Improved YOLOv7 Small Object Detection Algorithm for Seaside Aerial Images

Seaside aerial images due to the high number of small object instances, interference from the background, and occlusion caused by crowded personnel. These issues result in low accuracy of this scenario in the field of object detection. By improving the YOLOv7 algorithm, we proposed a YOLOv7-B model. We reconstructed the detection layer to reduce the miss rate of small objects. The Improved Bi-directional Feature Pyramid Network (IBi-FPN) replaced the Pyramid Attention Network (PANet) of YOLOv7, better integrating deep feature information with shallow feature information. Finally, we added Convolutional Block Attention Module (CBAM) to improve the utilization of effective features. Experiments show that the YOLOv7-B model can improve the detection accuracy of small objects at the seaside while reducing the number of parameters.