| With the rapid development of civil aviation business,the number of aircraft takeoffs and landings has gradually increased,and the airport runway carries hundreds or even thousands of aircraft takeoffs and landings every day.Overload operation,weather,temperature difference,groundwater level,lightning strike and other factors may cause damage to the runway,resulting in cracks,thus threatening the take-off,landing and taxiing safety of the aircraft.Therefore,the detection of runway cracks has gradually been paid more and more attention.The target detection algorithm based on deep learning can detect the target in real time and provide a feasible method for intelligent crack detection.Aiming at the time-consuming and laborintensive manual inspections and the insufficient real-time performance of traditional image recognition methods,this paper proposes an airport runway crack detection algorithm based on Yolov5 based on the knowledge of deep learning target detection to detect airport runway cracks.Aiming at the problem of missing detection of small cracks by Yolov5,an improved Yolov5 algorithm based on attention mechanism is proposed.CBAM and CA attention mechanisms are added to the backbone part of the Yolov5 network structure to improve the connection between crack features in channels and spaces.Through comparative experiments,it is found that,after adding CBMA and CA attention mechanisms,the model’s ability to extract effective features of cracks is strengthened,thereby improving the detection ability of small cracks.During the test experiment,it was found that some test results have errors of false detection and missed detection.After analysis,the test images with false detection and missed detection have motion blur.Combined with the conditional generative adversarial network,the blurred crack image is super-resolution.Modular processing,train a CNN network on the fuzzy image through a generative adversarial network as a generator network and a discriminant network,train the two networks in an adversarial way,and perform residual correction on the fuzzy image to improve the training speed and model generalization ability.,and finally fuse with the blurred image to obtain the deblurred result.After deblurring,a comparative experiment is carried out.The experiment shows that the model can effectively detect the test image after removing the motion blur.Finally,the airport runway crack detection platform is designed and implemented,and the test proves that the system can complete the real-time detection of cracks. |
