基于Logistic算法与遥感影像的棉花虫害监测研究  

Cotton pest monitoring based on Logistic algorithm and remote sensing image

作　　者：地力夏提•依马木 周建平[1,2] 许燕[1] 樊湘鹏 亚里坤•沙吾提 DILIXIATI Yimamu;ZHOU Jianping;XU Yan;FAN Xiangpeng;YALIKUN Shawuti(College of Mechanical Engineering,Xinjiang University,Urumqi 830047,China;Agriculture and Animal Husbandry Robot and Intelligent Equipment Engineering Technology Center of Xinjiang Uygur Autonomous Region,Xinjiang University,Urumqi 830047,China;Xinjiang Uygur Autonomous Region Agricultural and Animal Husbandry Mechanization Technology Promotion Terminal,Urumqi 830054,China)

机构地区：[1]新疆大学机械工程学院,新疆乌鲁木齐830047 [2]新疆大学新疆维吾尔自治区农牧机器人及智能装备工程研究中心,新疆乌鲁木齐830047 [3]新疆维吾尔自治区农牧业机械化技术推广总站,新疆乌鲁木齐830054

出　　处：《华南农业大学学报》2022年第2期87-95,共9页Journal of South China Agricultural University

基　　金：国家自然科学基金(51765063);国家级大学生创新创业训练计划项目(201810755079S);新疆维吾尔自治区研究生科研创新项目(XJ2019G033)。

摘　　要：【目的】借助多光谱遥感影像和Logistic算法,实现对棉田虫害的田间监测。【方法】以患虫害棉花区域为研究对象,利用无人机获取棉田多光谱遥感影像,并对影像进行预处理;结合受虫害棉花光谱特征,利用虫害敏感波段反射率与植被指数构建Logistic回归模型,开展棉花虫害识别监测研究。【结果】由土壤调节植被指数(Soil adjusted vegetation index,SAVI)模型和归一化植被指数(Normalized vegetation index,NDVI)模型构建的棉蚜虫、棉红蜘蛛、棉铃虫识别模型为最优模型,其训练样本准确率达到93.7%,测试样本准确率达到90.5%,召回率为96.6%,F1值为93.5%,对棉蚜虫、棉红蜘蛛和棉铃虫的识别模型决定系数分别为0.942、0.851和0.663。【结论】该模型可满足棉田中棉蚜虫、棉红蜘蛛和棉铃虫3种虫害的发生区域识别,且可基本满足棉田精准植保作业相关要求。【Objective】The purpose of this article is to monitor cotton pest in field based on Logistic algorithms and multi-spectral remote sensing images.【Method】The cotton areas with insect pests were selected as the research object.The multi-spectral remote sensing images of cotton field were acquired by UAV,and then pre-processed.Based on the spectral characteristics of cotton pests,the Logistic regression model was constructed by the reflectivity of pest-sensitive band and vegetation index to identify and monitor cotton pests.【Result】The cotton aphid,cotton red spider mite,and cotton bollworm identification models constructed by the soil adjusted vegetation index(SAVI)model and the normalized vegetation index(NDVI)model were the optimal models,and their accuracy for training sample and test sample reached 93.7%and 90.5%respectively the recall rate and F1 value were 96.6%and 93.5%respectively and the determination coeffecients of recognition models for three types of pests were 0.942,0.851 and 0.663 respectively.【Conclusion】This model can identify the occurrence area of cotton aphid,cotton red spider mite and cotton bollworm,which can basically meet the requirements of precision plant protection operation in cotton field.

关 键 词：无人机 光谱特征 遥感影像 植被模型 Logistic回归模型算法 虫害监测 

分 类 号：S251[农业科学—农业机械化工程] TP751.1[农业科学—农业工程]