Development of robotic systems and sensors for Autonomous Image-based Plant Phenotyping

Advancing plant phenotyping requires automated, high-throughput systems capable of capturing high-quality imaging data for crop analysis. This dissertation focuses on developing novel sensors and robotic technologies to enhance autonomous plant phenotyping, integrating automation with precision imaging. First, a target-to-sensor multispectral imaging device is introduced for soybean phenotyping, using an airflow-based mechanism to reposition and flatten leaves for optimal imaging. This approach increases imaging speed fivefold while ensuring high-resolution and noise-free data collection. This device also has controlled lighting conditions which enable higher accuracy nutrient deficiency detection, surpassing traditional techniques. Second, a portable snapshot multispectral device for corn nitrogen treatment classification is developed. By autonomously flattening leaves and using a transmittance imaging method, this device captures high-resolution multispectral images in six seconds. The high-resolution of this device enables the analysis of the combined spatial and spectral data significantly improves nitrogen classification accuracy compared to conventional index-based methods. Lastly, an autonomous robotic phenotyping system integrates a high-resolution hyperspectral camera with a robotic arm and vision-based tactile sensors. This system autonomously detects, localizes, and scans individual corn leaves, achieving over 90% success in under one minute. Advanced tactile sensor and computer vision algorithms ensure optimal leaf localization and scanning, reducing manual labor while enhancing imaging precision. These innovations collectively improve the efficiency, accuracy, and scalability of plant phenotyping by enabling faster, automated data collection with superior image quality.