Corn Canopy Image Segmentation And Growth Identification Method-Based On Convolutional Neural Network

Based on the image of corn growth monitoring and phenotypic analysis,it plays an increasingly important role in modern maize breeding,production and management.The corn canopy images collected in the field of daetian are generally affected by complex weather conditions,which makes the automatic processing of the image data in the whole growth period of maize face great challenges.How to extract the structure of maize and the information of the growth period with high quality from the coronal image sequence in the whole growth period of maize is one of the hot topics in the current crop phenotype.In the past decade,deep learning has made remarkable achievements in the field of computer vision,and it is accelerating its penetration into a wide range of industrial,medical,transportation,and agricultural applications.This paper takes corn canopy image as the research object in the whole growth period of maize,and combines with the deep convolutional neural network to propose an automatic segmentation method of corn canopy image and automatic identification method of growth period,which provides technical support for the automatic monitoring system of field corn growth.The main research contents and conclusions of this article are as follows:(1)In view of the vulnerability of corn canopy image to weather,shooting time and other factors,the traditional segmentation method cannot be effectively segmented automatically.The Maize-SegNet model was designed by using the deep convolutional neural network for the height invariance characteristics of image geometric transformation and ambient illumination,to achieve automatic segmentation of corn canopy images in the field.The Maize-SegNet model is based on a codec-based deep convolutional neural network model.The model uses a coding network to automatically extract features from the input image,and then uses the decoding network to restore the feature map to the size of the input image.Finally,Softmax was used for classification,and the pixel-level image segmentation was achieved through end-to-end training.Among them,using the Relu function to speed up the convergence of the model;then use the batch standardization algorithm to regularize,enhance the model's generalization ability.The Maize-SegNet method is compared with the traditional methods EGI,Hue,and NDI.The experimental results show that the Maize-SegNet method has better segmentation results.The pixel accuracy rate of the Maize-SegNet model reached 92.93%,and the average IoUI wasimproved to 75.9%,indicating that this method is more suitable for corn canopy image segmentation at different growth stages,and is more robust than the segmentation method based on color and pixel statistics.(2)Due to the significant changes in morphological structure during the whole growth period of maize,the traditional method of predicting growth period using color and morphological characteristics has certain limitations.Therefore,a method for identifying maize growth period with deep convolutional neural network was proposed.Firstly,the image datasets of maize growing period from the five-leaf stage to the nineteen-leaf stage were constructed;then the features were automatically extracted by using deep convolutional neural network to obtain a more comprehensive description of the characteristics of the growth period of maize;finally,the classification was carried out using Softmax to realize the growth of maize.Period for identification.Compared with the method of maize growth period identification based on the support vector machine,the maize growth identification method proposed in this paper has a higher accuracy of identification.