Airport Small Target Algorithm Based on Convolution Kernel Attention Mechanism

Abstract

Aiming at the complex background of the airport target and small targets, a YOLOv3 airport small target detection algorithm based on convolution kernel attention mechanism is proposed. Firstly, the convolution kernel attention mechanism is embedded in the three feature fusion layers of YOLOv3 to improve the score of the airport small target. Secondly, design a Gaussion IOU(Intersection over Union) algorithm to obtain the JS divergence by constructing a three-dimensional Gaussian function, and replace the original IOU with Gaussion IOU to improve the recall rate of target detection. Finally, an improved non-maximum suppression algorithm is proposed, which uses Gaussian function as the attenuation coefficient to screen anchor frames. Using the designed algorithm, the [email protected] for the three targets of airplanes, cars, and people is 0.951, which is 0.1 higher than the original YOLOv3. The results show that the designed algorithm can accurately detect small targets at the airport.