Project 54: Remote Sensing Image Classification with Deep Learning Methods
Contact Information:
Assoc. Prof. Zenghui Zhang
Email: zenghui.zhang@sjtu.edu.cn
Project Description and Objectives:
Remote sensing image classification aims to identify regions of unique or dominant land cover from their attributes of spectral signature, texture, and context. Examples include classifying images into water, buildings, forest, grass, road, and other classes. In general, remote sensing image classification techniques include unsupervised/supervised methods, pixel-based or object-based methods, and deep learning-based methods. Recent researches show that the deep neural networks, such as a fully convolutional network (FCN) and SegNet, can far outperform traditional segmentation methods providing with a large training dataset. This internship aims to realize the deep learning-based image classification methods and do some improvements with the dataset provided by the lab.
Eligibility Requirements:
Fundamentals of digital image processing.
Programming skills of MATLAB and Python.
Main Tasks:
Study the remote sensing concepts, principles, and traditional methods for image segmentation.
Realize the fully convolutional network (FCN) and test the performance on remote sensing image dataset.
Improve the FCN method with dense connection, pyramid pooling or multi-task training and do further experiments.
Website: