Considering the low small-target detection accuracy, dense targets and various scale patterns in remote sensing image target detection, small-scale and the higher-scale intermediate information is added to the feature fusion of the weighted Bi-directional Feature Pyramid Network (BiFPN) based on the lightweight network EfficientDet-D0 target detection algorithm, which reconstructs the BiFPN network, makes full use of different scale information, and improves the multi-scale target detection accuracy.At the same time, a feature enhancement module fusing dilated convolution with fast normalized fusion is added in the BiFPN to further improve the detection accuracy.In addition, the Dynamic ReLU activation function with dynamic parameters is used to improve the static Swish activation function in the original network.Without affecting the lightweight feature, the improved EfficientDet algorithm improves the mAP of target detection for public dataset Pascal VOC by 11.90 percentage points in comparison with the original algorithm, which is also better than that of other target detection algorithms.For the remote sensing image dataset RSOD, the existing 936 remote sensing image datasets are augmented by Imgaug data augmentation library, and the migration learning is performed by using the improved model.The target detection result is respectively 88.38% and 96.78% before and after data augmentation, which proves that the proposed method can meet the target detection requirement of remote sensing image in practical applications.