University of Twente Student Theses
Crop Type Mapping Using Multitemporal Hyperspectral Images and Deep Learning
Huang, Yuchian (2024) Crop Type Mapping Using Multitemporal Hyperspectral Images and Deep Learning.
|
PDF
2MB |
| Abstract: | This study investigates the application of hyperspectral imagery and deep learning techniques for crop type mapping. With global food security threatened by population growth and climate change, accurate crop type mapping becomes essential for estimating crop yields and optimizing agricultural management. Traditional methods using multispectral images face limitations in distinguishing crops with similar spectral characteristics. This research addresses these challenges by leveraging the high spectral resolution of hyperspectral imagery and the temporal analysis capabilities of deep learning models. The study area in Belgium provided a diverse range of annual crops. Hyperspectral datasets from the PRISMA satellite and labeled training samples from the Flemish government were utilized to classify the different crops. Two deep learning models, 2D CNN for single-date imagery and 3D CNN for multi-temporal imagery, were developed and evaluated. Results demonstrated that the 3D CNN model significantly improved classification accuracy by capturing the temporal dynamics of crop growth, achieving higher overall accuracy compared to the 2D CNN model. Most crops, including corn, “flax and hemp”, sugar beets, “grains, seeds and legumes”, and “vegetables, herbs, and ornamental plants”, exhibited improved classification performance with multi-temporal images compared to single-date images. However, for crops prone to classification ambiguity, such as grasslands and forages, the benefits of multi-temporal data are less pronounced. Early stopping techniques further enhanced the models' generalization capabilities, reducing overfitting. The findings explore the potential of integrating hyperspectral and multi-temporal data for precise crop classification, offering valuable insights for agricultural monitoring and management. |
| Item Type: | Essay (Master) |
| Faculty: | ITC: Faculty of Geo-information Science and Earth Observation |
| Programme: | Geoinformation Science and Earth Observation MSc (75014) |
| Link to this item: | https://purl.utwente.nl/essays/103427 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page