A deep learning approach to measure stress level in plants due to Nitrogen deficiency

Abstract

Stress due to nutrients deficiency in plants can reduce the agricultural yield significantly. Nitrogen, an essential nutrient, is a crucial growth-limiting factor and is the prime component of amino acids, proteins, nucleic acids, and chlorophyll. Nitrogen deficiency affects certain visible plant traits such as area, color, the number of leaves and plant height, etc. With the recent advancements in imaging technology, computer vision-based plant phenomics has become a promising field of plant research and management. Such imaging-based techniques are non-destructive and much faster with higher levels of automation. In this work, we have proposed an automatic image-based plant phenotyping approach for stress classification in plant shoot images. In this proposed phenotyping approach, a 23-layered deep learning technique is proposed and compared with traditional Machine Learning techniques and few other deep architectures. Results reveal that a simple 23-layered deep learning architecture is comparable to the established state of art deep learning architectures like ResNet18 and NasNet Large (having millions of trainable parameters) in yielding ceiling level stress classification from plant shoot images. In addition, the proposed model also outperforms traditional Machine Learning techniques by achieving an average of 8.25% better accuracy.