光学技术, 2017, 43 (1): 50, 网络出版: 2017-02-23
基于SoC软硬件协同设计的布匹瑕疵检测
Fabric defect detection of hardware-software co-design based on SoC
光学测量 布匹瑕疵 显著性分析 椭圆Gabor滤波 软硬件协同设计 optical measurement fabric defect salient analysis elliptical Gabor filter hardware-software co-design
摘要
针对布匹瑕疵检测存在的有效性不高、实时性不强等问题, 提出一种视觉显著性分析和最优椭圆Gabor滤波相结合的瑕疵检测方法。离线阶段采用差分进化算法对无瑕疵样本图像进行椭圆Gabor滤波器的参数寻优, 得到最优EGF; 在线阶段采用软硬件协同设计技术进行软硬件划分, 基于Zynq SoC的PL端实现中值滤波、RGB到CIE LAB格式颜色空间转换、LAB三通道EGF处理, 通过AXI VDMA总线将图像数据传输到PS端, 并计算滤波前后图像特征向量, 提取瑕疵显著区域, 确定阈值分离瑕疵。实验表明, 该方法有效地抑制了背景凸显瑕疵区域, 且软硬件结合提速明显, 具有较高的有效性和实时性。
Abstract
There are some shortages in present fabric defect detection algorithms, such as low accuracy, low real-time capability. A detection algorithm combined on visual salient analysis and optimized elliptical Gabor filter is presented. Differential revolution method are adopted by training the free-defect sample images to gain the optimized parameters of elliptical Gabor filter during the offline period and gain the optimized EGF. In the online period, the hardware and software of algorithm are partitioned by using hw/sw co-design technology, to realize median filter, color space conversion of RGB to the CIE LAB format, three-channel EGF processing in the LAB space based on PL of Zynq SoC. The image data are transferred to PS via AXI VDMA bus. The feature vectors of the pre-filter and filtered images are calculated. The defect salient region is extracted. The threshold is determined and the defect region is segmented. Experiments show, the background effectively and highlight the defect region accurately are suppressed. Combining software and hardware is accelerated clearly. The validity and real-time capability are better.
黄张祥, 白瑞林, 吉峰. 基于SoC软硬件协同设计的布匹瑕疵检测[J]. 光学技术, 2017, 43(1): 50. HUANG Zhangxiang, BAI Ruilin, JI Feng. Fabric defect detection of hardware-software co-design based on SoC[J]. Optical Technique, 2017, 43(1): 50.