农机化研究
農機化研究
농궤화연구
Journal of Agricultural Mechanization Research
2016年
6期
205-209
,共5页
徐明珠%李梅%白志鹏%胡耀华%何勇
徐明珠%李梅%白誌鵬%鬍耀華%何勇
서명주%리매%백지붕%호요화%하용
马铃薯%早疫病%高光谱成像技术%特征波长%识别模型
馬鈴藷%早疫病%高光譜成像技術%特徵波長%識彆模型
마령서%조역병%고광보성상기술%특정파장%식별모형
potatoes%early blight%hyperspectral imaging technique%effective wavelengths%identification of models
为实现马铃薯叶片早疫病的快速识别,达到尽早防治的目的,利用高光谱成像系统连续4 天采集375 ~1 0 1 8 nm 波段内的健康和染病马铃薯叶片的高光谱数据信息,并用ENVI 软件提取感兴趣区域的光谱反射率平均值. 分别建立基于全光谱( full spectrum , FS )、连续投影算法( SPA )和载荷系数法( x-LW )提取的特征波长的 BP网络和LS-SVM识别模型,其中FS-BP、SPA-BP、x-LW-BP 模型中预测集识别率分别为1 0 0%、1 0 0%、9 8 .3 3%, LS-SVM 模型的预测集识别率均为1 0 0%;SPA和x-LW提取的特征波长个数均仅占全波长的1 .4 7%,大大简化了模型,提高了运算速率. 实验表明:应用高光谱成像技术可以快速、准确地识别出马铃薯叶片早疫病,且 SPA和x-LW可以作为特征波长提取的有效方法,为田间马铃薯早疫病的在线实时检测仪器的开发提供理论依据.
為實現馬鈴藷葉片早疫病的快速識彆,達到儘早防治的目的,利用高光譜成像繫統連續4 天採集375 ~1 0 1 8 nm 波段內的健康和染病馬鈴藷葉片的高光譜數據信息,併用ENVI 軟件提取感興趣區域的光譜反射率平均值. 分彆建立基于全光譜( full spectrum , FS )、連續投影算法( SPA )和載荷繫數法( x-LW )提取的特徵波長的 BP網絡和LS-SVM識彆模型,其中FS-BP、SPA-BP、x-LW-BP 模型中預測集識彆率分彆為1 0 0%、1 0 0%、9 8 .3 3%, LS-SVM 模型的預測集識彆率均為1 0 0%;SPA和x-LW提取的特徵波長箇數均僅佔全波長的1 .4 7%,大大簡化瞭模型,提高瞭運算速率. 實驗錶明:應用高光譜成像技術可以快速、準確地識彆齣馬鈴藷葉片早疫病,且 SPA和x-LW可以作為特徵波長提取的有效方法,為田間馬鈴藷早疫病的在線實時檢測儀器的開髮提供理論依據.
위실현마령서협편조역병적쾌속식별,체도진조방치적목적,이용고광보성상계통련속4 천채집375 ~1 0 1 8 nm 파단내적건강화염병마령서협편적고광보수거신식,병용ENVI 연건제취감흥취구역적광보반사솔평균치. 분별건립기우전광보( full spectrum , FS )、련속투영산법( SPA )화재하계수법( x-LW )제취적특정파장적 BP망락화LS-SVM식별모형,기중FS-BP、SPA-BP、x-LW-BP 모형중예측집식별솔분별위1 0 0%、1 0 0%、9 8 .3 3%, LS-SVM 모형적예측집식별솔균위1 0 0%;SPA화x-LW제취적특정파장개수균부점전파장적1 .4 7%,대대간화료모형,제고료운산속솔. 실험표명:응용고광보성상기술가이쾌속、준학지식별출마령서협편조역병,차 SPA화x-LW가이작위특정파장제취적유효방법,위전간마령서조역병적재선실시검측의기적개발제공이론의거.
The purpose of this paper is to realize identification of early blight of potato leaves rapidly , and to achieve ear-ly prevention .Hyperspectral data of healthy and infected potato leaves were obtained by hyperspectral imaging system within the wavelength range of 375~1018 nm for four consecutive days , and the average spectral reflectance of the region of interests were extracted by ENVI software .Effective wavelengths were selected by successive projections algorithm (SPA) and x-loading weights (x-LW), respectively.Error back propagation (BP) neural network and least squares support vector machines ( LS-SVM) identification models were established based on full spectrum ( FS) , SPA, and x-LW, respectively.The results showed that the identification rates of the prediction set are 100%, 100%and 98.33%in FS-BP, SPA-BP, x-LW-BP models and 100%in all of LS-SVM models, respectively.The number of effective wave-length extracted by SPA and x-LW, respectively, accounts for only 1.47%of the total number of wavelengths , simplify-ing the models and improving the rate of operation greatly .The results indicated that it is feasible to identify early blight on potato leaves exactly and quickly using hyperspectral imaging , SPA and x-LW are effective methods to extract charac-teristic wavelengths and it provides a theoretical basis for the development of online real-time detection instrument of ear-ly blight of potato in the field .