Identifying drug-target interactions (DTIs) is a key task in drug discovery, and it plays an important role in the fields of virtual screening, drug repurposing and identification of potential drug side effects. Traditional biological experimental methods of identifying DTIs are often expensive and time-consuming. With the great success of deep learning in natural language processing and other fields, many studies have used deep learning to predict DTIs. However, many previous studies only utilized single-view features of drugs and proteins to predict DTIs, ignoring the potential value of multi-source descriptors. A drug-target interaction prediction framework with multi-source descriptors fusion (DFDTI) is proposed in this paper to make full use of the structural information of drugs and proteins from different view. First, low dimensional representations are generated for the multi-source descriptors through fully connected (FC) layers. Then, considering the different contribution of different types of descriptors, the channel attention mechanism is used to give weights to descriptors. In addition, one layer Transformer encoder is used to enhance the feature representation of descriptors. Finally, the enhanced features representations of drugs and proteins are concatenated and input into the deep neural network (DNN) to predict DTIs. Experimental results show that the DFDTI can effectively fuse different types of descriptors and outperform all baseline methods in this paper on all three types of evaluation metrics.