Novel deep learning methods for remote sensing imagery

• Grant number: 23/11556-1
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Start date: January 10, 2024
• End date: December 17, 2024
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Hélio Pedrini
• Grantee: Leandro Stival
• Supervisor: Ricardo da Silva Torres
• Host Institution: Instituto de Computação (IC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Institution abroad: Wageningen University & Research, Netherlands
• Associated to the scholarship: 22/12294-8 - Convolutional Networks with Attention for Video Color Propagation, BP.DR

Abstract

The analysis of remote sensing images is a task that is constantly receiving new approaches, whether for the improvement of current methods for solving problems of soil classification, cloud detection or changes in relief. In order to develop new techniques for this field and to improve existing ones, this project consists of two approaches. For this purpose, the first proposed solution is an architecture capable of extracting features capable of representing multi-spectral images in such a way that these representations can be used in the most common problems in the area, such as those mentioned above. In this implementation, given the difficulty of accurately labeling this type of data, we will use techniques that are currently considered state-of-the-art in visual computing and unsupervised training. With the training of this model, capable of generating generalist features for multispectral images, we will implement the second proposed solution, an architecture capable of removing the clouds present in the images. This model will be able to generate general features for multispectral images with the training of this model. In the second solution proposal, we will implement an architecture capable of removing clouds present in the images. For this, as in the first solution, the self-attention technique and training with emphasis on the textures of the regions will be applied in the architecture. In this way, we intend to surpass the current work considered as state of the art. Initially, feature extraction techniques currently used in Ph.D. research will be used in both solutions. This reuse will allow us to validate the domain extension capabilities of our proposed architecture by introducing them to the remote sensing domain. At the end of this project, we intend to have implemented two tools for remote sensing, where the first one has an emphasis on pattern recognition, this approach being useful for other areas of visual computing. While the second task has a more specific content for the field of Earth observation. (AU)

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

STIVAL, LEANDRO; TORRES, RICARDO DA SILVA; PEDRINI, HELIO. Video Colorization Based on a Diffusion Model Implementation. INTELLIGENT SYSTEMS AND APPLICATIONS, VOL 1, INTELLISYS 2024, v. 1065, p. 15-pg., 2024-01-01. (23/11556-1, 22/12294-8)

STIVAL, LEANDRO; TORRES, RICARDO DA SILVA; PEDRINI, HELIO. Enhancing Video Colorization with Deep Learning: A Comprehensive Analysis of Training Loss Functions. INTELLIGENT SYSTEMS AND APPLICATIONS, VOL 1, INTELLISYS 2024, v. 1065, p. 14-pg., 2024-01-01. (23/11556-1, 22/12294-8)