基于单目视觉的谷物联合收获机产量测量方法

为准确获取农田中作物产量信息，以联合收获机刮板式升运器为研究对象，提出了一种基于单目视觉的联合收获机产量测量方法。首先，根据真实的升运器内部谷堆图像，提出了一种更加精确的刮板上谷物堆积模型。然后，基于视觉测量和图像处理技术，开发了一种谷堆体积测量方法。在辅助光源照射下，通过工业相机采集升运器内刮板和谷堆的侧面图像。采用邻域微分法提取图像感兴趣区域，再利用Otsu法和形态学处理方法从背景中准确分割出谷堆。根据相机成像模型，计算谷堆在世界坐标系中的实际侧面积，并通过谷堆几何模型得到谷物的体积。最后，将每个刮板上的谷堆体积累加求取产量。为验证所提方法的有效性，搭建了基于单目视觉的谷物测产系统，并在升运器试验台上开展了试验验证。试验结果表明，在不同的升运器转速工况下，所提方法测量的相对误差为-4.08%~3.41%，能够满足联合收获机产量测量精度要求。

Yield Monitoring for Grain Combine Harvester Based on Monocular Vision

It is of great significance to accurately obtain the crop yield distribution information of farmland, which can provide decision-making basis for fine farmland management. However, due to the serious grain falling in the harvester elevator, the traditional photoelectric sensor is prone to triggering by mistake, and the grain falling is random, so the error is difficult to correct and eliminate. In order to improve the accuracy of yield monitoring, a method of grain yield measurement based on monocular vision was developed, which can be used in the combine harvester with scraper elevator. Firstly, a more accurate geometric model of grain heap on the scraper was established according to the real images of grain pile in elevator. Then, a volume measurement method of the grain heap was developed based on vision measurement and image processing technology. Under the illumination of the auxiliary light source, the image of the scraper and grain heap in grain elevator was collected by an industrial camera. The neighborhood differentiation-based method was put forward to extract the region of interest of the image, and then the Otsu method and morphological processing were used to accurately segment the grain piles from the background. According to the camera imaging model, the actual side area of the grain pile in the world coordinate system was calculated, and the grain volume was obtained through the geometric model of the grain pile. Finally, the volume of grain pile on each scraper was accumulated to obtain the grain yield. To verify the effectiveness of the proposed method, a grain yield measurement system based on monocular vision was built, and experiments were carried out on the elevator experiment bench. The results showed that the relative error measured by the proposed method was between -4.08% and 3.41% at different elevator speeds, which can meet the accuracy requirements of grain yield monitoring for the combine harvester.