Please use this identifier to cite or link to this item:
Title: Scene understanding based on heterogeneous data fusion
Authors: Zhu, Lingzhi
Keywords: DRNTU::Engineering::Electrical and electronic engineering
Issue Date: 2017
Abstract: Artificial intelligence has boosted human’s life; this technology has become something that will totally change people’s life in the future. Scene understanding is one of the most popular research areas under this topic. This project focuses on developing a high-performance deep learning neural network which could help scene understanding model perform well in image classification. This project uses convolutional neural network as the fundamental network architecture. Nearly ten thousand images are collected, and these images are classified into 20 different classes based on image descriptions. With pre-trained Keras VGG-16 model, several sets of features are extracted from different layers. New classifiers are created and trained by passing those features through. This network achieves 0.77 mean average precision without any fine tuning. Moreover, after fine tuning process, the highest mAP score it can reach is 0.804. Experiments on testing different variables are implemented, and the results are elaborated as well. Difference between these tests are discussed as well.
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 
Scene Understanding Based on Heterogeneous Data Fusion.pdf
  Restricted Access
Main Article4.29 MBAdobe PDFView/Open

Page view(s)

Updated on May 16, 2021


Updated on May 16, 2021

Google ScholarTM


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