In-situ vision data acquisition, feature extraction, and analysis are ongoing challenges for quality assessment in directed energy deposition (DED). This work proposes a method for detecting target regions in the laser-material interaction zone based on a low-cost high-dynamic-range (HDR) vision sensor. Adaptive image thresholding, connected component analysis, and iterative energy minimization are used to identify target regions. The method is designed to be adaptive, in terms of obtaining parameters based on simple training data, and robust, in terms of feature detection performance subject to under-melt, conduction and keyhole melting mode phenomena. The performance of the proposed region detection scheme is quantitatively and qualitatively evaluated against annotated data. It was found that the True Positive Rate in detection was above 90%, while the False Detection Rate was less than 10%. Extensive experimental results show that the proposed scheme is able to detect and follow target regions under a variety of power levels and process conditions.

现场视觉数据采集，特征提取和分析是定向能量沉积（DED）中质量评估的持续挑战。这项工作提出了一种基于低成本高动态范围（HDR）视觉传感器的激光材料相互作用区域中目标区域的检测方法。自适应图像阈值，连接成分分析和迭代能量最小化用于识别目标区域。就基于简单训练数据获得参数而言，该方法被设计为自适应的；对于易受熔解不足，传导和锁孔熔化模式现象影响的特征检测性能，该方法具有鲁棒性。所提出的区域检测方案的性能是针对带注释的数据进行定量和定性评估的。发现检测的真实阳性率超过90％，误检率小于10％。大量的实验结果表明，该方案能够在各种功率水平和工艺条件下检测并跟踪目标区域。