改进YOLO11n的雾天路面缺陷轻量化检测OA

Aiming at the problems of low detection accuracy and large number of detection model parame-ters in road defect detection methods in foggy scenarios,we proposed to improve the lightweight detection method of YOLO11n foggy road defects,aiming to improve the model detection accuracy while being more conducive to its lightweight deployment.First,a front-end Dehaze-Network(DH-Net)was con-structed in the backbone network,which maintained the consistency of the dehaze image structure while re-alizing the joint optimization of the detection task orientation through the channel normalization and cross-layer statistic transfer mechanism,and reduced the influence of the fog on the detection effect;second,in order to enhance the ability to extract the details of the defects,the Adaptive Downsampling Module(AD-own)replaced traditional convolution to reduce the number of parameters and retain key spatial features;then,to reduce the impact of foggy scenes and complex road conditions on detection,an efficient multi-branch auxiliary feature pyramid network was designed to enhance the cross-scale characterization of foggy targets through dynamic convolutional kernel adaptation and weighted bi-directional feature pyramid fu-sion;and lastly,the lightweighting of the detection header was improved by using part of the convolution to partially convolution operation to reduce the computational overhead.Experiments across various datas-ets demonstrate that the improved model achieves a mAP increase of 2.1%and 3%respectively over the baseline,whilst reducing the number of parameters by 47.2%.This method provides a high-precision and low-resource-consumption solution for foggy pavement inspection.