针对样本容量较小可能导致的过拟合和泛化能力弱的问题，提出新的特征融合网

Aiming at the problem of over-fitting and weak generalization ability that may be caused by the small sample size, a new feature fusion network Dense-YOLOv3 is proposed. By using two sets of dense network modules to replace part of the backbone's residual network modules, deep and shallow layers are realized. Feature interactive reuse to improve feature extraction capabilities.

In view of the problem of weak overfitting and generalization capability caused by the small sample size, a new feature fusion network, Denise-YOLOv3, is proposed, and the deep and shallow feature interaction can be reused by using two sets of dense network modules to replace the partial residual network modules of backbone, which can improve feature extraction capability.

针对样本容量较小可能导致的过拟合和泛化能力弱的问题，提出新的特征融合网络Dense-YOLOv3，通过使用两组dense network modules替代backbone的部分residual network modules，实现深层和浅层的特征交互复用，提高特征提取能力。<br>