Research On Classification Algorithm Of Potato Field Diseases And Pests Based On Digital Images

Potato is one of the most promising high-yield cash crops in China since the new century,and the frequent occurrence of potato diseases and pests often directly leads to a large reduction or low yield of this.Timely discovery and effective control of diseases and pests have an important role in ensuring the normal growth and health of potato.The traditional manual detection and diagnosis methods mainly rely on the naked eye observation of experts which cannot detect quickly and effectively.Therefore,the use of image processing technology to detect and diagnose potato diseases and pests can better meet the needs of farmers for disease diagnosis and prevention.This thesis took potato disease and pest images as the research object,and used machine learning and deep learning methods to extract key features to confirm the types of potato diseases and pests.First,use camera to take pictures in the field and establish a database.Then the image is preprocessed and segmented by K-means clustering method.Finally,the improved LBP and transfer learning methods are used to extract features and classify the image of potato diseases and pests.The main research contents and conclusions of this article are as follows:(1)Collect and establish an image database of potato diseases and insect pests.In this thesis,image collection equipment such as digital cameras and mobile phone cameras are used to collect images in the field.The collection locations include Wuchuan County,Helinger County,and Chahar Youyizhong Banner of Inner Mongolia.3 types of diseases and 4 types of insects were collected,a total of 1400 original image.In order to further expand the image database,the original images were rotated and flipped,and a total of 7000 images were obtained.(2)A segmentation method of potato diseases and pests based on G-R components and K-means was realized.The pictures collected in the field are easily affected by factors such as illumination and noise,and the background is complex.This method can accurately extract the target area from the background,and obtains an ideal segmentation effect,which lays a good foundation for subsequent classification and recognition.(3)The image classification experiment of potato diseases and pests based on BILBP and CILBP was realized.Two methods are used to extract the texture features of the high-frequency reconstructed image obtained after wavelet transform,and to classify them.Finally,compared with bilinear interpolation LBP,cubic interpolation LBP can more uniformly allocate the extracted feature vectors,thereby improving classification accuracy.(4)The image classification experiment of potato diseases and pests based on Inception v3 model transfer learning was realized.The image data set of potato diseases and insect pests was imported into the pre-trained model,retain all the parameters in the model,retrain the last layer,and use the softmax classifier for classification.It was proved that the method based on Inception v3 model transfer learning is effective for the classification of potato diseases and pests images by experiments.