应用改进YOLOX轻量化卷积神经网络的水下工程裂纹识别OA

Aiming at the drawbacks of excessive parameters and low efficiency in the convolutional neural network model,an underwater engineering crack recognition method based on improved YOLOX lightweight convolutional neural network is pro-posed.The depthwise separable convolution is used to replace ordinary convolution in YOLOX network,to reduce model re-dundant parameters and achieve lightweight design.On the basis of the YOLOX backbone feature extraction network frame-work,a Transformer visual module is embedded to replace the cross stage partial(CSP)loop structure at the end of the back-bone network.The attention mechanism is added to enhance the ability to extract key information.The experimental results show that the proposed method has good predictive ability,and the crack recognition error is generally between 1.10％and 1.40％.The ablation experiment shows that the underwater crack recognition accuracy of the proposed method reaches 96.5％,with a frame rate of 31.2,presenting the optimal recognition accuracy while ensuring a high crack recognition rate.