Description

The boundaries of agricultural fields are essential features that define agricultural units and allow one to spatially aggregate information about fields and their characteristics. This information includes location, shape, spatial extent, and field characteristics such as crop type, soil type, fertility, yield. Agricultural data is an essential indicator for monitoring agriculture policies and developments; thus, they need to be up-to-date, accurate, and reliable. Mapping the spatiotemporal distribution and the characteristics of agricultural fields are paramount for their effective and sound management. However, automatic recognition and delineation of farm fields from satellite imagery is not a simple task. Standard techniques based on edge detection and image segmentation are not accurate enough.

Objectives and Methodology

For this MSc research topic, the student will resort to deep learning methods based on convolutional neural networks to investigate agricultural fields' delineation [1], [2]. These methods allow us to effectively learn the spatial-contextual features to detect and delineate field boundaries accurately. One of the latest and most promising deep learning approaches consists of a deep network that can directly extract regularized object boundaries in a polygon (vector) format [3]. This strategy allows us to include the algorithm within an interactive scheme to edit the algorithm results easily. The approach has been successfully applied to the delineation of buildings [4]. This assignment aims to investigate the applicability of the system to delineate fields using open image data acquired by Sentinel-2. The initial experiments will be conducted over a case study area in the Netherlands. If successful, the students are encouraged to apply the method to a more challenging scenario involving the delineation of smallholder farm fields in Africa (Fig.2). The integration of the technique in Google Earth Engine is considered a plus.

Further reading

[1] K. M. Masoud, C. Persello, and V. A. Tolpekin, "Delineation of Agricultural Field Boundaries from Sentinel-2 Images Using a Novel Super-Resolution Contour Detector Based on Fully Convolutional Networks," Remote Sens., vol. 12, no. 1, p. 59, Dec. 2019.
[2] C. Persello, V. A. Tolpekin, J. R. Bergado, and R. A. de By, "Delineation of agricultural fields in smallholder farms from satellite images using fully convolutional networks and combinatorial grouping," Remote Sens. Environ., vol. 231, Sep. 2019.
[3] Z. Li, J. D. Wegner, and A. Lucchi, "Topological Map Extraction from Overhead Images," in ICCV, 2019.
[4] W. Zhao, I. Ivanov, C. Persello, and A. Stein, "BUILDING OUTLINE DELINEATION: From VERY HIGH RESOLUTION REMOTE SENSING IMAGERY to POLYGONS with AN IMPROVED END-TO-END LEARNING FRAMEWORK," in International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives, 2020, vol. 43, no. B2, pp. 731–735.