激光惯性约束聚变实验需要使用数十台套诊断设备从不同方位对瞬态微尺度物理过程进行诊断表征。大部分诊断设备通常需要进入巨型靶室真空环境内，在厘米到米级的不同工作距离上，对聚变靶上面毫米到数十微米的靶标进行瞄准，大部分诊断设备的瞄准精度需要达到50 μm水平。双目瞄准方法是在真空环境下实现远距离高精度瞄准的一种重要方法，但目前主要依赖人工判读图像识别靶标和手动操作诊断搭载平台运动实现对靶瞄准，特别是靶室内照明条件或诊断设备瞄准视线存在夹角等条件会严重影响靶标识别效果，对诊断设备瞄准精度造成较大影响。发展了一种基于机器视觉的诊断自动瞄准方法，采用Mask R-CNN算法并以大量模拟瞄准图进行靶标识别训练，有效解决了靶标自动判读问题，对靶标识别误差控制在8个像素点以内；同时基于实验室瞄准测试平台开展了靶标像素偏差与瞄准坐标偏离关系的离线标定，开展了算法引导下的瞄准精度测试，根据测试结果预估指向瞄准精度优于30 μm、径向瞄准精度优于50 μm，对实现诊断设备的高精度自动瞄准有一定的基础参考价值。

In the laser driven inertial confinement fusion experiments, several dozens of diagnostic instruments are needed, located in different sightlines of view. Most of the instruments are required to work inside the giant vacuum target chamber, in distances of centimeters to meters. They are always moved from several meters away, by the general diagnostic instrument manipulator and point the micro target in a precision of about 50 μm. The binocular pointing method is one of the continual alignment methods, which are able to work in the vacuum and in a faraway distance. However, at present this method needs target recognizing by eyes and pointing in a manual operation. Under some situations, such as under low illumination or pointing sightlines with an angle, the target marker may not be accurately recognized, as a result, the precision of the diagnostic alignment degenerates rapidly. In this work, a machine vision aided autonomic alignment method is proposed. The Mask R-CNN algorithm is used to recognize the target marker. Many simulated visual target graphs are created to train the algorithm. The accuracy of the target marker recognizing increases obviously. The error in the test is less than 8 pixels. Furthermore, the relation between the pixel shift in the visual graph and the pointing shift in the coordinate is calibrated in the laboratory. According to the machine vision aided alignment method, the tested alignment precision in pointing direction is estimated to be less than 30 μm, and in axial direction less than 50 μm. With the feedback of the alignment shift, the diagnostic alignment is able to be in an autonomic operation.