Research On Corn Disease Identification Based On GBDT Algorithm

As one of the main food crops,corn has been planted at home and abroad.Maize is currently the world’s highest crop yield and one of the most widely distributed food crops in the world.It ranks third in the world,second only to rice and wheat,and China’s planting area ranks second in the world.In recent years,China’s corn planting area is more than 3,500,000 hectares,and the total output exceeds 200 million tons,accounting for about 25% of China’s total grain output.The Northern Leaf Blight-Infected caused by fungal infection is one of the main diseases affecting corn production,and its damage has become more and more serious in recent years,covering many major corn producing areas around the world.In China,it is also called corn spot disease.According to statistics,the perennial disease area of corn spot disease in China has reached 69 million mu.Because the disease manifests as a large necrotic lesion,it is the best choice for remote detection with images.Due to the wide area of the disease,it is not easy to be detected at the beginning of the disease,and the cost of manual monitoring is high.And the disease traits are mainly manifested on the corn leaves,suitable for image recognition and detection and prediction,and large-scale monitoring by machine.So this paper designed a machine learning algorithm system based on gradient lifting tree to identify diseases,the main innovations are:(1)The distribution of image data in this study is highly unbalanced,and it is very difficult to establish an accurate classification for predicting samples,because classifiers tend to favor excessive representation or majority groups.The resampling technique of generating new samples by sampling aims to counter class imbalance by increasing the number of members belonging to minorities.Experiments show that after using the SMOTE algorithm to balance the data,the number of samples in the training set is increased,and the balance of the prediction model can also reach 1:1,which is useful.(2)In this paper,the GBDT machine learning algorithm is used to identify corn big spot disease.The data image is the largest single plant disease public data set in the world.Compared with Logistic Regression,Linear/RBF SVM,Decision Tree,Random Forest,Back Propagation,and Na?ve Bayes,the results show that the GBDT algorithm has higher accuracy for the corn disease identification problem,and the final recognition rate is above 92.5%.(3)Because the GBDT algorithm is slower to write and is written in Python,the Light GBMgradient promotion framework is used to speed up the iteration and increase the accuracy.Finally,based on the research basis of the previous chapters,this paper constructs a practical identification detection system with the function of detecting the northern leaf disease of corn,which has certain practical value.