Rapeseed Seedling Monitoring Based On UAV Visible Light Images

Seedling stage is the key stage of rape nutritional growth,timely access to rapeseed seedling information can provide farmers with a basis for decision-making,provide technical support for rapeseed seedling cultivation management,which is conducive to the healthy growth of rape.Seedling information mainly includes the number of seedlings per unit area,uniformity of growth,and growth indexes.Traditional seedling information is mainly obtained by experience and manual sampling,which cannot meet the rapid acquisition of seedling information.As an efficient,modern and intelligent technology,Unmanned Aerial Vehicle(UAV)remote sensing has become an important tool for crop growth monitoring and can provide effective data support for crop field management.In this paper,we took winter rapeseed as the research object,carry out the research on the detection of seedling number and estimation of rapeseed seedling growth index in winter rapeseed,and construct the growth index estimation models of rapeseed seedling count model,leaf area index(LAI)and chlorophyll content(SPAD)by using deep learning and machine vision technologies.The main research contents and results are as follows:(1)Rapeseed seedling counting method was proposed to achieve accurate counting of the number of rapeseed seedlings in a large,small and overlapping area.The images of rapeseed seedlings at the 3-leaf stage are collected by UAV,cropped and manually labeled to construct a dataset for rapeseed seedling detection.Based on the YOLOv5 target detection algorithm,the YOLOv5 network was improved to improve the detection accuracy of smaller and overlapping rapeseed seedlings by adding Coordinate attention mechanism(CA)and replacing the standard convolutional modules with GSConv modules.The improved YOLOv5 achieves 1.9%and 3.7%improvement in precision and recall on the test set compared to 96.2%and 93.7%of the original YOLOv5,respectively.Using the improved YOLOv5 network in combination with Deep Sort target tracking,a video-based count of the number of rapeseed seedlings was achieved.The final results showed that the average error between the proposed method and manual counting was 4.3%.(2)A model for estimating the growth index of rapeseed seedlings was constructed to enable monitoring of the growth condition of rapeseed seedlings.The visible vegetation index and texture features were extracted from 20m and 40m height images,and correlation analysis was performed with measured LAI and SPAD to filter combination of features for estimating LAI and SPAD.The SRRestnet super-resolution reconstruction method was used to reconstruct the 40m images to obtain more accurate growth index information.Three modeling methods,partial least squares(PLSR),random forest(RF),support vector machine regression(SVR),were used to estimate LAI and SPAD based on the features extracted from 20m,40m reconstructed and 40m images.The estimation accuracy of 40m reconstructed image(~2=0.768)was close to that of 20m captured image(~2=0.776),and40m reconstructed image can be selected for LAI estimation.SPAD estimation accuracy was positively correlated with the UAV image resolution.20m estimation accuracy was the highest(~2=0.538),followed by 40m reconstruction(~2=0.524)and finally 40m(~2=0.445).SPAD estimation of features extracted from 40m reconstructed images can be performed using RF method to balance operational efficiency and estimation accuracy.(3)Rapeseed seedling monitoring system.Py Qt5 was used as a development tool to integrate algorithms for seedling number detection and growth index estimation of rapeseed into a directly callable software.The software can detect the number of rapeseed seedlings,improve the image resolution of rape fields and estimate the LAI and SPAD of rapeseed plots.Users can input images or videos,and the software interface can give the values of rapeseed seedling number and estimated growth indexes.An effective tool for monitoring rapeseed seedlings was provided.In this paper,the number of rapeseed seedlings and the rapid monitoring of growth condition were achieved by collecting visible light images by UAV,combined with deep learning and machine vision.It provided feasible technical support for farmers to accurately grasp field seedling information and carry out field cultivation management.