Polyps segmentation is an important problem and receives a lot of attention in the medical field. The strong development of Deep Learning in recent years has opened up many positive solutions for this issue. Notable among these is the UNet model, known as a famous model for medical image segmentation tasks. UNet3 + is one of the latest upgraded versions of UNet that focuses on taking advantage of full-scale skip connections, thereby improving accuracy as well as reducing the number of network parameters. Although numerous enhanced proposals for this model have been put forth in recent years, the majority of them have attempted to integrate the attention mechanism which is transformer based method, leaving a large room for improvement by concentrating on traditional convolutional neural network (CNN) mechanisms. In this paper, we propose to improve the UNet3 + model through a transfer learning-based approach. Specifically, the Encoder part of the model will be constructed based on appropriate intermediate layers of the pre-trained model, with the pre-trained weights of these layers remaining intact to facilitate the transfer learning process. Experimental results on the Kvasir-SEG dataset demonstrate that the proposed method not only helps to improve the algorithm's efficiency but also effectively reduces the number of model parameters.