Crop Classification Using Deep Learning on Time Series SAR Images: A Survey

Synthetic aperture radar (SAR) images are an advanced alternative to all other traditional methods because they can collect data on soil moisture and crop structure while also penetrating clouds. Over the past 10 years, crop classification using a variety of deep learning techniques using time series data of SAR images has become an active research area. Deep learning models have been widely adopted for crop classification on this data. In this approach, the temporal information of SAR images over a certain period is exploited to capture the dynamic changes of crop growth stages. Various deep learning models have been proposed for crop classification on SAR image data, including transformers, convolutional neural network (CNN), and recurrent neural network (RNN). The performances of these models have been evaluated using different SAR datasets, such as ALOS PALSAR, RADARSAT-2, and Sentinel-1. The reported accuracies of these models range from 86 to 94.2%. These findings highlight the potency of deep learning models for precision agriculture and food security, as well as their efficacy in crop categorization using SAR image time series.