结合图像和图形特征的原木轮廓识别

Outline extraction of logs cross section base-upon both image and graphics features

针对原木端面存在污渍、裂痕和伐痕等导致原木轮廓难以被鲁棒地识别的问题,提出了结合图像和图形特征的原木轮廓识别方法.首先,引入YOLO V3,在对200多张原木图像样本进行学习后,实现利用图像信息进行目标检测;其次,利用原木轮廓图形上近似于圆的特点,采用轮廓重叠度计算和随机Hough变换校正YOLO V3识别结果的中心不准的问题.与同类原木图像的识别结果进行比较,结果表明,它能100%识别其中的完整原木,边缘的平均重叠度提高了10个点;在各类复杂端面轮廓情况下的识别结果中,该方法都有较好的鲁棒性,识别率达到98.8%.

Recognizing the cross sectional profiles of log is low in robustness due to disturbances from the stains, cracks and scratches on the cross section of logs. To solve this problem, we proposed an outline extraction method to distinguish the profiles based on both image and graphics features. At first, YOLO V3 is introduced to detect the target using the image information previously analyzed upon more than 200 log images. Then, the misalignment of the center from YOLO recognition is calibrated using contour overlap degree calculation and random Hough transform on the premise that the outline of the log contour is approximate to circle. Compared with the log recognition results from same type of study, 100% of the logs were well recognized using the proposed method, with a 10 points higher average overlapping degree of the edges. In all kinds of complex log contours, the recognition results showed good robustness, reaching a 98.8% recognition rate.