Research On Detection Method Of Airport Pavement Apparent Disease Under Complex Background

Due to the frequent impact of aircraft take-off and landing and the effect of bad weather,damages will gradually appear on the airport pavement and develop to various disease,which could cause serious flight safety risks,In order to ensure the safety while using the airport pavement,regular inspections and maintenance are top priorities.How to realize automatic airport pavement disease detection with high-performance by computer vision technology has become a hot topic in the current research.The diseases of airport pavement mainly include cracks,corner fractures,broken seams,and repairs.Therefore,this paper makes an in-depth study on the detection methods of apparent diseases on the airport pavement under the complex background.Main research contents and works are as follows:1、Aiming at the problem of cracks,corner fractures,broken seams and repairs,which has the characteristics of different length,less pixel proportion in the image and the contrast is low under the complex background,a deep neural network model,named as DetMSPNet,based on attention mechanism and feature fusion is proposed.The model adopts the attention mechanism to pay more attention to disease information,suppress interference information,and extract features at multiple scales and levels,making the fused features of diseases more representative.Experiments are carried out on the collected image dataset,named as APD,of airport pavement diseases,and the results show that the algorithm can extract the features of pavement diseases more accurately and realize the accurate detection of diseases.2、In order to further improve the detection accuracy of small target diseases in the thin strip shape on the airport pavement,an detection algorithm for small target diseases in the thin strip shape based on improved feature pyramid and feature fusion is proposed.Firstly,the maximum pooling branch is designed for the fusion of shallow and deep features to enhance the ability of the model to locate the small target diseases of thin strip-shaped structure.Secondly,the deep features are input into the pyramid pooling module,so that the disease features can obtain more context information.Finally,the feature pyramid is improved,and the flow alignment module is introduced to integrate the deep feature information into the shallow feature layer by layer to enhance the expression of disease feature.The experimental results show that the algorithm is better than its counterparts in terms of accuracy and robustness on the disease image dataset of small target diseases in the thin strip shape,which is named as STD.