Plant Height Measurement and Tiller Segmentation of Rice Crops Using Image Processing

Plant phenotyping is the process of completely assessing the basic and complex characteristics of the plant, which includes its height and tiller count. Automated plant phenotyping offers an effective substitute to manual visual assessments because it allows a regulated image analysis that can be reproduced and enables a high throughput because of the automation. This is to address the lack in accuracy, reproducibility and traceability in manual phenotyping. With this, the researchers developed an image processing system that automates the measuring of height and the counting of tillers of a rice crop, more specifically the C4 rice. The system is done by applying HSV and Thresholding for preprocessing, Canny Edge Detection (tiller) and Zhang-Suen Thinning Algorithm (height) for the plant structure and the Euclidean Distance for measuring the height. Tiller counting is done by counting the cluster of pixels in a given region of interest. The initial outputs were compared to the values manually measured by IRRI researchers from 50 plant images. There was a percentage error of 17.25% for height and 34.02% for tiller count. Errors may have been caused by plant not being able to fit the image frame and in result cut some parts of the plant. Another would be the effect of yellow leaves being removed during the preprocessing which produces an incomplete plant structure image. There are also leaves that are long that tend to bend and these leaves are then detected as the base instead of the real base.