Maize and sorghum are important cereal crops in the world. To increase the maize grain yield, two approaches are used: exploring hybrid maize in plant breeding and improving the crop management system. Plant population is a parameter for calculating the germination rate, which is an important phenotypic trait of seeds. An automated way to obtain the plant population at early growth stages can help breeders to save measuring time in the field and increase the efficiency of their breeding programs. Similar to what has been taking place in production agriculture, plant scientists and plant breeders have been looking for and adopting precision technologies into their research programs; and analyzing plant performance plot-by-plot and even plant-by-plant is becoming the norm and vitally important plant phenomics research and seed industry. Accurate plant location information is needed for determining plant distribution and generating plant stand maps. Two automated plant population detection and location estimation systems using different sensors were developed in this research.

A 2D machine vision technique was applied to develop a real-time automatic plant population estimation and plant stand map generation system for maize and sorghum in early growth stages. Laser sensors were chosen as they are not affected by outdoor lighting conditions. Plant detection algorithms were developed based on the unique plant stem structure. Since maize and sorghum look similar at early growth stages, the system was tested over both plants in greenhouse condition. The detection rate of over 93.1% and 83.0% were achieved for maize and sorghum plants from V2 to V6 growth stage, respectively. The mean absolute error and root-mean-error of plant location were 3.1 cm and 3.2 cm m for maize and 2.8 cm and 2.9 cm for grain sorghum plants, respectively.

Apart from using laser sensors, a 3D Time-of-Flight camera-based automatic system was also developed for maize and sorghum plant detection at their early growth stages. The images were captured by using a Swift camera from a side-view of the crop row without any shade during the daytime in a greenhouse. A serious of image processing algorithms including point cloud filtering, plant candidate extraction, invalid plant removal, and plant registration were developed for this system. By comparing with the manual measurement, for the maize plant, the average true positive detection rate was 89% with 0.06 standard deviation. For grain sorghum plants, the average true positive detection rate was 85% with 0.08 standard deviation.