal Labeled Data |
| title | Pill Recognition Using Minimal Labeled Data |
| title_full | Pill Recognition Using Minimal Labeled Data |
| title_fullStr | Pill Recognition Using Minimal Labeled Data |
| title_full_unstemmed | Pill Recognition Using Minimal Labeled Data |
| title_short | Pill Recognition Using Minimal Labeled Data |
| title_sort | pill recognition using minimal labeled data |
| url | http://hdl.handle.net/20.500.11937/70029 |