图像分割技术是图像分析的重要步骤。本文针对野外采集的植物图片存在多尺度及理想数据量不足等问题，提出一种多尺度Res-Att-Unet模型以实现自主植物图像分割。该方法在U-Net的收缩网络中增加多尺度机制，在输出端以相应的Label图进行深监督，并在模型的扩张网络部分加入双通道注意力机制，以抑制无关信息表达，各层新增的残差块使得模型可以提取更加丰富的特征信息，实验表明无论在花朵型图像还是叶片型图像中均取得不错的分割效果。

Image segmentation technology is an important step in image analysis. In this paper, a multi-scale Res-Att-Unet model is proposed to achieve autonomous plant image segmentation in order to solve the problems of insufficient multi-scale and ideal data of plant images collected in the field. Model in the contraction of U-Net network with multi-scale mechanism, increase the corresponding figure for deep Label supervision, and in the expansion of the network model part to join the dual channel attention mechanism, inhibition of irrelevant information expression, in addition the new residual blocks in each layer make the model can extract more abundant image information, the experiments show that good segmentation results are achieved in both flower-shaped images and leaf-shaped images.