【摘要】 钛合金由于导热率低，在磨削过程中工件表面容易产生烧伤或裂纹。采用图像法进行在线表面烧伤识别时，受到磨削液等现场因素的影响，采集的工件图像存在运动模糊或者目标区域被遮挡等现象，影响深度学习模型的识别效果。针对现场图像受损的问题，采用对偶学习和跳跃连接的方法，设计生成对抗网络的生成器、判别器和损失函数，对细节信息进行修复，重建退化图像。试验结果表明，经过重建的钛合金磨削现场图像的峰值信噪比（peak signal to noise ratio，简称PSNR）平均值达到25以上，结构相似度（structural similarity，简称SSIM）平均值达到0.77以上。采用基于模型微调的方法对重建后图像进行烧伤识别，准确率达到90%以上。更多还原

【Abstract】 Due to the low thermal conductivity of titanium alloys, burns or cracks are likely to occur on the surface of the workpiece during the grinding process. When the image method is used for online surface burn recognition, due to the influence of on-site factors such as grinding fluid, the collected workpiece image has motion blur or the target area is occluded, which affects the recognition accuracy of the deep learning model. To address the issue of image damage in the scene, the method of dual learning and skip connections is used to design the generator, discriminator, and loss function of the generative adversarial network to repair the detailed information and reconstruct the degraded image. The experimental results show that the peak signal to noise ratio(PSNR) average value of the reconstructed titanium alloy grinding field image is above 25, and the average structural similarity(SSIM) value is above 0.77. The burn recognition method based on model fine-tuning is used to recognize the reconstructed image, and the recognition rate is more than 90%.更多还原