基于改进SegFormer的轻量级坝体裂缝检测模型OA

Real-time detection of cracks in the dam body is crucial for ensuring the safe and stable operation of the power station and ensuring personnel safety.To address the low efficiency of traditional manual inspections and insufficient detection accuracy of existing deep learning models in complex environments,a dam crack detection method based on improved SegFormer model is proposed.This method innovatively introduces a lightweight multi-scale linear attention mechanism.Firstly,parallel multi-scale feature extraction is utilized to enhance the model's capability to capture crack features at different scales.Secondly,ReLU lin-ear attention is used to replace traditional Softmax attention,significantly reducing the number of parameters and improving com-putational efficiency.Finally,an improved loss function combining Focal loss and gradient extremum regularization is designed to effectively mitigate the class imbalance problem and improve the detection of fine cracks.The experimental results show that the improved model achieves 0.6954,0.7897 and 0.7875 in mIoU,mFscore,and mRecall metrics,respectively,representing improvements of 0.0275,0.0287 and 0.0710 over the original SegFormer model.The segmentation accuracy is maintained at a high level,while the number of parameters is significantly reduced,and processing speed is enhanced.This method offers a novel approach for efficient and accurate crack detection in dam body.