The Design And Implementation Of Forestry Pest Monitoring System Based On Deep Learning

pests outbreak will pose a great threat to forestry resources and ecological security,thus causing huge economic losses to the country.Therefore,it is necessary to monitor forest pests.Currently,there are few studies on forestry pests,and most studies are carried out based on stored grain insects and crop pests.Moreover,the main means of pest monitoring and control include trap trapping,pest situation measuring and reporting instrument forecast and manual identification and counting,which is time-consuming,laborious and error-prone.In this paper,the improved deep learning target detection model YOLOv4-TIA was used to extract features automatically and identify forest insect images,and on this basis,a forest pest monitoring system was designed and implemented.The main research work of this paper is as follows:Firstly,pest image datasets were constructed under two application scenarios.One was to collect and sort out images of five common forest pests in the wild natural environment from the network;Label Img was used to mark pest targets,with a total of 2,000 images;the other was the image of 7 kinds of forest pests under the background of trap,with a total of1973 images.Both datasets were expanded through data enhancement operation,and the final number of images reached 12,000 and 10,758 respectively,which can be applied to the related studies of forestry pest detection tasks based on deep learning.Secondly,the YOLOv4-TIA pest detection network suitable for forestry pest detection was constructed.The structure of CSPDarknet53 was improved by designing the Triplet Attention modules in the backbone network of YOLOv4,to capture cross-dimensional interaction and improve the feature representation generated by the standard CSPDarknet53,achieving to extract high-quality image features by strengthening the target feature representation without increasing the amount of network parameters;The feature fusion method of PANet in the YOLOv4 detection network was improved through skip connection and efficient multi-directional cross-scale connection,so as to better balance multi-scale feature information and obtain richer semantic information and location information,at the same time,Focal loss was used to optimize the loss function.By comparing the improved YOLOv4-TIA network with the other four models,it was found that the improved YOLOv4-TIA network has the best performance on the dataset of this study,and the m AP could reach 98.8%.Lastly,based on the YOLOv4-TIA network,a forest pest monitoring system was designed and implemented.The system mainly includes the functions of pest image information input,image file reading,pest detection,pest counting,pest early warning and pest situation analysis,which provides a new idea and basis for scientific forestry control.