Study On Methods Of Crop Growth Phenotypic Parameters Quantization Based On Vision And Correlation With Yield

Computer vision theory appeared in 1950 s,mainly used for two-dimensional image analysis and recognition.Core and motivation of computer vision research mainly focus on extracting visual information of object,and to propose appropriate processing method is a key step for the vision information.We propose a systematic method for extracting automatically and accurately the required crop featuring parameters based on computer vision theory and technology.The structural parameters of maize have great significance for evaluating the growth status.In the actual production process,traditional measurement methods such as manual measurement can't obtain enough structural parameters of maize because of the large amount of maize individuals.People usually use linear fitting to measure approximately the maize leaf inclination angle,and few of them study the characteristics of maize individual growth.In this paper,we propose a method that can obtain automatically individual parameters of maize based on monocular visual image processing.We propose an adjusted GrabCut algorithm with embedded preprocessor to segment maize image,and the results show that the adjusted algorithm has more accurate segmentation results compared with the Watershed algorithm and the MeanShift algorithm.We design a thinning algorithm to obtain the structure from segmented images of the maize individuals,and then,an eight neighborhood search algorithm is designed to select the curve line segments of maize structure and can obtain the backbone image of individual plant.Compared with manually segmented backbone images,the average error rate is only 2.85%.Finally,by using the monocular vision measurement technology,the parameters of leaf inclination and maximum height of maize leaves can be obtained by quadratic model fitting,and the change regulations of leaf inclination and maximum height of maize leaves in the period of growth are analyzed.Binocular vision technology is employed to obtain the points cloud data of maize binocular images.Those points are denoised by the bilateral filtering,and then,the three-dimensional reconstruction of maize is completed.The height and width of maize plant can be obtained.The error rate is only 5.81% and 0.52% compared with the manual measurement results.Meanwhile,Harris corner detection and main stem extraction algorithm are applied to obtain the pitch number of maize.Then the change regulations of the height,the width and the pitch number for an individual plant in a growth period are analyzed by correlation analysis and stepwise regression,the experimental results are consistent with the actual situations.Finally,the correlation analysis of maize plant fruit and visual quantitative parameters is implemented.The results showed that the perimeter of the bottom of the fruit and plant height are the most significant with the correlation coefficient being 0.7477,0.7103,respectively,followed by the length of fruit and the number of maize pitch with the correlation coefficient being 0.6516,0.6761,respectively.The results can be used as a reference for maize growth monitoring and virtual plant research.