Research On Soybean Grass Condition Judgment Based On UAV Image

The precise management of soybean fields plays an important role in improving their yield.In order to achieve rapid and accurate identification and classification of weeds in soybean fields within a certain range under complex field environments,and to conduct grass condition judgment and analysis,based on the analysis and summary of the current research status of unmanned aerial vehicle low altitude remote sensing imaging perception technology in crop information collection,image processing technology,and convolutional neural network algorithm in crop image information detection at home and abroad,Propose a YOLO-WEED convolutional neural network algorithm based on deep learning.The main research content and results are as follows:By studying the competition between the main weeds in the soybean field and the soybean,as well as the control threshold,the optimal time for the experiment is determined by determining the appropriate period for the drone to collect images.Data annotation tools and information preservation are used to prepare the dataset for operation.According to the experimental requirements,digital images of weeds at different periods are enhanced and expanded using geometric transformation and color transformation methods,The production of a soybean weed identification dataset using Label Img image annotation software can provide data support for the soybean field weed identification model.The YOLO-WEED recognition model trained in the self built dataset is superior to the YOLO v5 l,YOLO v5 s,and YOLO v5 x models.The F1 value of the YOLO-WEED recognition model is 0.903,and the accuracy and confidence of the model are higher than the other three models.The average recognition time is the shortest,indicating that the YOLO-WEED recognition model has good recognition performance and can quickly identify weed targets in high-resolution images.By analyzing the network structure of Alex Net,VGG-19,and Res Net50,the optimal model for weed classification in soybean fields,Res Net50,was determined.The average accuracy of the model after convergence was 90.62%.The YOLO-WEED identification model was used to analyze the number and area information of weeds in the field and fit the artificial results.The average correlation coefficient R2 of the linear regression model was 0.973,indicating a relatively high degree of fit.This indicates that estimating the number and area of weeds using the weed identification model estimation method can be used to analyze the distribution of weeds in soybean fields.Using Arcgis software to extract the longitude and latitude position information of visible light images,using Java Sript language to divide the land into chessboard grids,and drawing polygons based on the small grid corners obtained from the division.Java Sript synchronous programming and Cesium spatial polygon technology are used to display and deploy localized data distribution.