Research On Tomato Leaf Disease Identification Based On Convolutional Neural Network

Tomatoes are vulnerable to various diseases which will cause the decrease of tomato yield and the deterioration of fruit quality during its growth process.The primary prerequisite for the management of tomato diseases is to identify tomato diseases quickly and accurately so as to ensure the normal growth of tomatoes.The effect of traditional machine learning methods in identifying tomato diseases is easily affected by the experience of feature extraction personnel while deep convolutional neural networks can automatically extract features to improve the disease identification.At the same time,the disease identification method using the classification convolutional neural network model has poor performance in recognition effect under the condition of multiple targets and multiple diseases in the image.Therefore,this study proposes a multi-target tomato disease recognition method that uses a combination of target detection model and classification convolutional neural network model.In order to realize the recognition of multi-objective tomato diseases,this study carried out the theoretical research and model improvement on the convolutional neural network classification model and target detection model.The specific research contents are as follows:(1)This study improved the YOLOV3-Tiny model for target detection and cropping of tomato leaves in the image,and the obtained leaf images will be used to input the classification model one by one for disease recognition.The improved YOLOV3-Tiny model in this study was added SPP module,and I modified part of the network structure and the size of Anchor Box.The target detection data set of tomato leaves was used to train the model.After verifying the effectiveness of the improved method,the model was compared with the original YOLOV3 series,Faster R-CNN,SSD and other models in terms of accuracy and speed.It is proved that the improved YOLOV3-Tiny model in this study has high real-time performance and accuracy.(2)This study proposed an improved classification convolutional neural network model which is designed to recognize disease types in images and it is a parallel structure neural network model.This model modifies the network structure based on the Inception module,uses the ELU activation function instead of the RELU activation function,and introduces the Batch Normalization method.Subsequently,the model fuses the features extracted from the improved Inception model with the features extracted from the Resnet18 network in different dimensions to improve the accuracy of the model.The tomato disease data in Plant Village were used to train the network,and the model was compared with the classic convolutional neural network models such as Res Net18 and Alex Net to verify the performance of the model.(3)The tomato disease identification system that constructed images is based on Tkinter,making the system easier to use.On the basis of the two models above,this study devised GUI through Tkinter to further integrate the functions,so as to realize the detection and recognition of tomato leaf diseases with multiple objectives,thus facilitating users to quickly cope with the diseases.Based on the study of convolutional neural network theory,a multi-target tomato disease detection method is designed by the improvement of the target identification model and the image classification model and the integration of function.At the same time,a GUI application program is built to improve the practicability of the method,which provides some help to deal with the disease problem in the tomato planting process.