论文部分内容阅读
病害胁迫是造成小麦减产及危及世界粮食安全的主要因素之一。如何准确区分相似病害并科学诊断病害严重度,成为国内外研究热点。文中针对中国冬小麦种植区常见的两种真菌疾病——白粉病和条锈病,采用高光谱成像系统获取两种病害侵染的小麦叶片图谱合一数据,通过主成分分析法对影像数据进行降维、密度分割法对病害面积进行分割后,得到识别病斑准确率达到97%;进一步分析侵染白粉病和条锈病的叶片病斑区域的光谱特征差异,选择第二主成分图像筛选两种病害的敏感波段,得到识别白粉病的敏感波段为519、643、696、764、795、813 nm,条锈病的敏感波段为494、630、637、698、755、805 nm。最后对筛选出的敏感波段建立白粉病和条锈病支持向量机(SVM)判别模型并验证,得到两种病害的区分精度为92%。综上,利用高光谱图像协同解析可在叶片尺度实现小麦白粉病和条锈病的有效判别,这为开发病害区分仪器提供了重要的理论基础。
Disease stress is one of the main factors that cause wheat to cut down and endanger the world's food security. How to accurately distinguish similar diseases and diagnose the severity of the disease scientifically has become a hot research field at home and abroad. In this paper, two common fungal diseases of winter wheat in China - powdery mildew and stripe rust were surveyed. Hyperspectral imaging system was used to obtain the combined data of wheat leaf pattern inoculated with the two diseases. The principal component analysis was used to reduce the dimension of the image data , The density segmentation method was used to segment the lesion area to get the recognition lesion accuracy rate of 97%; further to analyze the differences of the spectral characteristics of the diseased leaf spot region of powdery mildew and stripe rust, and select the second principal component image to screen the two diseases Sensitive bands were identified, and the sensitive bands to identify powdery mildew were 519,643,696,764,795,813 nm. The sensitive bands of stripe rust were 494,630,637,698,755,805 nm. Finally, discriminant models of powdery mildew and stripe rust support vector machine (SVM) were established and validated in the selected sensitive bands, and the discrimination accuracy of the two diseases was 92%. In conclusion, the effective discrimination between wheat powdery mildew and stripe rust can be achieved at the blade scale by using the cooperative analysis of hyperspectral images, which provides an important theoretical basis for the development of disease differentiation instruments.