Machine learning based approach for wheat plant senescence quantification

Abstract

Wheat plant senescence is the result of the natural ageing process but also due to unfavorable conditions such as water deficiency. Water deficiency induces senescence that directly relates to the yield as a cause to reduce fertile wheat ears and the number of grains per ear. For precision farming, it is highly desirable to develop genotypes tolerable to drought stress. For selecting the best genotypes tolerable to drought stress, there is a need to measure the senescence percentage. Traditionally measurement of senescence is manual and time-consuming. In this paper, image-based non-destructive approach is proposed for the quantification of senescence percentage. In this study, wheat plant image data was taken from Nanaji Deshmukh Plant Phenomics Centre ICAR-IARI and six machine learning algorithms, Naïve Bayes, KNN, Decision Tree, Random Forest, Gradient Boosting classifier, and Artificial Neural Network algorithms were trained. These algorithms are trained to segment the senescence portion from the wheat plant. All the algorithms performed well but ANN outperformed among the above trained algorithms with 97.28% testing accuracy. Machine learning-based proposed approach was compared with binary thresholding approach on wheat plant dataset and it was observed that machine learning based approach provided best results in the quantification of senescence. A desktop application, named as m-Senescencica, has been developed to facilitate senescence quantification using the traine machine learning algorithms and to visualize senescence across different plant growth stages.