A Lightweight Foreign Object Debris Detection Algorithm for Airport Runway

Aiming at the problem that small target detection for foreign object debris(FOD) on airport runways, the deep learning target detection algorithm is difficult to deploy on mobile devices due to the large size of the model and too many parameters. A lightweight FOD detection algorithm YOLO-ACIR is proposed based on YOLOX detection model. The spatial pyramid structure is improved, and the spatial pyramid structure with atrous convolution is utilized to reduce the loss of local information and edge information while expanding the receptive field. Secondly, the inverted residual structure and depth-separable convolution are introduced to improve the feature extraction module to provide richer features and reduce the loss of image information. Faster model inference by fusing convolutional and BN layers. Finally, FOD-A data set is used to verify the algorithm. Experimental results show that, in comparison with YOLOX-Nano with A few parameters added, the improved YOLO-ACIR, mAP@0.5:0.95, increases by 3.5 percentage points, and the final inference speed increases by 35.34 percentage points after the convolution layer and BN layer are fused. The comprehensive performance is better than original model, which is a more suitable target detection algorithm for mobile devices.