In recent years, target tracking technology has been widely applied in areas such as video surveillance and autonomous driving. However, in some practical scenarios, issues arise where the background is similar to the tracked object, leading to tracking failures. Therefore, this paper proposes a target tracking algorithm (SiamMF) that utilizes dilated convolution and adopts a multi-scale feature fusion strategy. Firstly, the algorithm enhances the feature extraction capability of the network by fusing the deep-layer feature maps output from the deep network with the shallow-layer feature maps output from the shallow network through upsampling. Finally, the feature maps obtained through dilated convolution are fused with the feature maps obtained through standard convolution, addressing the issue of traditional convolution causing changes in the size of the feature maps after feature extraction, which can affect subsequent calculations. The performance of the proposed algorithm is validated using the OTB-100 dataset. Experimental results demonstrate that the accuracy and success rate of the proposed Siam MF algorithm on the OTB-100 dataset reach 81.6% and 62.3% respectively, achieving a significant improvement compared to the traditional Siam FC algorithm.