Abstract
Due to the frequent impact of aircraft takeoff and landing and the influence of weather temperature changes, airport roads will have different types of underground diseases (DT-CRACK, DT-GAP, DT-LACUNAS and DT-SUBSIDENCE), which affect the road performance and service life, cause safety accidents, and result in a great loss of manpower and material resources. Facing the radar data of underground hidden diseases of airport roads with low recognition and high noise intensity, it is inefficient to recognize the diseases by manual identification, and it is difficult to achieve accurate differentiation and localization of the diseases by the existing detection methods. We analyze and propose an improved algorithm EFA-SSD (Enhanced Feature Aggregation SSD) for automatic detection of airport road subsurface diseases, which solves the problems of strong noise background conditions, severe interference of morphological features of different types of subsurface diseases, and low target recognition. Our model designs RFB module with wider receptive field in the network layer, which effectively suppresses the noise interference around the disease and extracts more disease features from the original radar data; in addition, the detailed texture features of different types of diseases are captured by fusing the shallow features of the model network, which realizes the classification and localization of different types of diseases; and the attention mechanism of spatial channel is introduced to enhance the feature expression ability and improve the generalization ability of the model. The spatial channel attention mechanism is introduced to enhance the feature expression and generalization ability of the model. Compared with the existing classical target detection algorithms, EFA-SSD has the highest mean average precision (mAP) in detecting four types of subsurface diseases, which provides a new idea for subsurface disease detection and contributes to the protection of aviation safety.