Lightweight mesh target detection based on improved YOLOv3 and transfer learning

Automatic 3D Breast Ultrasound (ABUS) overcomes the shortcomings of traditional ultrasound and is successfully applied to the examination of lightweight mesh for abdominal hernia. However, manual inspection of ABUS ultrasound images is time-consuming and prone to missed diagnosis. Therefore, this paper proposes a target detection algorithm based on improved YOLOv3 and transfer learning to assist doctors to improve review speed and accuracy. Based on the original YOLOv3 model, a Spatial Pyramid Pooling (SPP) module is added in front of the detection layer to realize the fusion of local features and global features, which enriches the expression ability of the feature graph and solves the problem that it is difficult to detect small targets in the detection image. In the network training, the transfer learning strategy is used to train the network to overcome the problem of limited light mesh image data set and improve the robustness of the network to reduce over-fitting. The experimental results show that the mean Average Precision (mAP) of YOLOv3-SPP algorithm combined with transfer learning training method is 90.15%, and the image detection speed is 33.2 f·s⁻¹, which can effectively assist doctors to improve review efficiency.