Notice
This record is in review state, the data has not yet been validated.
Hierarchical Crop Mapping from Satellite Image Sequences with Recurrent Neural Networks
Date
2024-01-01Type
- Book Chapter
Altmetrics
Abstract
This chapter reviews different deep learning methods for image sequence analysis and provides an introduction to recurrent network architectures. It describes a recurrent network approach for mapping agricultural areas that is hierarchical . The approach exploits a tree-structured label hierarchy built by domain experts and encodes image evidence via convolutional layers. The chapter reviews an existing approach in literature for crop mapping from satellite data and looks at techniques that counter imbalanced data distributions. It explains in-depth how to exploit a crop label hierarchy for improving model performance, especially for rare classes. The chapter evaluates the proposed deep learning methods both quantitatively and qualitatively. It presents a hierarchical classification approach for multi-temporal crop classification from satellite images and demonstrates how the design of the Stackable Recurrent Cell can help training deeper recurrent neural networks models. Show more
Journal / series
Multitemporal Earth Observation Image Analysis: Remote Sensing Image SequencesPages / Article No.
More
Show all metadata
Altmetrics