Magnetic pulse crimping technology has high forming speed and efficiency

and is suitable for the connection of high strength steel and aluminum

carbon fiber or other lightweight materials. It has wide application prospect in aircraft industry. However

there are few online detection methods for magnetic pulse crimping pipe

which is not conducive to realize automated production of the technology. A visual detection method based on improved YOLOv4–Tiny (You only look once v4–Tiny) detection network and adaptive image processing was proposed to meet the requirement of online detection of crimping quality of magnetic pulse pressure tubing. Efficient channel attention (ECA) module was introduced to improve the YOLOv4–Tiny detection network

and an adaptive crimping depth extraction algorithm was designed based on adaptive threshold segmentation algorithm and Canny edge detection algorithm. A batch of magnetic pulse crimping pipes images were collected in a simulated industrial environment and were divided into training set and verification set. Finally

the algorithm was trained with the training data set

and the detection model obtained by training was verified by the verification set. The average precision (AP@0.5) of the crimping area detection model is 100% when the intersection ratio threshold is 0.5

and the average precision (AP@0.5:0.8) is 93.14% when the intersection ratio threshold is 0.5

0.6

0.7 and 0.8

and the running time per frame is 1.66ms. For image processing edge extraction algorithm

verification results show that the average deviation is 0.85 pixels

the maximum deviation is 2.6 pixels

and the running time of a single frame is 3.49ms. The average deviation of the whole crimping depth detection algorithm is 0.313 pixels

the mean square error is 0.115 square pixels

the deviation ratio is 1.35%

and the running time of a single frame is 124.49ms. In conclusion

the proposed algorithm can accurately and quickly extract the crimping depth of magnetic pulse crimping pipe without additional positioning. The algorithm has low deployment cost

high robustness and high application value.