Maize Plant Recognition And Leaf Inclination Angle Extraction Based On Image Machine Learning

The development of big data provides an opportunity for the development of deep learning.Studying the target detection in deep learning is of great significance for the development of intelligent agriculture,realizing the target detection in agricultural field,and further strengthening field management and providing high-precision and low-cost phenotypic analysis tools for agricultural workers.As one of the phenotypes,leaf dip angle is an important feature of maize canopy.Reasonable distribution of leaf inclination angle on population scale is of great significance to enhance light transmittance,reduce light leakage loss,intercept light radiation and improve photosynthetic use rate.(1)In this study,YOLO data set was constructed based on maize images from different regions,and maize plants were identified based on YOLOv5 s deep learning algorithm,and the prediction accuracy increased with the increase of data volume.For different data sets,the optimal learning rate is different.When the learning rate is set to 0.0001,the highest precision can be achieved for most data sets.Compared with the effect of adding different data to enhance data set on model precision,the effect of brightness enhanced data set on model precision is greater.(2)In this paper,the prediction accuracy of YOLOv5 s model is improved by adding CBAM,an attention module.The optimal learning rate of the improved model is 0.0001.The accuracy of the model increases with the increase of data samples.In view of the influence of adding different data to enhance the performance of the model,the contrast-enhanced data set can improve the precision of the improved model more obviously.The improved YOLOv5 s algorithm improves the precision average precision by 2%.(3)After digital image processing,including gray-scale,histogram equalization,median filter,Otsu threshold,morphological operations,connectivity identification and skeletonization,Freeman codes was used to identify maize leaf joints and then leaf inclination angles were extracted.Leaf apex and joints were correctly identified by 94.47% and93.94%,respectively.The correlation coefficient between observations and calculations is 0.94.This method showed high efficiency and accuracy in leaf inclination angle extraction.Maize leaf inclination angle decreases with the lowering of leaf positions,and increases with the growth of NDVI.The quick and precise extraction of leaf inclination angle will provide scientific basis for phenotype retrieval,growth and development monitoring and agricultural resource use.