Application Of Image Feature On Detection Of Rice Plathopper

Rice planthoppers is one important species of migratory rice pests that bring greatlosses to the rice yield every year. It is a key for forecasting accurately and controllingreasonably rice planthoppers that we know the dynamic changes of the planthopperpopulation density. The artificial survey methods of rice planthoppers in paddy oftencause the physical and visual fatigue of surveyors, and the low efficiency andprecision. Yao et al used image processing techniques to automatically count riceplanthoppers on rice plants based on three layers of detections.In order to improve the detection rate and reduce false detection rate ofplanthoppers, we have studied the effects on the detection results of planthoppers bydifferent image sizes, sources and amount of the training samples and differentparameters of the AdaBoost classifier based on Haar features in the first layer ofdetection. The results show that the AdaBoost classifier trained by the2400positivesamples with18*24pixels and the3300negative samples with different sizes, and thelargest false alarm rate of the classifier is0.48. We obtained the87.9%detection rateand99.1%false detection rate of. planthopper.The false detection rate of planthoppers in the first layer is high. We developed aAdaboost classifier based on HOG features. The result shows that this classifiertrained by the2400positive samples with36*48pixels and the3300negative sampleswith different sizes is better than the classifier based on the Haar features.To reduce the false detection rate in the first layer, we develop the SVMclassifiers based on the HOG, LBP and Gabor features to detect and identify theplanthoppers. The results show that the SVM classifier trained by4400dimensionalGabor features had the best results and we obtained the93.1%detection rate and the4.4%false detection rate of planthoppers.