Panicle Traits Measurement Of Rice Plant In Vivo

Rice is one of the world’s most important crops. Approximately one half of world’s population feed on rice. Yield has always been the key object of most of the rice breeding programs. Panicle is the reproduction organ of the rice plant where spikelets grow on. Panicle traits, including panicle number, panicle fresh weight, panicle dry weight, total spikelet number, filled spikelet number, plant yield and so on, directly influence rice yield. In the screening and evaluation of the rice varieties, measuring and evaluating panicle traits is essential.Traditional measurement of panicle traits mainly depends on manual operation, which is tedious, labor-intensive, subjective and error-prone. Utilizing modern technologies such as remote sensing technology and machine vision, it is feasible to automatically evaluate field rice yield of a large area or pot-grown rice plants, with the advantage of high efficiency and broad application. Concerning the researches about in vivo panicle trait evaluation in the field, remote sensing technology is utilized to predict field rice yield of a large area, which adopt specialized devices to acquire the spectral data. This technique is not capable of estimating the yield of a small area or individual rice plant. For individual port-grown rice plant trait measurement, academic publishment is unavailable. Meanwhile, modern breeding technologies are able to produce thousands of new varieties within a single day, and each variety needs to be measured to evaluate its value and potential for breeding and generalization. Therefore, a fast, simple and effective technique for in vivo rice panicle trait evaluation is urgently needed to withdraw the current limitations.This research aims to in vivo evaluate rice panicle traits for individual rice plant, including panicle number, panicle fresh weight, panicle dry weight, total spikelet number, filled spikelet number, and plant yield, using machine vision, image processing, pattern recognition and mathematical modelling. The main tasks of this work are:(1) Extraction of panicle regions from the rice plant using image processing algorithms. The algorithm should be effective and fast so as to achieve high-throughput measurement.(2) Recognition of panicle regions from the rest organs of the plant.(3) Extraction of panicle number and development of mathematical models for estimation of yield-related traits in vivo based on separation and recognition of the panicles. The results showed that, mean absolute error (MAE) for panicle number extraction was0.50and95.3%percent of the plants generated measuring error within±1. Prediction errors evaluated using5-fold cross validation were7.93%,7.37%,8.59%,7.72%and7.45%for panicle fresh weight, panicle dry weight, total spikelet number, filled spikelet number and plant yield respectively. These models are capable of estimating yield-related traits of rice plants grown in green house and outdoors. Posterior variance test indicated that the precision grade was the second grade (up to standard) for all the five regression models.The method presented in this work is capable of evaluating rice panicle traits in vivo for individual rice plant. This method would overcome the limitations in the current devices and researches, with no need of cutting off the panicles and threshing the spikelets. Integrated with automated growth, transportation and inspection platforms, this technique can be used for automatic and high-throughput evaluation of panicle traits. The main contributions includes:(1) This work presents a new method for in vivo, automatic and high-throughput measurement of panicle traits for individual pot-grown rice plant.(2) This work illustrates a new image processing pipeline for extraction and recognition of panicle region from the rest organs of the rice plant.(3) This work proved the new idea of estimating yield-related traits, including panicle fresh weight, panicle dry weight, total spikelet number, filled spikelet number and plant yield, based on projected area of panicle region, projected area of leaf area and stem area, and fractal dimension. This work would have the potential of accelerating the evaluation of panicle traits and would be a promising impetus for plant phenomics. The work would also be meaningful in revealing the function of some key genes of rice and be a powerful tool in enhancing the researches of functional genomics and crop genetic improvement in our country.