Abstract: Aiming at the problems of welding seam image transmission，storage，and detection，this research designed a remote welding seam image detection system based on YOLOv5. The hardware system consisted of the camera module， power step-down module，main control unit module，and radio frequency module. The software system was developed by the WinForms application program，and the original welding seam image and marked image were displayed on the monitoring end in the form of a visual interface. In this study，an attention mechanism was added to the YOLOv5 network model to enhance the ability of weld seam feature extraction. A small target detection layer was added to the Neck part of the YOLOv5 model to enhance the generalization ability of the model. The YOLOv5 convolution neural network was trained with 870 images and tested with 130 images. The experimental results showed that the mAP value of the improved model was finally stable at 93. 42%，0. 53% higher than that of the original model.

Key words: object detection, YOLOv5, deep learning, attention mechanism, WinForms