Small Target Detection Algorithm in Remote Sensing Image Based on Improved Yolo

摘要

Aiming at the problem of low detection accuracy of small remote sensing image targets with small size and unobvious features in existing target detection algorithms, a small target detection algorithm based on improved yolo_v4 is proposed. The algorithm expands the scale to four during target detection, and removes the feature scale map of the minimum perception field when output; replaces the convolution layer of the feature fusion network with a hole convolution to maintain a higher resolution and a larger receptive field; The deconvolution operation performs up-sampling of high-level features, so that low-level features can learn richer semantic information. The experiment analyzed the remote sensing image with the high-voltage electric tower as the small target, and the average accuracy was increased from 90.06% to 91.47%, indicating that the algorithm can effectively improve the detection accuracy of the small target in the remote sensing image.