A method of underwater bridge structure damage detection method based on a lightweight deep convolutional network

The problem of the underwater structure disease bridge is increasingly obvious, which has seriously affected safe operation structure, so it necessary to detect regularly. There are many kinds diseases. This paper targets structural crack diseases adopts multiple image recognition networks for verification, compares advantages different networks, and takes YOLO-v4 network as main body build a lightweight convolutional neural network.Mobilenetv3 replaced CSPDarkent backbone feature extraction network, while layer scale Mobilenetv3 was modified, extracted preliminary input into enhanced fusion. PANet by depthwise separable convolution. Using ablation experiments compare performance four algorithm combinations in networks. At same time, identification accuracy each tested various experimental environments, feasibility verified application damage identification.