Research On Key Techniques Of Illumination Model On The Canopy Of Apple Trees

The apple industry is a key industry in the development of agricultural economy in China.So,the study of the growth factors of apple trees plays a vital role in increasing the income of farmers,enhancing the development of region economy and improving the progress of science and technology of our country.Thanks to the devoted and diligent researchers on studying apple tree's growth environment,light distribution and tree structures.There has been great progress in quality and quantity of apple industry.What's more,a good structure of tree canopy can absorb enough light source,which has satisfied the photosynthesis and growth of plants.Therefore,the study of illumination model of plants canopy has been one of hotpots in digital plants treatments.In the view of the problem of low calculation accuracy and low calculation efficiency in the indoor and out door illumination model of plants canopy,the research and analysis of the high precision and high efficiency of the light distribution of the plants canopy needs to be solved urgently.So,the goal of this study is to build a model of illumination distribution in three dimensional space of potted plant and apple tree canopy based on two kinds of potted plants named Guangdong Rohdea japonica and Peperomia as well as outdoor apple trees.Moreover,the research results can provide theoretical method for the scientific pruning of plants and management of apple orchard,as well as promoting the development of the apple industry further.After analyzing and summarizing previous research,this work mainly contains:(1)The schedule of plants point cloud data acquisition and preprocessing was completed.The two sides of point cloud data of two kinds of potted plants named Guangdong Rohdea japonica and Peperomia as well as outdoor apple tree were obtained through Kinect 2.0.The point cloud data of plants was denoised by the methods of pass through filter and down sampling voxel grid.The point clouds of potted plants leaves were segmented.And the plants point cloud data was aligned through the way of geometrical feature constraint of neighborhood.As for the apple tree,the plants skeleton was extracted by improved spatial colonization algorithm.And last,the plants pointd cloud data was reconstructed.This procedure can supply accurate input data for the following study.(2)On the problem of calculation of light distribution model of plants canopy,an improved ray tracing algorithm was proposed.This algorithm can be divided into 3 parts.Firstly,calculating the location and intensity of the light source;Secondly,simulating the interception of the light by the plant;Thirdly,outputing and the light intensity distribution of plants.Since that raytracing algorithm can reflects the process of intersecting between light and objects in the nature,it can be applied to simulate the light distribution of plant canopy reasonably.(3)As for the problem of low efficiency of traditional ray tracing,a parallelized ray tracing was proposed.Since the experimental scenary is static,and ray tracing algorithm needs to divide the space of the object in advance.However,building a spatial subdivision structure constructed by KD-tree is time-consuming.Therefore,we use GPU to accelerate the process.Besides,ray tracing based on parallelized computing was performed through memory copy.In this paper,a parallel computing platform CUDA(Compute Unified Device Architecture)based on GPU,which greatly reduces the threshold of parallel programming and provides a powerful parallel computing interface.(4)The analysis and verification of the light distribution model of canopy of plants were performed.In order to verify the effectiveness of the proposed algorithm,a controlled experiment was designed to analyze the light distribution between the virtual plant grid model and real plants.Firstly,the light meter was used to get the light intensity of the real plants;Secondly,after the reconstructed plant model was processed by ray tracing algorithm,the illumination information is distributed on the two-dimensional image;Finally,the difference between simulated and real illumination was calculated.The experimental results showed that,the RMSE between simulative light intensity and the real one of two kinds of potted plants as well as the apple tree are 0.1872,0.0085 and 0.0519,respectively.Moreover,the technology of KD-tree space subdivision based on GPU is faster than the original one,which increases by 81.76%,and the whole algorithm's efficiency is 6.8 times of that on CPU.