Hierarchical classification for improving parcel-scale crop mapping using time-series Sentinel-1 data.

Area of Science:

• Agricultural Science
• Remote Sensing
• Geographic Information Systems

Background:

• Parcel-scale crop classification is crucial for precision agriculture.
• Utilizing hierarchical crop structures enhances classification accuracy.
• Time-series satellite data offers valuable insights into crop phenology.

Purpose of the Study:

• To develop a hierarchical classification framework for crop mapping using time-series Sentinel-1 data.
• To integrate crop hierarchy knowledge into a deep learning model for improved classification.
• To generate accurate parcel-scale crop type classification maps.

Main Methods:

• Preprocessing Sentinel-1 data and overlaying onto parcel maps for feature extraction.
• Constructing a hierarchical crop type system based on phenology and labeled samples.
• Developing a deep learning model to classify crop types within a hierarchical structure.

Main Results:

• The hierarchical classification framework achieved over 4.0% improvement in overall accuracy.
• The F1 score saw an improvement of over 5.0% compared to traditional methods.
• The study demonstrated effective multi-scale time-series feature learning for crop classification.

Conclusions:

• The proposed hierarchical approach significantly enhances parcel-scale crop classification accuracy.
• Leveraging crop hierarchies and Sentinel-1 data is effective for precision agriculture.
• The deep learning model successfully utilizes hierarchical crop knowledge for improved mapping.