Presentation | 2021-03-05 A Consideration on Suspicious Object Detection by Mixup and Improved U-Net Naruki Kanno, Wataru Kameyama, Toshio Sato, Yutaka Katsuyama, Takuro Sato, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In this paper, on suspicious object detection by using semantic segmentation, we study the effectiveness of Mixup data augmentation, propose several improved U-Net models, and compare them by precision. The target images are obtained by passive-imaging technology using W band. The objects to be detected are 9 classes including background, human, 6 types of suspicious objects (blade, gun, cell phone, liquid, powder, dummy bomb), and phantom. Annotation images are manually generated for the obtained 1,008 images, and taken as ground truth. On Mixup data augmentation, we compare it with conventional data augmentation methods including horizontal flip and scale augmentation by using U-Net, where mIoU (Mean Intersection over Union) of Mixup achieves 86.0% accuracy which is 8.4 point higher than others. On improved U-Net, we compare 7 models with Mixup, including normal U-Net, U-Net applying RB (Residual Block) to each encoder and decoder, U-Net applying RB to both of encoder and decoder, U-Net applying dense block to each encoder and decoder, and FC-DenseNet. The experimental results show that mIoU of U-Net applying RB only to decoder achieves the highest accuracy of 86.9%. These results suggest the effectiveness of Mixup data augmentation for the experimental images and applying RB for decoder to U-Net in order to reduce the noise in the experimental images. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Semantic Segmentation / Data Augmentation / Suspicious Object Detection / U-Net / Mixup |
Paper # | PRMU2020-90 |
Date of Issue | 2021-02-25 (PRMU) |
Conference Information | |
Committee | PRMU / IPSJ-CVIM |
---|---|
Conference Date | 2021/3/4(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Online |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | Computer Vision and Pattern Recognition for specific environment |
Chair | Yoichi Sato(Univ. of Tokyo) |
Vice Chair | Akisato Kimura(NTT) / Masakazu Iwamura(Osaka Pref. Univ.) |
Secretary | Akisato Kimura(Mobility Technologies) / Masakazu Iwamura(Chubu Univ.) |
Assistant | Takashi Shibata(NTT) / Masashi Nishiyama(Tottori Univ.) |
Paper Information | |
Registration To | Technical Committee on Pattern Recognition and Media Understanding / Special Interest Group on Computer Vision and Image Media |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A Consideration on Suspicious Object Detection by Mixup and Improved U-Net |
Sub Title (in English) | |
Keyword(1) | Semantic Segmentation |
Keyword(2) | Data Augmentation |
Keyword(3) | Suspicious Object Detection |
Keyword(4) | U-Net |
Keyword(5) | Mixup |
1st Author's Name | Naruki Kanno |
1st Author's Affiliation | Waseda University(Waseda Univ.) |
2nd Author's Name | Wataru Kameyama |
2nd Author's Affiliation | Waseda University(Waseda Univ.) |
3rd Author's Name | Toshio Sato |
3rd Author's Affiliation | Waseda University(Waseda Univ.) |
4th Author's Name | Yutaka Katsuyama |
4th Author's Affiliation | Waseda University(Waseda Univ.) |
5th Author's Name | Takuro Sato |
5th Author's Affiliation | Waseda University(Waseda Univ.) |
Date | 2021-03-05 |
Paper # | PRMU2020-90 |
Volume (vol) | vol.120 |
Number (no) | PRMU-409 |
Page | pp.pp.121-126(PRMU), |
#Pages | 6 |
Date of Issue | 2021-02-25 (PRMU) |