Lightweight detection model of foreign objects in airport runway based on the improved NanoDet-Plus

Foreign objects on the airport runway have a crucial impact on the daily rise and fall of airport flights, and accurate target detection is particularly important for airport safety. To reduce the detection complexity, this experiment uses a lightweight algorithm model NanoDet-Plus for the task of detecting foreign matter on the airport runway. And on the basis of this, the ShuffleNetV2 and FPN channels are changed to reduce the number of parameters. The activation function and the loss function are modified simultaneously to improve mAP accuracy. Experiments on the FOD data set show that the Params parameter volume of the improved algorithm model decreased by 59.5%, while its mAP increased by 2.0%. Compared with other algorithms for airport runway foreign matter detection, this experiment not only improves the accuracy of the detection, but also greatly reduces the parameter volume of the model, and has lower requirements for experimental equipment, which can better meet the accuracy and rapid requirements of airport runway foreign matter detection.