基于复杂性测度的柑橘病虫害识别

为进行柑橘病虫害的机器识别，提出了柑橘病虫害为害状特征的复杂性测度表达与病虫害识别方法。首先，对柑橘病虫害为害状主要色调区间[0,120°]进行长度为1°的等分割，形成120个色调子区间；其次，统计各色调子区间像素分布密度，以此作为柑橘病虫害为害状复杂性测度的结构性序列；再次，依据此结构性序列计算病虫害为害状统计复杂性测度，并将其作为病虫害特征值；最后，将Shannon信息熵和统计复杂性测度作为输入变量建立3层前馈神经网络柑橘病虫害识别模型来识别柑橘病虫害。柑橘蓟马、褐圆蚧、柑橘树脂病、糠片蚧各30个测试标本最低正确识别率、最高正确识别率、平均正确识别率分别为93.3%、96.7%、95%。试验结果表明，柑橘病虫害为害状的复杂性测度较充分地表达了柑橘病虫害的典型特征，能用此方法进行柑橘病虫害识别。

Machine Recognition of Diseases and Insect Pests on Citrus Fruits Using Complexity Measurement

To automatically recognize citrus diseases and insect pests, the ideas of complexity measurement expression and diseases and insect pests’ identification of citrus diseases and insect pests’ damage pattern features were researched. Firstly, citrus diseases and insect pests damage pattern main hue range [0,120°] was equally divided for the length of 1° to form 120° hue subinterval; secondly, pixel spread density of all hue subinterval was counted up to make as a structural sequence of citrus diseases and insect pests damage pattern complexity measurement; and then diseases and insect pests damage pattern statistical complexity measurement was calculated according to this structural sequence to make as diseases and insect pests damage character; finally a three-layer feedforward neural network citrus diseases and insect pests identification model was established to identify citrus pests and diseases in the situation that Shannon information entropy and statistical complexity measurement was regarded as input variables. Each 30 of Pezothrips Kellyanuses, Chrysomphalus aonidums, Phomopsis citri Pawcetts, and Parlatoria Pergandi Comstocks were tested as specimens, the minimum correct recognition rate, the highest recognition rate and the average correct recognition rates were 93.3%, 96.7% and 95%. The results shows that citrus diseases and insect pests damage pattern complexity measurement can fully express the typical characteristics of citrus pests and diseases, thus can be used as citrus diseases and insect pests identification.