Study On Apple Leaf Diseases Detection Method Based On Improved Faster R-CNN

As one of the important cash crops in Our province and even the whole country,apple occupies an important position in agricultural economy.In the process of planting,apple is affected by a variety of diseases,among which the leaf part of photosynthesis is greatly affected by diseases,thus affecting the economic benefits of apple.Therefore,the efficient and convenient detection of apple leaf disease has a very important role.The traditional apple leaf disease detection requires a lot of image preprocessing and segmentation,which is of great workload and lack of flexibility.The color,shape,texture and other information of apple leaf disease area are complex.In view of the above problems,this topic is to study three common apple leaf diseases,mainly build suitable apple leaf disease detection model for research,improve the detection efficiency of apple leaf diseases,the specific research content is as follows:Aiming at the problem that the target of Apple disease spot detection is small and the detection accuracy still needs to be further improved,this paper proposes a network model based on the combination of Faster R-CNN model and feature pyramid.When extracting target features,the model fuses deep semantic information with gradual feature information,reduces the loss of feature information and improves the accuracy of small target extraction.Through a series of experimental comparison,the results show that in the public data set voc2012,the average accuracy of this model is 3.03% higher than that of Faster R-CNN model.Map increased by 3.88%;On the extended Apple data set,map is 2.43% higher than the unexpanded Apple data set,which shows the feasibility of this model for apple leaf disease detection.Aiming at the problems of extracting feature parameters,large amount of calculation and long detection time of fused feature pyramid,this paper proposes a problem based on the combination of SE-Res Net-50 network and feature pyramid.Based on the fusion of deep feature information and shallow feature information,the network uses se module to solve the dependence between channels,so that the model has more nonlinearity and alleviate the complex relationship between channels,The amount of parameters and calculation are greatly reduced.Through a series of experimental comparison,the results show that on the public data,the average accuracy of this model is 6.15% higher than that of Faster R-CNN model,and the map is 3.18%;On the extended Apple data set,the map is 2.05% higher than the unexpanded Apple data set,indicating the superiority of the apple leaf disease detection model.Finally,an apple leaf detection system is designed based on the Faster R-CNN-SE-FPN model in this paper.Through 150 apple leaf disease pictures,the total detection rate is as high as 96%,which shows the feasibility of this system.