基于AI多模型技术对储粮害虫防治应用的研究  

Application of AI Multi-model Technology in Prevention and Control of Stored Grain Pests

作　　者：张壮 傅慧 王殿轩[2] 荆世华 李志信 王泽 刘曼 Zhang Zhuang;Fu Hui;Wang Dianxuan;Jing Shihua;Li Zhixin;Wang Ze;Liu Man(Wave Digital Grain Storage Technology Co.,Ltd.,Jinan 250098;Henan University of Technology,Zhengzhou 450001)

机构地区：[1]浪潮数字粮储科技有限公司,济南250098 [2]河南工业大学,郑州450001

出　　处：《中国粮油学报》2024年第11期17-25,共9页Journal of the Chinese Cereals and Oils Association

摘　　要：基于Transformer、高斯混合模型、ResNet18等AI深度学习、视觉分析等多维度模型,构建粮食储藏害虫的种群活动与防治评估数智化应用平台,实现了害虫在线监测、预测和防治评估。涉及害虫监测、害虫预测和储粮防治评估等3个核心业务模型:1)基于仓内视频流数据进行监测,通过高斯混合模型和三帧差分法进行害虫运动目标检测,基于ResNet18构建害虫分类模型,实现害虫数量和种类在线实时检测和确定虫粮等级;2)在害虫预测模块中提出了基于Transformer预测模型,结合害虫种群活动,基于粮温等数据建模对害虫发生概率预测;3)对害虫防治治理效果在害虫监测模型基础上,进一步对害虫监视视野场内的运动速度、活动姿态、活跃程度进行分析,评价害虫防治效果。Based on multi-dimensional models,such as AI deep learning and visual analysis,including Transformer,Gaussian mixture model and ResNet18,in this paper,a digital intelligent application platform was constructed for population activity and control evaluation of grain storage pests,and online monitoring,prediction and control evaluation of pests were realized.Three core business models,i.e.pest monitoring,pest prediction and grain storage prevention and control evaluation,were involved:1)Two-stage monitoring based on video stream data in the warehouse,detection of pest moving targets by Gaussian mixture model and three-frame difference method,and construction of pest classification model based on ResNet18 to realize online real-time detection of pest number and species and determination of pest grain grade;2)In the pest prediction module,a prediction model based on Transformer was proposed,and the probability of pest occurrence was predicted based on environmental data,such as grain temperature combined with pest population activity;3)The effect of pest control based on the pest monitoring model,the movement speed,activity posture and activity degree in the pest monitoring field were further analyzed to evaluate the effect of pest control.

关 键 词：AI多模型技术 害虫监测 害虫预测 防治评价 

分 类 号：S3[农业科学—农艺学]