A systematic review of the use of Deep Learning in Satellite Imagery for Agriculture

TL;DR

This paper conducts a systematic review of 150 studies to map how deep learning is applied to satellite imagery for agricultural analysis. It categorizes tasks into crop segmentation, soil monitoring, plant physiology, crop damage, and yield estimation, benchmarking modern deep learning against traditional methods. The review finds that deep learning generally outperforms traditional algorithms across tasks, though LSTM variants do not consistently beat Random Forests for yield prediction, and cross-study comparability is hampered by the lack of standardized benchmark datasets. It also notes that many studies reuse generic computer vision methodologies and often underutilize satellite-specific properties such as multi-spectral resolution and diverse spatial scales, signaling opportunities for standardization and exploitation of spectral-temporal information in future work.

Abstract

Agricultural research is essential for increasing food production to meet the requirements of an increasing population in the coming decades. Recently, satellite technology has been improving rapidly and deep learning has seen much success in generic computer vision tasks and many application areas which presents an important opportunity to improve analysis of agricultural land. Here we present a systematic review of 150 studies to find the current uses of deep learning on satellite imagery for agricultural research. Although we identify 5 categories of agricultural monitoring tasks, the majority of the research interest is in crop segmentation and yield prediction. We found that, when used, modern deep learning methods consistently outperformed traditional machine learning across most tasks; the only exception was that Long Short-Term Memory (LSTM) Recurrent Neural Networks did not consistently outperform Random Forests (RF) for yield prediction. The reviewed studies have largely adopted methodologies from generic computer vision, except for one major omission: benchmark datasets are not utilised to evaluate models across studies, making it difficult to compare results. Additionally, some studies have specifically utilised the extra spectral resolution available in satellite imagery, but other divergent properties of satellite images - such as the hugely different scales of spatial patterns - are not being taken advantage of in the reviewed studies.