Research On Monitoring And Early Warning Technology Of Rice Pests And Diseases Based On Machine Visio

Rice is one of the most important food crops in China and plays an extremely important role in solving China’s food security problems.China’s rice production has the characteristics of large planting area,wide distribution of planting areas,frequent outbreaks of diseases and insect pests,which has caused huge economic losses to China’s grain production,and the prevention and control of rice diseases and insect pests is facing great challenges.The traditional identification method of rice pests and diseases based on manual inspection and identification has problems such as low accuracy,high cost and low efficiency.In order to solve this problem,this paper takes the image of rice pests and diseases as the research object,based on the theory of deep neural network,and around the problem that the accuracy of rice pests and diseases recognition is reduced due to factors such as the change of image scale and the insufficient number of labeled samples in the actual scene,and carries out the research of rice leaf diseases diagnosis method based on multiscale feature fusion,and rice pests recognition method based on migration learning,and adopts the mainstream Springboot framework,Develop an early warning system for rice disease and pest identification based on mobile internet,and provide application demonstration for automatic identification of rice disease and pest.The research contents and innovations of this paper are as follows:(1)A rice leaf disease diagnosis method based on multi-scale feature fusion is proposed to address the issue of reduced accuracy in rice disease recognition caused by changes in image scale in practical application scenarios.Due to changes in shooting distance,the resolution of rice leaf images has changed.The traditional rice leaf disease recognition method based on deep convolutional neural networks only adapts to feature learning at a single image scale.The last few convolutional layer feature maps are concatenated and fused at the back end of the network,which is difficult to adapt to situations with significant scale changes.This article adopts a multi task learning framework to simultaneously perform feature learning on two related tasks of rice leaf disease type and disease degree discrimination,sharing underlying visual features and effectively improving the accuracy of rice leaf disease type and disease degree discrimination;Introduce channel and spatial attention models to improve the classic Mobile Net V3 network model,and construct a multi task deep convolutional neural network model based on feature pyramid.In addition,multiple data augmentation methods are used to expand the dataset and solve the problem of imbalanced sample distribution in the dataset.The experimental results show that compared with traditional methods,the proposed method in this paper improves the average accuracy of rice disease identification by 1%.(2)Aiming at the problem that the lack of labeled samples of rice pest images in practical application scenarios leads to the decline of recognition accuracy of rice pest images,a rice pest recognition method based on transfer learning is proposed.This article proposes a method of using semantic feature consistency to transfer the feature expression from a sufficient sample dataset to a small annotated dataset under the premise of a small number of annotated samples.Therefore,in practical application scenarios,high-precision identification of pest types in rice leaf images can be achieved with only a small number of annotated samples;In addition,this article constructs a challenging large-scale rice pest dataset,and enhances the data based on the dataset’s features.It introduces data augmentation methods such as random noise,Mixup,Cutout,to enable deep learning models to learn visual features of pest discrimination from deeper dimensions;By introducing a self-attention model,the YOLOv7 network is improved and a multi-scale neural network model based on feature pyramid is constructed to improve the recognition accuracy of small individual pests.The experimental results show that the improved YOLOv7 model has a1.6% improvement in recognition accuracy compared to the original model.(3)Aiming at the problem that the traditional convolution neural network model is too complex to deploy in the resource-constrained mobile computing environment,a model compression method based on knowledge distillation is proposed.In practical application scenarios,the traditional convolutional neural network needs to consume a lot of computing and storage resources,while in mobile computing environments such as mobile phones,computing,storage and power resources can not meet the demand.This paper proposes to use the knowledge distillation theory to transform the complex neural network model with high accuracy to the simple neural network model with low complexity,so that the simple network model has the ability to distinguish the complex model,but its consumption of resources is greatly reduced.On the basis of the above research,this paper establishes a set of early warning system of rice diseases and insect pests for mobile computing environment,realizes remote automatic diagnosis of rice diseases and insect pests,and provides precise technical guidance for early warning and prevention measures of rice diseases and insect pests.To sum up,the research work in this paper mainly solves the problems of multi-scale feature expression,knowledge transfer,model simplification and other problems of rice leaf image disease and pest disaster early warning technology under practical application scenarios,realizes the remote and rapid diagnosis of rice disease and pest,and provides technical reference for the early prevention and control of disease and pest in the agricultural field,which has certain theoretical value and good application prospects.