Review of Crop Yield Estimation using Machine Learning and Deep Learning Techniques

Abstract

The agriculture sector is subjected to constant challenge of yield deficit due to rising population, improper resource management and shrinking agricultural land. Advance yield estimates help in systematic planning to reduce such losses. However, prediction of accurate estimates is still an open challenge due to geographical diversity, crop diversity and crop area. Recently non-destructive approach has gained attention due to its robustness and provides easy availability of data from heterogeneous resources compared to its counterpart; destructive approach which is computational, resource intensive and hence less utilized. This paper conducts a detailed study on utilization of non-destructive approach to estimate yield taking into account, input feature, and methodology. We consider five major observations namely, data acquisition, pre-processing techniques, features, methodology, and result. Moreover, we summarize analysis of each observation, extract most prominent technique, the adopted methods, and finally recommends integration of different models that can be explored to improve accuracy.

Code and data availability

This is a review/survey paper on crop yield estimation. All URLs in the text are cited third-party data portals and datasets (EarthExplorer, MODIS, USDA, MangoNet, SEN12MS, COCO, etc.) referenced as background resources, not the authors' own phenotype datasets, images, code, or models reproducing this paper's analysis.

No evidence-backed public reproduction asset is currently recorded.