遥感信息
遙感信息
요감신식
2015年
2期
94-98
,共5页
遗传算法%支持向量机%加工番茄%早疫病%病虫害识别
遺傳算法%支持嚮量機%加工番茄%早疫病%病蟲害識彆
유전산법%지지향량궤%가공번가%조역병%병충해식별
genetic algorithm%support vector machine%processing tomato%early blight%blight recognition
为了快速监测加工番茄早疫病发病率和加工番茄的产量和质量,防止病虫害的扩大,该文基于高光谱遥感数据和田间早疫病调查数据,以新疆天山北坡典型加工番茄种植区为研究区,分析加工番茄早疫病的病叶光谱响应特征,寻找早疫病的敏感波段,再利用遗传算法优化支持向量机的惩罚参数 c 和核函数参数 g ,对不同病害严重度的病叶进行识别。结果表明:不同病害严重度加工番茄早疫病病叶的敏感波段为628nm ~643nm 和689nm~692nm;遗传优化算法得出支持向量机最佳惩罚参数 c 为0.129,核函数参数 g 为3.479;分别利用多项式核、径向基核函数、Sigmoid 核进行分类训练和测试,最佳分类模型为径向基核函数模型,训练准确率为84.615%,预测准确率为80.681%,高于默认参数 c 和 g 的支持向量机模型。说明通过遗传算法优化支持向量机的识别方法具有更高的精度,支持向量机为多波段协同识别病害严重度提供了新的思路。
為瞭快速鑑測加工番茄早疫病髮病率和加工番茄的產量和質量,防止病蟲害的擴大,該文基于高光譜遙感數據和田間早疫病調查數據,以新疆天山北坡典型加工番茄種植區為研究區,分析加工番茄早疫病的病葉光譜響應特徵,尋找早疫病的敏感波段,再利用遺傳算法優化支持嚮量機的懲罰參數 c 和覈函數參數 g ,對不同病害嚴重度的病葉進行識彆。結果錶明:不同病害嚴重度加工番茄早疫病病葉的敏感波段為628nm ~643nm 和689nm~692nm;遺傳優化算法得齣支持嚮量機最佳懲罰參數 c 為0.129,覈函數參數 g 為3.479;分彆利用多項式覈、徑嚮基覈函數、Sigmoid 覈進行分類訓練和測試,最佳分類模型為徑嚮基覈函數模型,訓練準確率為84.615%,預測準確率為80.681%,高于默認參數 c 和 g 的支持嚮量機模型。說明通過遺傳算法優化支持嚮量機的識彆方法具有更高的精度,支持嚮量機為多波段協同識彆病害嚴重度提供瞭新的思路。
위료쾌속감측가공번가조역병발병솔화가공번가적산량화질량,방지병충해적확대,해문기우고광보요감수거화전간조역병조사수거,이신강천산북파전형가공번가충식구위연구구,분석가공번가조역병적병협광보향응특정,심조조역병적민감파단,재이용유전산법우화지지향량궤적징벌삼수 c 화핵함수삼수 g ,대불동병해엄중도적병협진행식별。결과표명:불동병해엄중도가공번가조역병병협적민감파단위628nm ~643nm 화689nm~692nm;유전우화산법득출지지향량궤최가징벌삼수 c 위0.129,핵함수삼수 g 위3.479;분별이용다항식핵、경향기핵함수、Sigmoid 핵진행분류훈련화측시,최가분류모형위경향기핵함수모형,훈련준학솔위84.615%,예측준학솔위80.681%,고우묵인삼수 c 화 g 적지지향량궤모형。설명통과유전산법우화지지향량궤적식별방법구유경고적정도,지지향량궤위다파단협동식별병해엄중도제공료신적사로。
The yield and quality of processing tomato are seriously affected by early blight.Our study area is the main growing area of the north of Tianshan in Xinjiang.Based on the data of hyperspectral remote sensing and the data of survey in the field of early blight,we analyzed the spectral characterization in order to look for the sensitive wave bands and recognized the different disease severity with the genetic algorithm and support vector machine model.The result show that:①Sensitive bands of different disease severity levels of processing tomato early blight is 628nm~643nm and 689nm~692nm.②Using genetic algorithm optimize parameters of support vector machine,we get that the best penalty parameters is 0.129 and kernel function parameters is 3.479.③ We make classification training and testing by polynomial nuclear,radial basis function nuclear,and sigmoid nuclear,where the best classification model is the radial basis function nuclear of SVM.Training accuracy is 84.615%and testing accuracy is 80.681%.Those are higher than SVM with default parameters.So the method of support vector machine optimized by genetic algorithm has higher accuracy and support vector machine are offered a new idea of combined band to identify disease severity.