Root Segmentation And Length Measurement Of Soybean Seedling Based On Deep Learning

Root length is a better quantitative index reflecting the root absorption capacity of soybean seedlings,which is of great significance for soybean genetic breeding.Intelligent segmentation of root images is a key technology for automatically and quickly measuring root length.In order to solve the problems of over segmentation,uneven root edges,and disconnected roots caused by background interference such as water stains,noise,and unclear contrast in the root images of soybean seedlings during hydroponic or washing,and to meet the needs of rapid and accurate measurement of root length,the research on root segmentation and root length measurement of soybean seedling based on deep learning was carried out.The main research contents and results are as follows:(1)Construction and preprocessing of soybean seedling root image dataset.The high-precision scanner is used to collect the root image of soybean seedling under hydroponic and soil culture environments,distinguish the root and background in the above image by manual labeling,assign the corresponding label category,and save it as the corresponding png image to construct the soybean seedling root image segmentation dataset,and complete the division of training set,validation set and test set.In view of the problem that the network is prone to over-fitting,the data augmentation and data normalization preprocessing experiments of soybean seedling root image were carried out using the self-built soybean seedling root image segmentation dataset.The results show that the data augmentation operation can increase the diversity of root image training samples and provide data support for the subsequent research on the semantic segmentation of soybean seedling root image.(2)Construction of semantic segmentation model of soybean seedling root image based on improved U-Net network.Based on the basic theory of soybean root image segmentation,the U-Net network is improved.In the process of down-sampling,the double attention mechanism is introduced,and the Attention Gate mechanism is added in the skip connection part to enhance the weight of the root region and suppress the interference of background and noise.The feature map and category activation map are used to visually explain the model prediction process,and the residual background noise is removed through connected domain analysis.The results show that the Accuracy,Precision,Recall,F1-Score and Intersection over Union of the model are99.62%,98.83%,97.94%,98.37%and 96.83%respectively.The segmentation time of a single image is 0.153 s.This method can extract more accurate root edges and complete detailed information,which can achieve automatic and accurate root segmentation of different varieties of soybean seedlings in different culture environments,and provide theoretical basis and technical support for quantitative evaluation of root length parameter in soybean seedling stage.(3)Rapid and accurate measurement of root length in soybean seedling stage.Based on the segmentation model of soybean seedling root image,by comparing the effect of Hilditch serial thinning algorithm and Zhang-Suen parallel thinning algorithm on extracting the root post-processing image,the Zhang-Suen thinning algorithm was determined to extract the root skeleton,and the root length of soybean seedlings was calculated using the chain code statistical method.Using the Win RHIZO root analysis software and the soybean seedling root length measurement model,the root length data of 21 single variety root images,30 multi-variety root images under hydroponic environment and 30 multi-variety root images under soil culture environment were analyzed quantitatively.The first case:Spearman=0.9909,R~2=0.9979,RMSE=4.1626 cm;The second case:Spearman=0.9978,R~2=0.9989,RMSE=4.7781 cm;The third case:Spearman=0.9964,R~2=0.9988,RMSE=2.4136 cm.The results show that compared with the root length parameters of Win RHIZO software,the measured results of the algorithm model have high reliability and small error in different cultivation environments and different varieties,and the constructed model could meet the actual measurement requirements.(4)Design and implementation of root segmentation and root length measurement system in soybean seedling stage.Through the analysis of the system function,the root image segmentation software and root length measurement software are developed based on Py Qt5framework and Qt Designer tool using Python language.The results show that the system can realize the main functions of root image reading,intelligent segmentation of root images,root post-processing,root skeletonization,automatic measurement of root length,etc.At the same time,it can achieve the real-time display of root image operation effect,root processing time,root length and other key information,and output information to the computer for data storage,which has strong practicability.It can provide a reference for more convenient use of the model.