基于混合像元分解的分蘖期水稻基本苗数量估测方法研究

Estimation of Rice Basic Seedling Number Based on Mixed Pixel Decomposition

摘要:

基本苗数量是反映水稻健康水平的重要依据，在分蘖期精准估测水稻基本苗数量可以指导后期的施肥量，从而调控水稻的最佳分蘖数。同时，对水稻长势监测和产量预测具有非常重要的意义。针对传统田间人工统计基本苗数量耗时长、成本高等问题，以江苏大学附属农场镇江润果农场分蘖期水稻为研究对象，利用大疆无人机（M600 Pro型）搭载多光谱相机（Rededge-MX型）获取水稻分蘖期多光谱数据，对原始图像进行图像拼接、辐射校正、几何校正等预处理操作，根据像元纯度系数提取土壤端元和植被端元，建立波谱库，然后按照完全约束最小二乘法的方法执行混合像元分解，构建植被覆盖度和水稻基本苗数量的回归模型。该研究方法获得的模型决定系数R2为0.891，均方根误差RMSE为4.6株/m2。而传统的像元二分法模型（基于NDVI、VDVI和GNDVI植被指数计算植被覆盖度），其决定系数R2为0.834、0.744、0.642，其RMSE为5.7、7.1、8.4株/m2。试验结果表明，基于完全约束最小二乘法的混合像元分解模型评价指标均优于像元二分法模型。本文基于混合像元分解方法有效提高了水稻基本苗统计精度，并且生成了水稻基本苗数量反演图，可以直观统计基本苗数量，为分蘖期水稻补苗、间苗提供指导。

关键词:水稻;基本苗;混合像元分解;完全约束最小二乘法;像元二分法

Abstract:

The basic seedling number is an important basis to reflect the health level of rice. Accurately estimating the basic seedling number at tillering stage can guide the fertilizer and nitrogen amount in later stage, so as to regulate the optimal tillering number of rice. At the same time, it is of great significance for rice growth monitoring and yield forecasting. Considering that traditional manual field statistics on the number of basic seedlings are time-consuming and costly, this experiment took rice at tillering stage in Zhenjiang Runguo Farm, affiliated farm of Jiangsu University, as the research object, and used DJI UAV (M600 Pro) equipped with multi-spectral camera (Rededge-MX) to obtain multi-spectral data of rice at tillering stage. After image splicing, radiometric correction, geometric correction and other pretreatment operations were carried out on the original image, the soil end elements and vegetation end elements were extracted according to the pixel purity coefficient, and the spectral library was established. Then the mixed pixel decomposition was performed according to the fully constrained least square method, and the regression model of vegetation coverage and the number of basic rice seedlings was constructed. The model determination coefficient obtained by this method was 0.891, and the root mean square error RMSE was 4.6 plants/m2. In the traditional pixel dichotomy model (based on NDVI, VDVI and GNDVI vegetation index), the determination coefficients of R2 were 0.834, 0.744 and 0.642, and the RMSE were 5.7 plants/m2, 7.1 plants/m2 and 8.4 plants/m2. The experimental results showed that the evaluation indexes of the model based on the hybrid pixel decomposition method were superior to the pixel dichotomy model. The statistical accuracy of rice basic seedlings can be effectively improved based on the decomposition of mixed pixel decomposition, and the inverse map of rice basic seedling number is generated, which can directly count the basic seedling number and provide guidance for rice seedling replacement and thinning at tillering stage.

Key words:rice;basic seedlings;decomposition of mixed pixels;completely constrained least squares;pixel dichotomy