摘要：As a brand-new monitoring technology,Unmanned Aerial Vehicle (UAV) aerial photography has gradually been used to monitor forest diseases and insect pests in production due to its unique advantages such as large-scale and real-time monitoring.With the continuous development of artificial intelligence and big data technology,in view of low accuracy and efficiency of conventional forest pest identification methods,this paper proposes a forest aerial image pest identification method using improved GoogLeNet network based on transfer learning.Firstly,this paper builds a pest image collection platform based on an eight-rotor UAV,and establishes a dataset for pest identification in UAV aerial images.Secondly,based on GoogLeNet,an improved GoogLeNet deep convolutional network model was designed according to the characteristics of aerial images.Then,this paper adds a new network model composed of forward propagation and back propagation to replace the original support vector machine classification module.Finally,the experimental results based on TensorFlow framework show that improved GoogLeNet network model has an average recognition accuracy of 95.44% for forest pest images.Moreover,the pest identification model for forest aerial image using improved GoogLeNet based on transfer learning proposed in this paper has stronger robustness and applicability,and it can provide references for forest pest identification and intelligent diagnosis.
关键词：Aerial image pest identification;transfer learning;GoogLeNet network;deep learning;back propagation
