Research on application of insect species image recognition based on convolutional neural network

摘要:

昆虫种类图像识别是农业智能化识别虫害的重要方式,精准高效识别昆虫种类是进行针对性防治虫害的前提.利用昆虫数据集ArTaxOr及Insect_det,基于卷积神经网络下图像分类如MobileNet,ResNet及目标检测(FasterRCNN)、Yolo技术,运用迁移学习进行模型训练,并对比分析训练结果,获取最优昆虫种类图像识别模型.将构建的最优模型采用EasyEdge平台进行部署,从而实现了模型到端的全流程开发模式,为后续昆虫种类图像识别场景化应用研究提供依据参考.

Image recognition of insect species is an important way of agricultural intelligent identification of pests,and accurate and efficient identification of insect species is the premise of targeted pest control.In this paper,we use the insect datasets ArTaxOr and Insect_det,based on image classification under convolutional neural network,such as MobileNet and ResNet,and target detection under convolutional neural network,such as FasterRCNN and YOLO technology,uses transfer learning to train the model,compares and analyzes the training results to obtain the optimal image recognition model of insect species.The constructed optimal model is deployed on EasyEdge platform,thus realizing the whole process development mode from model to end,which provides a basis and reference for the follow-up research on the scene application of insect species image recognition.

作者:
魏甫豫 张振宇 梁桂珍

Wei Fuyu;Zhang Zhenyu;Liang Guizhen(School of Landscape and Ecological Engineering,Hebei University,Handan 056000,China;Academy of Fine Arts,Xinxiang University,Xinxiang 453000,China;School of Mathematics and Statistics,Xinxiang University,Xinxiang 453000,China)

机构地区:
河北工程大学园林与生态工程学院 新乡学院美术学院 新乡学院数学与统计学院

出处:
《betway官方app 学报：自然科学版》 CAS 北大核心 2022年第6期96-105,共10页

Journal of Henan Normal University(Natural Science Edition)

基金:
国家自然科学基金(11871238) 河南省科技厅科技攻关项目(132102310482) 河南省高等学校重点科研项目(20B110014) 国家级大学生创新训练项目(202011071015).

关键词:
昆虫种类图像识别 卷积神经网络 图像分类 目标检测 模型场景应用

insect image recognition convolutional neural network image classification target detection model scene application

分类号:
TP391.41 [自动化与计算机技术—计算机应用技术]

基于卷积神经网络下昆虫种类图像识别应用研究.pdf