Research On Tomato Leaf Disease Recognition Method Based On Convolutional Neural Network

As an important crop widely cultivated in my country,tomato is not only a frequent visitor on the table of Chinese residents.Moreover,ensuring the safety of tomato planting is of great significance for promoting the growth of economic crops in our country and improving the income of farmers.However,due to the weak disease resistance of tomato,it is easy to get sick.This seriously restricts the development of the tomato industry.The traditional manual identification of tomato diseases not only requires a lot of experience and time,but also the accuracy is not high.The method based on machine learning requires manual feature extraction,and the feature extraction is also affected by human subjective consciousness,so the practical application is also very difficult.With the development of technologies such as deep learning and artificial intelligence in recent years,these technologies can be used to automatically extract and identify diseased tomato plant leaves.Based on this,this paper takes tomato foliar disease as the research object,studies the tomato foliar disease identification method based on convolutional neural network,and proposes two models,improved Res Net50 and improved Efficient Net B0.The main research contents are as follows:(1)A tomato foliar disease identification method based on improved Res Net50 is proposed.This method inserts the SK attention module after each residual module of Res Net50.The adaptive receptive field of SK attention can help the model extract more effective features,and adds DCANet deep connection attention on the basis of SK attention.force mechanism.The original Relu activation function is also replaced with the Mish activation function.This paper uses the accuracy rate and F1 value to evaluate the model performance.The experimental results showed that the accuracy of the model for identifying tomato foliar diseases reached 97.89%,and the F1 value reached 0.972.And the effect is better than other convolutional neural network models.(2)In view of the fact that the benchmark network Res Net50 used in the previous chapter has a large number of parameters and is difficult to run on some devices with low computing power,an improved Efficient Net B0 identification method is proposed.Efficient Net is a model that can take into account both the amount of parameters and performance.The specific improvement method includes introducing a lightweight attention module ECA,replacing the SE attention module of the MBConv module in Efficient Net B0 with the ECA attention module,followed by using Ghost convolution to replace the ordinary convolution used in MBConv for dimension increase and Efficient Net B0 The first convolutional layer.After experiments,the recognition accuracy reached 95.25%,and the F1 value reached 0.942.Compared with the original Efficient Net B0,the number of parameters is reduced by about 1.2M while the accuracy is slightly improved.Compared with the convolutional neural network model proposed in the previous chapter,in the case of slightly lower accuracy and F1 value,the number of parameters is only about 1/8,and for the actual use problem,this paper builds a tomato based on Py Qt5 For the leaf disease identification system,the user only needs to click the upload button to upload the picture to the system,and then click the identify button to get the type of disease and the corresponding probability.