The process of welding is a crucial industrial production method that establishes a permanent connection between components and is widely utilized across various sectors. However, welding operations are susceptible to factors such as setting adjustments, environmental conditions, and operator compliance with standards, leading to the occurrence of defects such as cracks, incomplete fusion, and pores which significantly compromise production safety. The conventional manual inspection method lacks efficiency and standardization. Advancements in machine vision research provide effective solutions to these challenges. Currently, advancements in ultrasonic technology and C-scan imaging indicate unique advantages for detecting welding defects through ultrasonic imaging. This paper utilizes the YOLOv7 network model as the foundational framework to establish a machine vision-based welding defect feature detection model in C-scan images. By adjusting the model and incorporating an SE attention mechanism module, the localization and identification of welding defects have been enhanced with an accuracy exceeding 90% and a recall value reaching approximately 0.03. This indicates that our model holds practical reference value for welding defect detection.