Robust Detection of Cercosporo Leaf Spot of Sugar Beet in Real Field

Abstract

<abstract> <bold>Abstract.</bold> In this paper, we propose a novel and robust scheme by matching-based method of orientation code matching (OCM) for Cercospora leaf spot (CLS) detection in sugar beet under real-field environmental conditions. Field monitoring of plant disease is the foremost step which provides quantized evaluation of disease severity for fungicide spraying guidance. However, there are challenges of lacking of continuity, robustness, and precision in field detection due to complex real field environment and non-rigid plant object. Different from current disease detection algorithms, our proposed method combined with two frameworks of OCM and supporter vector machine (SVM), which achieves good robustness and continuity for single-leaf based CLS observation to resist environment and plant variations. In first framework, we employ the robust template matching method of OCM for tracking a single leaf from a holistic plant throughout a time sequence field plant images. In the second framework, a three-dimensional (3D) feature, which integrates pixel-based intensity of L*, a* components in L*a*b* color space and region-based orientation code entropy with density derived from orientation code histogram, is as input into SVM classifier for CLS classification from complex-sandy soil background. Furthermore, we implemented a real field experiment and test the proposed method with field beet images. Experiment results shows that our integrated frameworks perform well for continuous CLS detection and quantization under complex real-field conditions. Besides, the robust observation of foliar disease development will facilitate better analysis of disease mechanism detection and fungicide-spraying guidance.