Intelligent modeling method for plant morphology based on gene expression programming

Abstract: The plant simulation based on computer modeling and visualization has become an important topic in the scientific researches, such as the researches of computer graphics, agroforestry, and ecology. Due to the complexities of plant structure, especially in the modeling of large-scale scenes of natural environments, how to quickly and efficiently establish models of the scenes using computers has been a research focus in the area of plant modeling. It is a key step to select appropriate morphogenetic model to simulate morphology and architecture in plant modeling. However, it is needed to artificially extract the parameters of the model based on the priori knowledge of the plants in order to simulate the realistic plant morphology as required, no matter what kind of morphogenetic model is chosen. Larger number of parameters for the rules will be needed in the modeling of large-scale scenes. Extraction of the parameters for the rules based on the artificial method is time-consuming and laborious. Thus it is particularly important to develop a method for efficiently extracting the rule parameters in simulating different types of plants. In this study, an intelligent method for simulating and visualizing plant shape was proposed, aiming at solving the problems caused by blindness and low efficiency when only using L-systems to simulate plant shapes by manual way. The production rules and the initial axioms of the model with L-systems were obtained by this method. Then the spatial structure of specific plants based on the concepts of gene expression programming was simulated. We proposed a restrictive strategy to design the initial population with the control of the branch number and the morphology of individuals, which can be used to guide the evolution of simulated plants towards the target shape and reduce the searching scope with the algorithm. The method was developed based on the analysis of previous studies, the most of which were using traditional genetic algorithms to generate the initial population in a completely random way. Besides, we proposed a selecting strategy to automatically select the optimal individuals in each generation, and thus preserve better traits of the population, which can further improve the efficiency of the algorithm. Using the genetic manipulations to simulate the evolution processes, i.e. one point crossover, two point crossover, gene recombination, transposition, and mutation, the population with morphological diversity can be generated. We also proposed an individual fitness evaluation function which is integrated algorithm of plant outline comparison with Hausdorff distance calculation method. Combined the fitness evaluation function with the proposed evolutional algorithm, the optimal individuals in each generation can be selected, so that the evolution speed can be increased greatly. The method proposed in this study has been implemented with the graphics hardware, NVIDIA GeForce3, using OpenGL functions. The simulation results indicate that the proposed methods can not only simulate the special plant morphology, but also the normal plant morphology with various types. The method will promote the development of plant simulation models and also provide a reference in the exploring of new methods for virtual plant modeling.