【摘要】 针对工业内窥镜在缺陷检测时遇到较大、较长的缺陷无法一次性完成识别、测量的问题，提出一种基于改进SIFT算法的图像拼接方法。在建立尺度空间的过程中，使用灰度阈值的方法，缩小了检测范围，解决SIFT算法在使用过程中计算量过大的问题。使用YOLOv3算法完成对缺陷区域的识别，并通过像素尺寸与实际尺寸的关系，实现对拼接后缺陷真实长度的测量。借助工业内窥摄像头，开展了缺陷拼接、识别和测量实验。实验结果表明，利用改进的SIFT算法对缺陷图像的拼接识别准确率可达92.83%；角点检测速度提高了44.3%；拼接后的缺陷测量相对误差可达0.63%。所提的方法具有较高的识别准确率和测量精度，可广泛应用于工业内窥镜的缺陷测量中。更多还原

【Abstract】 Aiming at the problem that the large and long defects in the defect detection of industrial endoscope cannot be identified and measured at one time,an image mosaic method based on improved SIFT algorithm was proposed. In the process of establishing the scale space,the method of gray threshold is used to reduce the detection range and solve the problem of excessive calculation amount in the process of using SIFT algorithm.The YOLOv3 algorithm was used to identify the defect area,and the true length of the defect after stitching was measured by the relationship between the pixel size and the actual size. Experiments of defect stitching,identification and measurement were carried out with the help of industrial intramural camera. The experimental results show that the accuracy of defect image mosaic recognition by the improved SIFT algorithm can reach 92.83%. Corner detection speed increased by 44.3%. The relative error of defect measurement after splicing can reach 0.63%. The proposed method has high identification and measurement accuracy and can be widely used in defect measurement of industrial endoscope.更多还原