Algorithm for the Detection of Thin Strip-Shaped Structural Small Diseases on Airport Pavement based on improved pyramid and feature fusion

Abstract

Airport Pavement disease detection is an important part of ensuring operational safety. The small target diseases of thin strip-shaped structure in cracks, cornerfractures, and broken seams, have the characteristics of narrow width, different length, and less pixel proportion in the image, and the contrast is low under the complex background. These factors lead to fail when using the existing detection algorithms. Therefore, a deep Neural Network algorithm, named as DFAMNet, based on improved pyramid and feature fusion is proposed. Firstly, the maximum pool branch is designed to facilitate the fusion of different features from shallow and deep layers, and improve the positioning ability of the model for diseases. Then, the deep features are fed into pyramid pooling module, so that the disease features contain more global context information. Finally, the feature pyramid is improved, and the flow alignment module is introduced to better integrate the deep feature information into the shallow feature layer by layer, to enhance the expression of disease features. Compared with four classical object detection algorithms, the experimental results show that the proposed algorithm achieves a mAP of 50.35%, which is better than its counterparts.