3D Fully Convolutional Neural Networks with Intersection Over Union Loss for Crop Mapping from Multi-Temporal Satellite Images
This addresses crop mapping for food security studies by improving accuracy over existing methods, though it is incremental as it builds on known neural network approaches.
The paper tackled crop type mapping from multi-temporal satellite images by using a 3D Fully Convolutional Neural Network with an Intersection Over Union loss function, achieving a Kappa coefficient of 91.8% for identifying soybean and corn in the US corn belt.
Information on cultivated crops is relevant for a large number of food security studies. Different scientific efforts are dedicated to generating this information from remote sensing images by means of machine learning methods. Unfortunately, these methods do not take account of the spatial-temporal relationships inherent in remote sensing images. In our paper, we explore the capability of a 3D Fully Convolutional Neural Network (FCN) to map crop types from multi-temporal images. In addition, we propose the Intersection Over Union (IOU) loss function for increasing the overlap between the predicted classes and ground reference data. The proposed method was applied to identify soybean and corn from a study area situated in the US corn belt using multi-temporal Landsat images. The study shows that our method outperforms related methods, obtaining a Kappa coefficient of 91.8%. We conclude that using the IOU loss function provides a superior choice to learn individual crop types.