【摘要】 针对Canny算法在焊缝缺陷检测中出现边缘信息丢失、检测精度不足以及实时性偏低等问题，提出一种基于FPGA的改进型Canny焊缝缺陷检测方法。首先，该方法将自适应中值滤波和双边滤波技术结合起来，有效抑制噪声的同时最大限度地将焊缝边缘细节保留下来；其次，为了提升边缘提取的完整性，利用高斯加权的Sobel算子计算图像梯度；最后，通过引入基于梯度的自适应多级阈值选择策略，有效连接焊缝边缘的同时提升检测的自适应性。本研究采用FPGA的高速并行处理技术与分层流水线架构，开发了一种在硬件加速平台下的实时边缘检测系统。实验结果表明，该方法在焊缝区域内的边缘连通性较“自适应中值滤波+Canny算法”提升了24%,边缘强度提升了22.7%,并在100 MHz时钟频率下处理一张640×480的图片仅耗时4.100 5 ms。本算法在检测精度及实时性方面更具优势，满足复杂环境下对焊缝缺陷检测的相关需求。更多还原

【Abstract】 To address the issues of edge-information loss, insufficient detection accuracy, and suboptimal real-time performance when applying the Canny algorithm to weld-defect detection, this study proposes an FPGA-based advanced Canny weld-defect detection method. Firstly, the method integrates adaptive median filtering with bilateral filtering to effectively suppress noise while maximally preserving weld-edge details. Secondly, image gradients are computed using a Gaussian-weighted Sobel operator to enhance edge extraction completeness. Finally, a gradient-based adaptive multi-level threshold selection strategy is introduced to promote robust weld-edge connectivity and improve detection adaptability. Leveraging the high-speed parallel processing capabilities and hierarchical pipelined architecture of FPGA, a real-time hardware-accelerated edge detection system has been implemented. Experimental results demonstrate that, compared to the “adaptive median filtering + Canny” baseline, the proposed method achieves a 24% increase in edge connectivity within weld regions, a 22.7% enhancement in edge amplitude, and processes a 640×480 image in just 4.100 5 ms at a 100 MHz clock frequency. The proposed algorithm exhibits superior detection accuracy and real-time performance, meeting the demands of weld-defect detection in complex environments.更多还原