Yield-SpikeSegNet: An Extension of SpikeSegNet Deep-Learning Approach for the Yield Estimation in the Wheat Using Visual Images

Abstract

High-throughput plant phenotyping integrated with computer vision is an emerging topic in the domain of nondestructive and noninvasive plant breeding. Analysis of the emerging grain spikes and the grain weight or yield estimation in the wheat plant for a huge number of genotypes in a nondestructive way has achieved significant research attention. In this study, we developed a deep learning approach, “Yield-SpikeSegNet,” for the yield estimation in the wheat plant using visual images. Our approach consists of two consecutive modules: “Spike detection module” and “Yield estimation module.” The spike detection module is implemented using a deep encoder-decoder network for spike segmentation and output of this module is spike area and spike count. In yield estimation module, we develop machine learning models using artificial neural network and support vector regression for the yield estimation in the wheat plant. The model’s precision, accuracy, and robustness are found satisfactory in spike segmentation as 0.9982, 0.9987, and 0.9992, respectively. The spike segmentation and yield estimation performance reflect that the Yield-SpikeSegNet approach is a significant step forward in the domain of high-throughput and nondestructive wheat phenotyping.