Multi-modal neural networks for tree species classification from remonte sensing data

The goal of the project is to develop a classifier that uses a neural network to classify tree species from multimodal remote sensing data. This data includes aerial images from a hyperspectral camera and lidar. The main challenge is the lack of class balance and the small amount of training data, due to the cost of ground-based inspections. A neural network with an encoder-decoder architecture featuring two dedicated encoders will be used to process data from different sensors simultaneously. Additionally, the classifier will utilize a weighted loss function and data augmentation through the automatic generation of (pseudo) labels. The method will be tested on a unique dataset available to IITiS PAN as part of a collaboration with MGGP Aero Sp z o.o. and the University of Warsaw.

Machine-translated
Przetłumaczone maszynowo