Since the curvature induced variations of mode interference in multimode fiber (MMF) can be well represented by the fiber specklegrams, a bending recognition scheme based on the analysis of MMF specklegrams is proposed and verified. Amounts of specklegrams from the facet of MMF under different bendings were detected and used for training and testing in an image recognition algorithm based on deep learning. Good recognition results are provided by the specklegrams from MMF with diameter being 105 and 200 $\mu \mathrm{m}$, for which the average accuracies of bending status recognition are respectively 92.8% and 96.6%. Because specklegram can represent the status of the whole section of MMF, such scheme indicates the capability to distinguish the specklegrams even when the MMF is under complicated bending. In this sense, the scheme presents the potential of a single MMF being used as a bending indicator or status monitor independently, which may find applications in distinguishing the status of certain structures, such as robotic arms, mechanical fingers and some disabled auxiliary equipments.

由于多模光纤（MMF）中模干涉的曲率引起的变化可以用光纤散斑图很好地表示，因此提出并验证了基于MMF散斑图分析的弯曲识别方案。在基于深度学习的图像识别算法中，检测了来自不同弯曲程度的MMF刻面的散斑图，并将其用于训练和测试。直径为105和200的MMF的散点图可提供良好的识别结果$\mu \mathrm{米}$，其弯曲状态识别的平均准确度分别为92.8％和96.6％。由于散斑图可以表示MMF整个部分的状态，因此即使MMF处于复杂的弯曲状态，该方案也可以区分散斑图。从这个意义上讲，该方案提出了将单个MMF单独用作弯曲指示器或状态监视器的潜力，这可能会在区分某些结构（例如机械臂，机械手指和某些禁用的辅助设备）的状态中找到应用。