A Robust and Lightweight Detector for Ship Target with Complex Background in SAR Image

Algorithm 1

3S-YOLO detection algorithm.

Preliminary work : data set input, training parameter setting, training period epoch

Pruning rate : R

While(FPGM pruning model does not converge or T<epoch)

1.Backbone

   Feature information was extracted using Focus, C3 and CBL structures.

2.Neck

   Scale change using up-sampling and down-sampling, feature fusion using Concat.

   Computational fusion features :

3. Prediction layer

   Use Varifocal loss to highlight positive sample weights. EIoU anchor frame positioning.

   EIoU loss:

   Varifocal-EIoU:

End While

Network compression, redundant parameter elimination after pruning.

Output : pruning and model training convergence network model and weight