Abstract:

Above-ground biomass is an important parameter for studying rice growth. The traditional measuring method is labor-intensive and time-consuming. This paper proposed a modified method based on biomass estimating modle by image analysis, and compared its performance at tillering stage and elongation stage. This proposed model using the average of multiple side views projected area and top view projected area had a better prediction performance than the models using single side view projected area. With the determination coefficient (R2) of 0.93 and mean relative error (MAPE) of 9.26%, the new model was better than the others. The result also showed that the biomass predictive effects depended on the length of rice plant growth stage.

Key words: biomass, non-destructive, growth-stage, multi-view, image processing, regression analysis

摘要：

地上部分生物量是水稻生长研究中的重要参数，传统的测量方法主要依靠人工剪取称重，不仅具有破坏性，而且费时费力。提出了一种改进的基于图像特征参数的生物量预测模型，并比较了其在分蘖期和拔节期的效果。优于使用单幅图像投影面积预测的方法，该模型使用多幅侧视图像投影平均值和顶视图像来降低植株不对称带来的影响。基于分蘖期和拔节期的两批数据，新模型的预测平均相对误差(MAPE)达到9.26%，决定系数(R2)为0.93，优于其他模型。实验结果还显示出，当水稻生育期跨度较大时会造成植株结构差别较大，进而影响生物量的预测效果。

关键词: 生物量；无损；生育期；多视角；图像处理；回归分析