Study On Rapid Identification Of Rice Leaf Diseases Based On Image Processing

Rice is one of the important grain crops in China.At present,the primary goal in rice production is to increase rice yield and quality.However,during the growing season,rice is prone to various diseases,resulting in a decline in yield and quality,which in turn brings huge economic losses.The correct identification of diseases is an important means to effectively reduce the economic losses of diseases.Traditional rice disease identification methods mainly rely on the experience of farmers,and they have the disadvantages of strong subjectivity and low recognition accuracy.Rice blast,sheath blight,and bacterial blight are three diseases with high incidence in the rice growth period,and the early symptoms of the three diseases are not very different.This study aimed at these three diseases,based on image processing technology to carry out research on identification methods of rice leaf disease.The main work includes:(1)Choose the smart phone with low cost,simple operation and high popularization rate as the disease image acquisition equipment;(2)Through image preprocessing methods such as graying,image denoising,and lesion segmentation,a rice leaf disease database was established.(3)Analyze the characteristics of the three lesions,extract the characteristic parameters from three aspects: color,shape,and texture,and use the color moments of the three components of the HSV color space as the color features of the lesion image;use the correlation function to obtain the target in the lesion image.The relevant attributes of the region are used to describe the shape features of the lesion;the description and selection of texture features are achieved by using a gray-level co-occurrence matrix;(4)On the premise of not affecting the accuracy of model recognition,the feature parameters were optimized.A total of 13 different feature parameters were selected from the aspects of color,shape,and texture.The single factor variance analysis method was used to distinguish all the extracted feature parameters.For the parameters with small discrimination,the verification results show that the parameters with small discrimination can be removed without affecting the recognition result.This effectively removes the redundant parameters.(5)Using the neural network and Bayesian classifier to construct the rice leaf disease recognition model respectively,three different rice leaf disease parameter sets were identified,and the identification effect was compared.The best model was selected.The results show that the recognition accuracy of neural network is obviously higher than that of Bayesian classifier.Therefore,this study will use the BP neural network as a model to identify the three rice leaf diseases.Among them,the disease recognition rate based on color parameters is below 80%,and the recognition rate based on shape and texture features is higher than that based on color.The highest recognition rate is the texture feature;in the combination of feature parameters,the disease identification.The highest rate is the combination of three parameters,reaching about 98%.This study satisfies the requirements of the development of contemporary precision agriculture,accords with the development direction of modern agricultural automatic diagnosis technology,has a good application prospect,and also provides a reference for follow-up research,which has far-reaching significance for the defense of rice diseases.