Detection Of Rose Leaf Diseases And Insect Pests Based On Deep Learning

With the improvement of people’s living standards,the scale of China’s flower industry and the number of people who raise and enjoy flowers are expanding year by year.Flowers are widely used in environmental protection,landscaping courtyard,edible medicine,edifying sentiment,export sales and so on.Rose is the most popular flower variety with the highest yield.It has considerable industrial scale and complete industrial chain.However,as an unavoidable factor in the process of rose cultivation,rose diseases and insect pests are closely related to the final yield and ornamental value of rose.At present,rose diseases and insect pests prevention is mostly through manual fixed-point observation and highly dependent on individual experience,for large-scale rose planting base,this method has high labor cost,long time-consuming,low efficiency,and high learning cost.At the same time,it is difficult to provide the corresponding guidance and help for the initial practitioners and planting enthusiasts,and there is no practical and effective way for Rose diseases and insect pests diagnosis.Therefore,the research on rose diseases and insect pests detection by deep learning and computer vision has application value,which is an important way to help the rose planting base to dectct rose disease and insect pest and implement the application for rose diseases and insect pests detection.In this thesis,rose diseases and insect pests as the research object,aiming at the problem of multi leaf diseases and Insect Pests Detection in natural environment,a step-by-step rose diseases and insect pests detection method PDDIP(Phased Detection of Diseases and Insect Pests)is proposed.The rose diseases and Pests Detection in natural background is divided into two steps: first,detect the rose leaves,and then detect diseases and pests detection on the leaf.According to the characteristics of rose diseases and insect pests and the characteristics of each target detection model,the improved optimization was carried out,so as to realize the rose multi leaf diseases and Insect Pests Detection in the natural environment.The main research contents and results of this article are as follows:(1)In leaf detection,aiming at the common problems of rose leaf occlusion and it is difficut to detect small leaf in the natural environment,based on the one-stage target detection model YOLOv3 which has the advantage of detection speed,Inception modules is designed and add to construct the YOLO-in model,which improve the detection accuracy of small-scale targets and occlusion targets by multi-scale fusion of leaf and building leaf dataset with occlusion labels.(2)In rose diseases and insect pests detection,the scales of diseases and insect pests are quite different,the features of small-scale diseases and insect pests such as aphids are not prominent enough in feature map,and the dense targets are difficult to detected.This thesis improves VGG-16 to optimize the capability of feature extraction of Faster R-CNN,with optimizing Anchor box size and NMS,On the basis of(1)leaf detection,pest detection was carried out,integratively improve the detection accuracy of small-scale target and dense pest and disease.(3)The PDDIP method is realized by the joint detection of(1)and(2),in which(1)the average detection accuracy of the Yolo-in combined with the classification marker method is 14.06% higher than that of the original YOLOv3 model;Compared with the original Faster R-CNN,the improved Faster R-CNN in(2)improved the detection accuracy by 6%.The results show that in the natural environment,In natural environment,the detection accuracy of PDDIP method is 14.48% higher than that of YOLOv3,and 8.74% higher than that of Faster R-CNN,and it can effectively reduce the false detection caused by areas outside the region of interest in the image background.