Deep Learning-Enhanced Monitoring and Analysis of Leaf Traits of Arabidopsis thaliana
Tracking dynamic changes in plant leaves using deep learning models represents a new approach to plant trait analysis. Combining deep learning techniques with botany and agronomy can be of great significance in the future. This indeed marks a crucial step towards addressing the increasingly prevalent problems in agriculture, especially considering that the issue of food scarcity represents a real problem we might face in the coming period. In this paper, we present an adaptation of the Speedy Measurement of Arabidopsis Rosette Traits (SMART) program - a robust, parameter-free system for plant image segmentation and trait extraction. Our goal was to optimize performance on a diverse dataset that differs significantly in characteristics from the one originally used to evaluate SMART. To this end, we replaced the traditional feature extraction methods with a custom-designed semantic segmentation approach. This modification enabled significantly improved results on our target dataset. Furthermore, the enhanced model offers promising potential for future applications, particularly in estimating plant developmental stages.