Aiming at the problem that the detection accuracy of hidden objects on mechanical external damage is not high

which is prone to appear false detection and missed detection in the case of complex background

large scale change and occlusion

a hidden target detection algorithm for mechanical external damage based on improved YOLOv7(you only look once version 7) is proposed. Firstly

the Swin Transformer attention mechanism is added to the Head to improve the extraction ability of multi-scale features. Then

some convolutions in the Backbone are replaced by depth-wise separable convolution to reduce the operation cost of the model. While

the Focal-EIOU loss function is used to optimize the prediction box. Finally

the Mish activation function is introduced to enhance the generalization ability of the network

and improve the detection performance of the model when the complex background and the object are partially obscured. The experimental results show that the improved algorithm is 5.2%

10.6%

and 5.2% higher than the original YOLOv7 in accuracy

recall

and mean average precision

respectively. Compared with other mainstream algorithms

it has obvious advantages in detection accuracy and model volume. The effectiveness of the improved method is verified

which provides algorithm support for edge recognition of mechanical external damage hidden in complex backgrounds.