基于卷积神经网络的原烟仓库虫情检测算法研究
2023,31(10):76-82
摘要：随着精细化管理的不断深入，传统的仓储养护管理模式也面临着巨大的挑战，原有的人工养护管理模式已经无法满足精细化和智能化仓储管理的需要。为此，以原烟仓储环节中虫情监测为探究应用方向，基于深度学习技术与目标检测技术，设计适用于烟草粉螟、烟草甲虫检测的检测网络，该网络采用轻量化结构设计，同时基于Transformer机制构建了主干网络，以实现快速、精准的检测目的。经过实验证明，该网络模型在GPUs上的检测速度达50FPS可实现实时检测、且检测精度达96.7%。通过实时获取的检测数据结合虫情检测系统能够快速评估出原烟仓储环节中的虫情实况，实现原烟仓储虫情监测管理的信息化和智能化，进而为打叶复烤行业仓储虫情监测管理提供可行参考方案。
关键词：卷积神经网络 、原烟仓库、虫情监测、深度学习、目标检测
Research on Pest Monitoring Detection Algorithm in Raw Tobacco Storage Based on Convolutional Neural NetworkJIANG Fuhuan1,YAN Xianghui2,FAN Mingdeng1
Abstract：With the deepening of the fine management, the traditional storage maintenance management model is also facing great challenges. The original artificial maintenance management model has been unable to meet the needs of fine and intelligent storage management. Therefore, in order to explore the research and application of insect pest monitoring in the storage of raw tobacco, a detection network, based on the deep learning technology and target detection technology, was designed for the detection of tobacco meal borer and tobacco beetle. The detection network is designed with lightweight structure. And in order to achieve fast and accurate detection, the main network is built based on Transformer mechanism. The experimental results show that the network can realize real-time detection with the detection speed of 50FPS and the detection accuracy of 96.7% in the GPUs. The insect pest situation in the raw tobacco storage link can be rapidly evaluated by the real-time detection data which combined with the pest detection system. It’s also be realized that the information and intelligence of pest monitoring management in the raw tobacco storage by the way. Then it provides a feasible reference scheme for monitoring and management of pest situation in storage of threshing and rebaking industry.
Key words：Convolutional;neural network,Raw;Tobacco Storage, Pest;Monitoring, Deep;Learning, Target;Detection
收稿日期：2023-04-13
基金项目：
