Research On Soybean Leaf Morphology Identification And Disease Diagnosis Based On Convolutional Neural Network

Soybean is one of the most important food and feed crops in China,and its yield and quality directly affect people's daily life.On the one hand,differences in soybean germplasm genes can cause different tolerances in soybeans.Soybean leaf morphology as an effective external expression of soybean germplasm genes,in order to support research on improving soybean varieties and improving soybean tolerance,Continuously and accurately identify soybean leaf morphology.On the other hand,soybean leaf diseases have always restricted the improvement of soybean yield and quality,and have brought great economic losses to farmers.In order to minimize the impact of disease,soybean leaf diseases need to be diagnosed quickly and accurately.Based on the above two reasons,in order to further improve the yield and quality of soybeans,it is of great significance to use image processing,computer vision,and deep learning methods to timely and effectively identify the morphology and disease of soybean leaves.This article takes soybean leaf images as research objects.First,a soybean leaf image acquisition device is set up to avoid non-linear deformation of soybean leaf images collected in the field.Then,the region of interest detection and the leaf segmentation extraction are performed on the collected images to avoid background noise.Disturb and prevent soybean leaves from sticking to each other.Finally,based on the Convolutional Neural Network(CNN)algorithm,research on the method of morphology identification and disease diagnosis of soybean leaves is realized.The main research contents and innovation results are as follows:(1)This article establishes a soybean leaf image acquisition device to detect the Region of Interest(ROI)of soybean leaves based on the collected images,and segment and extract the soybean leaf based on this region,effectively avoiding the influence of noise interference on soybean leaf morphology recognition and disease diagnosis.(2)This paper proposes a deep voting model that integrates multiple convolutional neural networks to achieve automatic identification of soybean leaf morphology.By using different voting weights to fuse multiple convolutional neural networks,and comparing with the classification and identification algorithms commonly used in the field of computer vision,explore the identification performance of the Deep Voting Model(DVM)proposed in this paper for soybean leaf morphology.The experimental results show that using the deep voting model proposed in this paper,the identification accuracy on the independent test set consisting of 960 pictures is 96.9%,which can better automatically identify the morphology of soybean leaves.(3)This paper proposes an improved Goog Le Net model to realize automatic diagnosis of soybean leaf diseases.By using different improved methods and initialization methods to train soybean leaf diseases,and comparing with other convolutional neural network diagnosis results,explore the diagnostic performance of the improved Goog Le Net model proposed in this paper for soybean leaf diseases.The experimental results show that using the improved Goog Le Net model proposed in this paper,the diagnosis accuracy on the independent test set consisting of 5160 pictures can reach 98.29%,which can well complete the automatic diagnosis of soybean leaf diseases.