基于通道加权FPN的熔池检测与特征提取方法

High Speed Molten Pool Image Detection Based on Weighted Fusion and Parameter Extraction

为了获取使用熔池特征参数对增材制造进行反馈控制的输入参数,提出了一种基于通道加权FPN的激光增材制造熔池语义分割算法和基于图像像素阈值的熔池方向、面积和宽度特征参数提取算法.语义分割算法主要包含轻量级的主干神经网络、通道加权特征FPN网络.实验结果表明,熔池图像的分割速度可达79.76张/s,mIoU和mAP分别可达90.53%和95.79%,且模型大小仅为90MB.与其他相同类型的深度学习模型相比,该算法在保证精度的同时,提高了检测速度,减少了模型参数量和大小.熔池图像特征参数提取算法则结合了相机拍摄的原始图像与分割完成的图像的像素阈值分布情况,能够准确分析并计算出熔池的方向、宽度与面积特征参数.

In order to obtain the input parameters for feedback control of additive manufacturing using molten pool feature parameters,we propose a laser additive manufacturing molten pool semantic segmentation algorithm based on channel-weighted FPN,and a molten pool orientation,area and width feature parameter extraction algorithm based on image pixel threshold.The semantic segmentation algorithm mainly consists of a lightweight backbone neural network and a channel-weighted feature FPN network.The experimental results show that the segmentation speed of the molten pool image can reach 79.76 frames/s,and the mIoU and mAP can reach 90.53%and 95.79%,respectively,and the model size is only 90MB.Compared with other similar types of deep learning models,this algorithm improves the detection speed and reduces the model parameter size and size while ensuring accuracy.The molten pool image feature parameter extraction algorithm combines the pixel threshold distribution of the original image captured by the camera and the segmented image,and can accurately analyze and calculate the orientation,width and area feature parameters of the molten pool.