Quantitative Analysis Of Crop Phenotyping Based On Multi-view Automatic Imaging System

Crop variety breeding requires the support of a large number of phenotypic data.Intelligent,efficient and reliable crop phenotypic observation equipment and quantitative analysis methods can assist breeders to quickly acquire crop phenotypic characteristics.In this paper,based on low-cost non-penetrating imaging sensor and automatic control technology,a general multi-view automatic imaging system suitable for different crops,different growth periods and different plant parts(shoots and roots)was constructed.Furthermore,three-dimensional point clouds of crops were reconstructed based on the SFM-MVS algorithm,and the reconstruction strategy of three-dimensional point clouds of crops was optimized by evaluating the reconstruction efficiency and accuracy.On this basis,the intelligent extraction of plant phenotypic parameters was realized based on the reconstructed high-precision three-dimensional point cloud.The main results are as follows:(1)Universal high-throughput multi-view automatic imaging system.Based on non-standard mechanical automation design,motion control,imaging control,wireless communication and other technical means,a universal high-throughput multi-view automatic imaging system based on low-cost non-penetrating imaging sensors are designed and constructed.The results showed that the self-constructed multi-view automatic imaging system could efficiently acquire multi-view image sequences with hemispherical distribution around the plant(image efficiency of 3 min/plant,432 images from different perspectives).Through accurate control of imaging perspective,the success rate of three-dimensional crop reconstruction was effectively improved,and standardized,large-scale,systematic and automatic plant phenotype monitoring was realized.In addition,the system can flexibly adjust the camera imaging perspective and position according to the plant size,which is suitable for multi-perspective image acquisition of different crops(monocotyledon and dicotyledon),different growth stages(seedling to maturity),and different plant parts(aboveground part and root system).(2)Accuracy evaluation and strategy optimization of point cloud reconstruction.Based on the multi-view automatic imaging system and the SSFM-MVS algorithm,three-dimensional crop point clouds were reconstructed with images obtained from different perspectives and different camera numbers.The reconstruction efficiency and accuracy(Hausdorff distance)were evaluated,and the reliability of point cloud extraction phenotypic parameters was evaluated to optimize the reconstruction strategy of three-dimensional crop point clouds.The results showed that Potted plants with relatively loose structure and less shielding(seedling stage,bud stage,flowering stage,rapeseed at maturity stage),plants with relatively compact structure and more shielding above ground(cotton at flowering stage,rice at heading stage,jointing stage and wheat at filling stage),and plants with dense organs and serious shielding and more elongated above ground and roots(tillering stage and mature stage of rice)The root system of maize and rape at maturity stage),3～4,6 and 10 cameras were used as the optimal reconstruction strategy(Hausdorff distance was less than or close to 0.20 cm,and the sum of reconstruction time and Hausdorff distance normalized value was minimum).More reliable phenotypic parameters.(3)Quantitative analysis of phenotypic parameters of rapeseed aboveground and roots.Reconstruction of 3D point cloud in the upper part of rapeseed and root system based on multi-view image sequence;The intelligent algorithm was used to extract phenotypic parameters of rapeseed shoots and roots based on three-dimensional point cloud,and then the dynamic changes of phenotypic parameters of rapeseed aboveground and roots of different varieties under different nitrogen treatments were quantitatively analyzed.The results showed that the 3D point cloud model reconstructed based on this method could well restore the real 3D morphology of the upper part of rapeseed field and root system.By using Meshlab software and root global phenotypic parameter extraction algorithm,the dynamic variation trend of phenotypic parameters(plant height,width,convex hull volume and total surface area)and global root phenotypic parameters(root depth,root width,convex hull volume,root volume,surface area and total root length)of the upper part for rapeseed at different growth periods could be quantitatively analyzed.The root global phenotypic parameter extraction algorithm has good reliability.The phenotypic parameters of the rapeseed aboveground with high nitrogen fertilizer were higher than those of the rapeseed field without nitrogen fertilizer during the whole growth period.The root surface area,volume and total length of NY22 were higher than those of NZ1818 during the whole growth period.Based on the constructed general multi-view automatic imaging system,the optimized three-dimensional reconstruction strategy and the constructed intelligent extraction algorithm of plant phenotype parameters can meet the requirements of intelligent,efficient and accurate extraction and analysis of crop phenotype parameters.This technology improves the efficiency of systematic collection of standardized phenotype information and provides powerful hardware,software and technical support for researchers to monitor crop phenotypes.