JP Journal of Heat and Mass Transfer
Special Issue II, Advances in Mechanical System and ICT-Convergence, Pages 89 - 98
(December 2020) http://dx.doi.org/10.17654/HMSIII20089 |
|
DEEP LEARNING NETWORK FOR MAIN OBJECT DETECTION COMBINING VGG NETWORK AND GRAD-CAM
Nae Joung Kwak, Sun Jin Kim and Dong Ju Kim
|
Abstract: In this paper, we propose an optimal deep learning network structure for detecting the location of the main object through weakly supervised learning. In the proposed network, convolutional blocks are added to improve the accuracy of detecting the location of the main object through weakly supervised learning. The additional deep learning network was trained by a weakly supervised learning method. And Grad-CAM was used to detect the location of an object. According to the number of added layers, performance of the detection of location of main object is analyzed and we determine the optimal network for the detection of location of main object. The performance of the proposed network was tested using the CUB-200-2011 data set and CND data set. When the Top-1 Localization Error was calculated, the result is obtained 49.87% for CUB-200- 2011 data set and 96.96% for CND data set. In addition, the proposed network shows higher accuracy in detecting the main object than the existing method. |
Keywords and phrases: weakly supervised learning, object detection, location detection, CNN.
|
|
Number of Downloads: 319 | Number of Views: 485 |
|