The food supply requirement of a growing global population leads to an increasing demand for agricultural crops. Without enlarging the current cultivated area, the only way to satisfy the needs of increasing food demand is to improve the yield per acre. Production, fertilization, and choosing productive crops are feasible approaches. How to pick the beneficial genotypes turns out to be a genetic optimization problem, so a biological tool is needed to study the function of crop genes and for the particular purpose of identifying genes important for agronomy traits. Virus-induced gene silencing (VIGS) can be used as such a tool by knocking down gene expression of genes to test their functions.

The use of VIGS and other functional genomics approaches in corn plants has increased the need for determining how to rapidly associate genes with traits. A significant amount of observation, comparison, and data analysis are required for such corn genetic studies. An autonomous maize functional genomics system with the capacity to collect data collection, measure parameters, and identify virus-plants should be developed. This research project established a system combining sensors with customized algorithms that can distinguish a viral infected plant and measure parameters of maize plants.

An industrial robot arm was used to collect data in multiple views with 3D sensors. Hand-eye calibration between a 2D color camera and the robot arm was performed to transform different camera coordinates into arm-based coordinates. TCP socket-based software written in Visual C ++ was developed at both the robot arm side and the PC side to perform behavioral bidirectional real-time communication.

A 3D time-of-flight (ToF) camera was used to reconstruct the corn plant model. The point clouds of corn plants, in different views, were merged into one representation through a homogeneous transform matrix. Functions of a pass-through filter and a statistical outlier removal filter were called from the Point Cloud Library to remove background and random noise. An algorithm for leaf and stem segmentation based on the morphological characteristics of corn plants was developed. A least-squares method was used to fit the skeletons of leaves for computation of parameters such as leaf length and numbers.

After locating the leaf center, the arm is made ready to position the 2D camera for color imaging. Color-based segmentation was applied to pick up a rectangular interest of area on the leaf image. The algorithm computing the Gray-Level Co-occurrence Matrix (GLCM) value of the leaf image was implemented using the OPENCV library. After training, Bayes classification was used to identify the infected corn plant leaf based on GLCM value.

The System User Interface is capable of generating data collection commands, 3D reconstruction, parameter table output, color image acquisition control, specific leaf-probing and infected corn leaf diagnosis. This application was developed under a Qt cross-platform environment with multithreading between tasks, making the interface user-friendly and efficient.