Research on Part Edge Detection Algorithm Based on Deep Learning

When the traditional edge detection algorithm is used for edge detection of parts, it will be affected by its surface reflective properties, machining and grinding tool marks, and uneven light intensity, resulting in noise points, missing edge points and reduced edge detection accuracy. This paper proposes an improved deeplabv3+ network and a part edge detection algorithm that combines the most peripheral constraint algorithm. First, the improved deeplabv3+ network is used to obtain segmented images with obvious boundaries, and then the most peripheral constraint algorithm is used for edge detection. The experimental results show that the algorithm proposed in this paper can accurately extract the complete and single-pixel edge contour of the part under the interference of factors such as uneven illumination intensity, and there are no noise points and burrs. the RMSE reaches 13.36, the PSNR reaches 25.60, and the SSIM reaches 0.9852, which meets the requirements of part edge detection accuracy.