Research On Monitoring Models And Methods Of Wheat Seedlings Based On Deep Learning

The growth state of wheat seedlings has an important impact on the entire growth and development process of wheat.The timely and accurate acquisition of wheat seedlings information provides guidance for a series of wheat management measures such as replanting,irrigation,and pesticide spraying in the later stages of wheat.At present,the acquisition of seedling condition information of wheat seedlings mainly relies on manual observation,which is time-consuming and laborious,and does not have timeliness.Therefore,how to quickly and accurately monitor the condition information of wheat seedlings is a problem that needs to be solved urgently in current wheat cultivation and management.This paper studies the application of drones to quickly obtain images of wheat seedlings,and uses deep learning methods to detect the images of wheat seedlings.Then,threshold segmentation methods,morphological methods,clustering algorithms,and deep convolutional networks are used to obtain acres of wheat seedlings.Seedling number,individual seedling classification,ground coverage and other information.The specific work is as follows:(1)Research on the detection model of wheat seedlings.Use drones to acquire multi-angle images of wheat seedlings,crop the images,and build a training data set.In order to quickly and accurately detect the individual wheat seedlings in the image,this study uses a self-made data set in VOC and other formats,adjusts the training parameters according to the characteristics of Faster RCNN,SSD,YOLOv3,and YOLOv4,and optimizes the YOLOv4 algorithm based on the characteristics of the data set.,Trained 5 target detection models.Experiments show that the optimized YOLOv4 performs best in terms of speed and accuracy,with an average detection accuracy of 93.89% and a detection speed of 1.317 s per image.(2)Research on monitoring methods of wheat seedling situation.Using maximum entropy threshold segmentation algorithm,image corrosion expansion method and clustering algorithm to obtain the position information of the sampling block in the image,combined with the detection result of the wheat detection model,the number of wheat seedlings in the sampling block is obtained,and the experimental field is obtained according to the area of the sampling block.The number of seedlings is around 55280.The grading research of the individual wheat detected by the detection model was carried out,and the individual grading model of wheat was trained,and the classification accuracy rate on the test set was 95.07%;tested on the sampling block image,the grading model can correctly classify the number and level of individual wheat at each level.The Supergreen algorithm and threshold segmentation algorithm are used to process the image of wheat seedlings taken vertically by the drone,and then binarized,and the results are statistically analyzed.The average ground coverage of the wheat seedlings in the experimental field is obtained as 3.411%.(3)Based on the research of wheat seedling condition monitoring method,the TKinter module of Python was used to develop a prototype system of wheat seedling condition monitoring,which was tested on the real data set to verify the stability of the system;and the system was tested on the image of the sampling block to verify the system’s stability.Effectiveness and robustness.