POSTERS: Remote Sensing and Sensor Technology
Blackbird: A novel high-throughput laboratory phenotyping system to quantify incidence and severity of powdery mildews.
Dani Martinez - Plant Pathology and Plant-Microbe Biology Section, Cornell AgriTech. Surya Sapkota- Cornell University, Lance Cadle-Davidson- USDA Grape Genetics Research Unit, Mark Rea- Lighting Research Center, Rensselaer Polytechnic Institute, David Gadoury- Cornell University, Timothy Plummer- Lighting Research Center, Rensselaer Poly
Phenotyping of responses in the earliest stages of pathogen growth has required microscopy and observation by human experts, an expensive and slow process that severely restricted throughput. We developed a high-throughput phenotyping system that combined high-resolution, low magnification (4X) optics, a 47 MP 24X36 mm CMOS sensor with a pixel width of 4.2µm, long working distance, consequent maximum depth of field; and paired this with an X-Y-Z robotic stage, image-stacking to create fully-focused composite images, and a pre-trained convolutional neural network. Projecting a 4X image of the 4 µm-wide hyphae of the grape powdery mildew pathogen (Erysiphe necator) onto the camera sensor provided 4-pixel width coverage of the hyphal diameter. The system synoptically captures a fully-focused high-resolution image covering a 1-cm leaf disk in as little as 13 seconds. We named the system Blackbird, after the SR-71 reconnaissance plane. The current throughput rate ranges between 1200 to 2400 samples imaged per 8-hour workday. The assessments are also non-destructive; allowing repeated assessments of samples through multiple time points. Our present convolutional neural network estimates the percentage of infected area per sample at an agreement ratio of 91% compared to human expert assessments. Further QTL analysis showed that this automated phenotyping methodology has higher QTL association than the traditional manual method.