为了对现有机器人的物体识别进行优化和改进,提出了一种新的权重计算方法进行室内场景图像识别。 该方法通过对输入场景的转换获取无向带权图,在表面法方向的基础上,使用表面凹凸度这一指标来 进行表面粗糙度综合判定,取代了传统的布尔判定,大大提升了抗噪性能,避免出现错误传递放大的 情况。基于快速图像分割算法及时识别未知物体。实验结果表明:提出方法的鲁棒性及抗噪能力均较 强,优于单纯基于法方向的方法。与基于深度学习与推测的同类方法相比,所提方法性能更好,更适用于实际识别。

Damage experiments are conducted by irradiating 351 nm laser near-filed to fused silica based on the damage threshold definition of near-filed irradiating. Damage areas are extracted from damage images by the marker-based watershed algorithm with gray control. The initial damage threshold is defined as the fluence of critical site between damage region and no damage region, which is calculated by comparing the damage image with 351nm laser near-field. This damage testing method is sample, quick, which can effectively decrease the effect of beam diameter for damage threshold due to the bigger aperture. The algorithm of image segmentation effectively removes the noise and suppresses the over-segmentation, and improves the accuracy of initial damage and damage growth. The research shows that the damage of fused silica is induced by 351 nm laser near-field with pulse width of 3.13 ns. The initial damage threshold is 2.94 J/cm2. With multiple irradiation of fused silica, the damage growth of exit surface is exponential, and the coefficient of damage growth is 0.32.

概述了红外图像处理系统对图像预处理的要求，并针对传统图像分割算法中 容易出现的目标错分问题，提出了一种基于灰度梯度映射函数的Otsu法。该方法克服了传统的基于灰度直方图的Otsu法不能 利用图像局部信息和边界信息的缺点，通过将像素点的梯度信息与灰度信息相结合，较好地从背景中分离出了目标。 仿真结果表明，该算法达到了预期的技术指标，很好地实现了红外图像预处理的功能。

针对351 nm激光诱导熔石英样品进行损伤实验。采用激光近场辐照样品，并基于近场辐照的损伤阈值定义方法进行损伤测试。利用带有灰度抑制的分水岭标记分割算法对损伤图像进行损伤区域提取，与相应近场分布进行比较，得到损伤区域边界点对应的局域能量密度即为损伤阈值。这种损伤测试方法简单、快速，有效降低了损伤阈值口径效应，且采用了能够有效去除噪声和抑制图像过分割问题的算法对损伤图像进行处理，使得到的损伤阈值和损伤增长系数有较高的准确性。实验结果表明，脉冲宽度为3.13 ns的 351 nm激光诱导熔石英样品的损伤阈值为2.94 J/cm2；在多次辐照样品情况下，熔石英样品出射面的损伤尺寸随着辐照次数的增加呈指数增长，其损伤增长系数为0.32。