小麦是全球非常重要的粮食作物之一。解决粮食安全问题最为关键的 是提高粮食的产量。世界上由于病虫害导致的小麦减产达到 7.6 亿吨[1]。 减少病虫害的影响，是保证粮食安全的重要途径。截至 2020 年，河南的 耕地面积为 1.22 亿万亩，约占 47.5%的全省面积，其中小麦的种植面积达 到 5690.69 千公顷。河南小麦的病虫发生 1.5 亿亩左右[2]。遥感是获取大 范围农作物病虫害发生范围和程度的有效技术手段。

本文以河南新乡市、信阳市、周口市、安阳市等地为研究区，研究基 于中等空间分辨率光学遥感数据的小麦病虫害探测方法。本文采用 Sentinel-2 遥感数据，分别用红边一阶微分曲线不对称性（Ar）、支持向 量机算法（SVM）和概率神经网络算法（PNN）探测研究区小麦病虫害 的发生范围，并对三个算法的检测精度进行对比分析。直接将红边一阶微 分曲线不对称性（Ar）与病虫害指数（DI）进行相关性分析，所得结果较 差，P 值为 20.7959529%，无法通过显著性检验。在取了对数之后，再次 进行显著性分析，P 值为 9.0686764%,勉强通过显著性检验，不足以完全 识别小麦的病害情况，但是，红边一阶微分曲线不对称性（Ar）与病虫害 指数（DI）有一定程度的相关性，可以协助提高小麦病虫害的识别能力。 支持向量机算法（SVM）在识别小麦病害的过程中，普通支持向量机算 法分类结果总体准确性为 68.1818%，在 Ar 值协助下，SVM 分类的总体 准确性为 77.2727% 。其中，可以很好地识别重度小麦的病害情况，可以 达到 100%的程度。但是，在轻度和中度小麦的识别中，效果不佳。小麦 的病害后期可以方便监测，但是，在小麦病害初期识别还是存在困难。

本文选择概率神经网络算法（PNN）来解决小麦病害初期的监测。无 论是普通 SVM 算法还是 Ar 值协助的 SVM 算法都无法准确识别小麦轻度 和中度的小麦病害，PNN 算法在数据充足的情况下，有较好的分类结 果。本文选择了多个月份的遥感数据，最高准确性达到了 90%。

Wheat is a very important staple food globally and plays an important role in global food security. The most important thing to solve food security is to increase the self-sufficiency rate of grain, and increasing the self-sufficiency rate of food will also face many contradictions, and the area of grain planting is limited. As of 2020, Henan's arable land area is 122 million mu, accounting for about 47.5% of the province's area, which has seriously affected Henan's economic development, social and cultural development, and the development of higher education. Wheat yields worldwide have also declined due to pests and diseases. In the face of these problems and contradictions, it is urgent to ensure food security by reducing the impact of pests and diseases.After wheat is affected by diseases, it will produce obvious changes in certain bands, and the disease situation of wheat can be judged according to the changes in the remote sensing band of wheat.In this paper, three methods are selected to study the identification of Henan wheat, namely red-edged first-order differential curve asymmetry (Ar), Support Vector Machine algorithm (SVM) and Probabilistic Neural Network algorithm (PNN).The correlation between the asymmetry of the firstorder differential curve of the red edge (Ar) and the pest index (DI) was directly analyzed, and the results obtained were poor, and the P value was 20.7959529%, which could not pass the significance test. After taking the logarithm, the significance analysis was carried out again, and the P value was 9.0686764%, which barely passed the significance test and was not enough to fully identify the disease of wheat, but the asymmetry (Ar) of the first-order differential curve of the red edge had a certain degree of correlation with the pest index (DI), which could help improve the identification ability of wheat diseases and pests.

In the process of identifying wheat diseases, the overall accuracy of the general support vector machine algorithm classification results was 68.1818%, In the process of identifying wheat diseases, the overall accuracy of the general support vector machine algorithm classification results was 68.1818%, and the overall accuracy of SVM classification with the assistance of Ar value was 77.2727%. Among them, the disease situation of severe wheat can be well identified,and it can reach 100%. However, in the identification of mild and moderate wheat, the effect is not good. The disease of wheat can be easily monitored in the later stage, but it is still difficult to identify the disease in the early stage of wheat.

In this paper, the probabilistic neural network algorithm (PNN) was selected to solve the initial monitoring of wheat diseases. Neither the ordinary SVM algorithm nor the Ar-assisted SVM algorithm can accurately identify mild and moderate wheat diseases, and the PNN algorithm has good classification results when the data are sufficient. This paper selects multiple months of remote sensing data, and the highest accuracy reaches 90%.