Image-based Analysis And System Implementation Of Leymus Secalinus Root System Parameters

The root system is the main organ for nutrient uptake and the first organ for adversity perception when the soil environment changes,and plays an important role in the whole cycle of plant growth.Leymus secalinus is a perennial grass in the genus Ragwort of the family Gramineae,which is characterized by high resistance and wide habitat adaptability,and its high resistance depends,to a certain extent,on its longitudinal and horizontal underground rhizome network and well-developed underground root system.Therefore,the collection and analysis of root system characteristics and the implementation of targeted agronomic cultivation measures are effective ways to increase yield and improve quality.At present,the root system phenotypic parameters of Leymus secalinus roots are mainly measured by the digging method,which has the disadvantages of low efficiency,affecting the growth direction and mistakenly injuring the root system,and it is also difficult to meet the demand of high throughput measurement.In this study,computer vision technology was used to extract and analyze the phenotypic parameters of the root system of Leymus secalinus roots in batch.In this thesis,we take Leymus secalinus as the research object,acquire and preprocess its root system images,and propose a Seg Net＿Root network based on a non-local self-attentive mechanism using deep learning method combined with semantic segmentation theory to realize segmentation of in situ images and acquisition of root system parameters.The main research contents and results of this thesis are as follows:(1)Production of the dataset.In this thesis,we take Leymus secalinus roots as the experimental object,obtain root images through hydroponic experiments,crop,noise reduction,and annotation of the collected root images,and expand the dataset using random cropping and other image enhancement methods,and the expanded dataset is 7200 images.The root images were segmented using five types of traditional segmentation methods:threshold segmentation,edge segmentation,region segmentation,cluster segmentation,and graph theory segmentation,and the experimental results showed that the traditional segmentation algorithms did not achieve pixel-level segmentation of the root images;(2)Seg Net＿Root network model is proposed based on the non-local self-attentive mechanism.The Seg Net network is optimized based on the Encoder-Decoder architecture fused with the non-local self-attention mechanism:multi-scale feature fusion is added to extract image features so that the model retains more semantic information;the non-local self-attention module is added to the decoder structure part of Seg Net so that the model can take into account the global contextual information.The Seg Net＿Root model is trained and optimized using the Py Torch deep learning framework,and the improved segmentation model can retain more fine root systems.Finally,the correlation analysis between the total root length obtained by Seg Net＿Root and the total root length calculated by Win RHIZO,where R~2=0.9719 and RMSE=3.4906 cm,proved the usability of the model;(3)Application of Seg Net＿Root root parameters analysis under adversity conditions.Through hydroponic experiments,sequence images were collected at 24h intervals,and the analysis of the root system(including rhizome)parameters of Leymus secalinus under adversity conditions was completed using the root analysis system.One control and three treatment concentrations were set up to investigate the effects of adversity conditions on the developmental parameters of the rhizosphere and rhizome components of Leymus secalinus roots.The experimental results showed that the salt tolerance of Leymus secalinus was mainly related to its rhizome components,and the efficiency of the model in extracting and analyzing the rhizome parameters was also verified;(4)Design and implementation of root system image analysis system.The system embeds the root image segmentation algorithm proposed in this thesis and provides users with an automatic root image segmentation and parameter analysis system to meet root analysis applications.It realizes the functions of automatic segmentation,calculation,plotting,and storage of the root system images of Leymus secalinus roots,which significantly improves the efficiency of the parametric analysis of root morphology.