Crop Classification with Synthetic Multitemporal Satellite Images
Synthetic Multitemporal Images in CNNs
Published in: Remote Sensing (Volume 13)
Publisher: MDPI
DOI: 10.3390/rs13173378
Abstract: The demand for new tools for mass remote sensing of crops, combined with the open and free availability of satellite imagery, has prompted the development of new methods for crop classification. Because this classification is frequently required to be completed within a specific time frame, performance is also essential. In this work, we propose a new method that creates synthetic images by extracting satellite data at the pixel level, processing all available bands, as well as their data distributed over time considering images from multiple dates. With this approach, data from images of Sentinel 2 are used by a deep convolutional network system, which will extract the necessary information to discern between different types of crops over a year after being trained with data from previous years. Following the proposed methodology, it is possible to classify crops and distinguish between several crop classes while also being computationally low cost. A software system that implements this method has been used in an area of Extremadura (Spain) as a complementary monitoring tool for the subsidies supported by the Common Agricultural Policy of the European Union.
This peer reviewed paper introduces a novel approach to crop classification by generating synthetic multitemporal and multispectral composite images from Sentinel 2 satellite data, specifically designed as inputs for Convolutional Neural Networks.
The Innovation
Standard crop classification approaches feed time-series data sequentially into neural networks. We proposed a different strategy: encoding temporal information spatially. By compositing multiple acquisition dates into a single synthetic image, each spectral band from each date becomes a separate channel in a multi-layer image stack.
This transforms the temporal classification problem into an image recognition problem: exactly the type of task CNNs excel at.
Architecture
We designed custom CNN architectures tailored for these high-dimensional inputs. The synthetic images contain far more channels than standard RGB images, requiring careful architectural choices around kernel sizes, pooling strategies, and depth.
The models were trained on terabytes of Sentinel-2 imagery covering the Extremadura agricultural region (specifically focusing on Summer crops in the Mérida-Don Benito area), classifying crops across multiple growing seasons.
Results
The synthetic image approach achieved strong classification performance, demonstrating that temporal encoding through spatial composition is a viable and effective strategy for remote sensing applications.
Citation
Published in MDPI Remote Sensing, August 2021.