Light field images incorporate the angular and spatial distortion of the captured light. Benefiting from the abundant information contained in the light field, depth estimation emerges as one of its key applications. In this article, we presented a neural network in order to precisely estimate the depth of the light field. The proposed network fully considers the rich feature information of the full light field image, designs an attention mechanism to select more valuable views in the same channel branch and performs multi-feature fusion to enhance the inference accuracy of the occluded region. Meanwhile, the network extracts the edge detail features of the target for guidance to provide more effective information for depth estimation. The proposed network achieves positive results and performance on the commonly used 4D HCI dataset in quantitative, qualitative, and computational cost evaluations.