LARGE-SCALE GENOMIC PHENOTYPING OF RICE CROP HEALTH: INTEGRATING SENTINEL-2 NDVI COMPOSITES FOR CHARACTERIZED TRAIT SELECTION

Abstract

Crop management practices that helps to detect the presence of stress or any abnormality at the right time are essential to ensure food security and maximum agricultural output. The crop stress leads to massive loss of crop yield since it affects the plant health. Stress in crops is caused by poor irrigation, deficiency or any infection by pests. Modern agricultural methods manual scouting to determine the health condition of the crop is not efficient and not viable to large scale farming or large-scale monitoring of the area by the government agencies. This paper offers a dependable and effective method of determining and mapping stresses area within the farmed area. An unsupervised clustering framework that is based on deep learning was chosen to segment stressed area that provides valuable benefit as it can detect clusters of pixels that share certain common features. Unsupervised learning method does not presuppose previous knowledge or marked training data. It thus becomes an effective and scalable technique to study the health of crops, and it is also possible to map the levels of different stresses in the field without any problem. To begin with, an Auto-Encoder was applied to produce a small lower dimensional representation of the multi-band images. The AE model uses the abundant spectral information of several bands of Sentinel-2 to extract salient features that describe the underlying biophysical attributes of the crops. Then the encoded features are further grouped with k-Means clustering algorithm to outline the healthy vegetation, moderately stressed and stressed vegetation areas. The efficacy of the proposed approach was checked with the help of multiple performance measures that illustrate the effectiveness of this approach to offer a scalable solution to crop remote monitoring. This solution helps farmers to implement specific interventions and improve the management of the farms.