Research On Automatic Measurement Platform Of Potted Maize Stem Diameter Based On Deep Learning

Maize is one of the main crops in our country.Researchers often need to measure the phenotypic data of Maize such as stem diameter for the selection of seed.Among the current methods of automatic crop stem diameter measurement,the most useful way is to collect crop image through darkroom,however,it costs a lot and it’s inefficient of carrying pots.Because of diverse environment,the direct and automatic measurement of maize stem diameter outdoors meets great challenge.Uncertain factors,such as lighting condition and weeds,can directly affect the accuracy of automatic navigation and the accuracy of maize stem diameter measurement.In order to solve the difficulties existing in the automatic measurement of maize stem diameter outdoors,this topic,which is based on deep learning,accomplished the improvement of automatic navigation system,the identification system of potted plant license and the measurement system of maize stem diameter respectively.And then we realized the automatic measurement of potted maize stem diameter.The main research contents and results are listed as follows:(1)The improvement of automatic navigation system based on semantic segmentation network.This topic uses the existing automatic navigation platform and the Enet semantic segmentation network to distinguish the straight road,turning road and background in the picture.By improving the annotation method,a better model is trained,which can realize the function of autonomous turning based on straight navigation platform.The system uses gyroscope module to assist visual navigation.By using proportional differential algorithm,the yaw angle of the navigation system changes between-1 ° ～ 2 °,and the standard deviation is 0.85 °.(2)The implementation of pot license detection and recognition system based on SLPNet neural network.In this system,The camera is mounted on the automatic navigation platform to realize automatic navigation as well as detect and identify the potted plant license at the same time.And then we enhance the data,so that the model is suitable to different lighting conditions.The detection accuracy of pot labels is 97.8%,and the license recognition accuracy is about 91.3%.(3)The implementation of automatic measurement system of potted maize stem diameter based on depth camera.The Realsense D435 i depth camera is used to obtain the depth information and color information at the same time,The fixed threshold method is used to segment the maize plant in the depth image.After binarization,the segmentation result is masked with the RGB image to segment the maize plant in the RGB image.Then the maize stem in the processed RGB image is accurately segmented.Through the conversion of color space,the RGB image is converted into YCr Cb color space,and the fixed threshold method is used to segment the maize stem,so that the segmentation method of maize stem can be better adapt to the change of light intensity.Finally,according to the coordinate transformation,the actual stem diameter of maize in the world coordinate system is calculated from the depth information in the depth camera and the segmented stem binary map.Through the experiment,the average measurement error of maize stem diameter is 0.12 mm,and the average relative error is 2.61%.(4)The experiment is designed for measuring the stem diameter of potted maize based on automatic navigation system and potted license recognition system.As well as automatic navigation system is working,the identified pot license is matched with the corresponding maize stem diameter one by one to realize the on-line automatic measurement of pot maize stem diameter.The average measurement error of maize stem diameter of the final system is 0.50 mm,and the average measurement relative error is0.12%.At the same time,the accuracy of pot license recognition was 93.3%.