Deep Learning based image fusion for mapping coffee plantations in Brazil

M-GEO
M-SE
ACQUAL
M-SE Core knowledge areas
Spatial Information Science (SIS)
Additional Remarks

Suggested elective courses: Advanced Image Analysis, Radar Remote Sensing, Computer vision related courses

For this topic, the student must have good programing skills. 

Topic description

Brazil is the biggest coffee productors in the world, with millions of hectares of plantations. Mapping these plantations is very important for planning, optimizing crop management and harvesting, estimation of production, and ensure the sustainability. However, since coffee is usually mixed with other crops and usually cultivated in the shadow, mapping those plantations represent a big challenge. To tackle this challenge, multiple sources of data are needed. Therefore, this topic will focus on fusing multiple sources of freely available Remote Sensing data (like Optical images and SAR) for mapping coffee plantations in Brazil.

Topic objectives and methodology

The student will initially revise the state of the art of semantic segmentation models, image fusion and multimodal deep learning, aiming to find the most efficient and accurate models. The next step will be the comparison between those methods, and their optimization, aiming in this way to achieve a better model with a good computational cost and accuracy.

References for further reading

Silva, S.d.A., de Queiroz, D.M., Ferreira, W.P.M., Corrêa, P.C. and Rufino, J.L.d.S. (2016), Mapping the potential beverage quality of coffee produced in the Zona da Mata, Minas Gerais, Brazil. J. Sci. Food Agric., 96: 3098-3108. https://doi.org/10.1002/jsfa.7485

Tassis, L. M., de Souza, J. E. T., & Krohling, R. A. (2021). A deep learning approach combining instance and semantic segmentation to identify diseases and pests of coffee leaves from in-field images. Computers and Electronics in Agriculture, 186, 106191.