Please use this identifier to cite or link to this item:
Title: Deep convolutional neural networks for object detection under rainy conditions
Authors: Du, Kaiwen
Keywords: DRNTU::Engineering::Electrical and electronic engineering
Issue Date: 2019
Abstract: As we know, artificial intelligence is developing rapidly nowadays. Particularly, with the appearance of the theory of deep learning, object detection is allowed to maintain a balance between accuracy and speed. However, many methods are affected under rainy days at present. The rain streaks are likely to resulting in the failure in detecting the objects. As a result, we developed a deep network for removing rain streaks from single images and improved it constantly. We used the ResNet framework as the parameter layers in the de-rain network. In addition, we used the detail image and negative residual mapping as the input of the parameter layers, to reduce the amount of information to be processed. After training the de-rain network, we placed VGG16 network before and after the de-rain network, respectively, in order to observe the effect of removing rain. According to our results, the confidence level increased significantly after removing rain. On top of that, some other filters were used to see the effect. We wish that more contributions about the improvement of the network can be done in the future. The setting of some parameters’ value influences the performance of this model a lot. If the setting of these parameters is optimal, we believe that the effect of de-rain will be almost perfect. Therefore, it is important to find the optimal setting, which can be gained by modifying the parameters and retraining the model.
Schools: School of Electrical and Electronic Engineering 
Rights: Nanyang Technological University
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:EEE Student Reports (FYP/IA/PA/PI)

Files in This Item:
File Description SizeFormat 
Final Report.pdf
  Restricted Access
3.5 MBAdobe PDFView/Open

Page view(s)

Updated on Jul 15, 2024


Updated on Jul 15, 2024

Google ScholarTM


Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.