Automatic Plant Water Stress Phenotyping for Genotype Classification Using Lightweight CNN

The utilization of phenotypic characteristics observed in plants subjected to water stress holds significant potential for enhancing drought-tolerance plant breeding programs and enabling precise water stress detection. Contemporary advancements in computer vision technology, particularly deep learning algorithms, have emerged as an advanced and high-throughput approach for automated plant phenotyping. Convolutional neural networks (CNNs) have garnered substantial attention as a powerful tool for image analysis and processing. However, conventional CNN models predominantly exhibit large-scale architectures demanding substantial computational resources. In this investigation, we employed lightweight CNNs to discern various genotypes of Setaria plants by leveraging distinctive features (phenotypes) extracted from corresponding images. Our study introduced a customized lightweight CNN model, equipped with an embedded attention mechanism, which demonstrated superior performance in genotype classification, achieving an impressive accuracy rate of 98.38%. Additionally, the training time of our proposed model was comparable with that of smaller parameter models, while delivering enhanced computational efficiency. These encouraging outcomes establish a significant milestone in genotype classification using lightweight CNNs, representing the foremost published findings in this domain.