基于三维模型的谷子表型特征分析研究

Abstract

As one of the ancient cultivated crops in China, millet has the characteristics of high nutritional value, drought resistance and barrenness. It also plays an important role in ensuring the supply of food in our country. At present, most of the millet breeding work uses manual extraction of phenotypic information, which is laborintensive and inefficient. Therefore, the development of an automated, efficient and accurate millet phenotype detection method has practical significance for the extraction of the millet genome. In this study, a combination of sparse reconstruction based on Structure from Motion (SfM) and Patch-based Multi-View Stereo (PMVS) was used to select three different varieties of millet. A total of 81 samples of 9 samples in each period were reconstructed to obtain a 3D model of millet. The combination of conditional filtering and statistical filtering is used to remove the noise points generated during the photographing process, and finally the obtained point cloud data is used to measure the agronomic traits of millet such as plant height and leaf area. The results show that the interval angle of 5° is the best reconstruction angle of millet. The coefficient of determination R2 of point cloud measurement results and manual measurement data regression analysis is higher than 0.94, indicating that the method used for 3D reconstruction has high applicability to different millet in different periods and high-throughput measurement of millet by the method in this paper is feasible. This study provides a theoretical basis for a millet phenotypic information measurement device