Research On Rice Pest Detection Method Based On Computer Vision

Rice is one of the most important crops in the world.Accurately identifying rice pests is the foundation for preventing and eliminating pests,and it is of great significance for improving rice yield and quality,and protecting the ecological environment.Traditional methods for detecting rice pests rely heavily on human resources,which can easily lead to the misjudgment of similar pests and cannot meet the current needs for the healthy growth of large-scale rice.In recent years,with the rise of the smart agriculture concept,an increasing number of researchers have used sensors such as cameras to capture images or videos and automatically detect rice pest hazards.However,existing models have insufficient accuracy in detecting rice pests with varying shapes or similar appearances.Therefore,this study conducts research on rice pest detection based on object detection models in computer vision,and proposed a rice pest detection model that introduces self-attention mechanism feature fusion,aiming to improve the accuracy of rice pest detection.The research content of this study includes the following aspects:(1)Four types of neural networks commonly used in object detection at present:Faster RCNN,SSD,Retina Net and VFNet are used to establish models,and rice pest data sets are used for model training.During the training process,multi-scale training,data enhancement and transfer learning strategies were introduced to enhance the training effect.The detection performances of these four models in the same experimental environment were compared,and the optimal model was selected based on the experimental results.(2)VFNet cannot achieve good results for some rice pests with variable shapes and extremely similar appearance.To address these issues,a deformable convolutional module and a self-attention mechanism fusion feature for rice pest detection are proposed,and their ideas and structures are introduced.(3)A rice pest detection model incorporating self-attention mechanism fusion features was proposed.This model is based on VFNet and adds a deformable convolutional module to the feature extraction network to improve the feature extraction ability for pests with variable shapes.Second,by acquiring the balance features of multiple feature maps,the self-attention mechanism was introduced to refine the balance features,aiming to better restore the semantic information of some similar pests.Then,the group normalization method was used to replace the batch normalization method in the original model,with the aim of reducing the impact of batch processing size on model training.The model was trained and validated using the IP102 rice pest dataset.The experimental results showed that the model proposed in this paper can accurately detect 9 types of rice pests,including rice leaf curlers and rice borers.Compared with Faster RCNN,SSD,Retina Net,and VFNet,the average accuracy of the algorithm increased by 33.7%,14.1%,6.5%,and 2.9%.