The bin packing problem is a classic combinatorial optimization problem with widespread applications across various industries. This paper proposes an end-to-end deep reinforcement learning method for predicting the packing solution of 2D regular items. The method decomposes the problem into two steps: item selection and placement. In the selection step, both the item and its corresponding rotation angle are predicted. In the placement step, a placement strategy is integrated to pre-generate candidate placement positions, reducing the action space for placement. The model then predicts the optimal placement position for the selected item. Additionally, multi-dimensional container features are extracted, allowing the model to make better selection and placement decisions. Experimental results demonstrate that the proposed method performs well on the 2D bin packing task.