Adaptive aerial target detection method based on multi-scale deep learning

The invention discloses a self-adaptive aerial photography target detection method based on multi-scale deep learning. The method comprises the following steps: obtaining a training image set and a test image set; extracting target features in the unmanned aerial vehicle image by using multi-convolution; an adaptive module CBMA is added into the convolved feature map to change original feature weight distribution; performing up-sampling on the multi-layer feature mapping, and fixing the multi-layer feature mapping to a specific dimension through 1 * 1 convolution; fusing the multi-scale features to obtain a high-resolution feature map; and target classification positioning is completed through the fused high-resolution image. According to the method, the feature extraction capability is improved, the small target detection effect is enhanced, the aerial image target detection accuracy is effectively improved, and the method has good generalization capability and wide application range. 本发明公开了一种基于多尺度深度学习的自适应航拍