Weld surface defect detection is as the significant part during the weld quality assessment. The weld surface defect detection method based on deep learning can improve the identification efficiency greatly. At present, it is focused on achieving the high precision for the complex and tiny weld surface. Therefore, a feature-focused method by adaptive genetic segmentation is proposed for weld surface defect detecting with high precision in laser welding, which facilitates automated industrial quality assessment and reduces inspection turnaround time. The adaptive genetic segmentation method segments the weld images according to the image complexity. The features of weld surface defect are effectively enhanced by employing the adaptive genetic segmentation method. By utilizing these feature-enhanced weld images as training input, the convolutional neural network (CNN) is guided to focus exclusively on the crucial morphological details of weld surface defects, which enhances the focus on important features and improves the classification precision. The better information on complex and tiny weld surface defects can be captured by suppressing the redundant background noise. The results show that the excellent weld surface detection accuracy without compromising processing speed is achieved. The feature-focused method by adaptive genetic segmentation is of great importance for the quality control of laser welding.