Abstract: To address the issues of low accuracy, missed detections and false detections in Ground Penetrating Radar (GPR) defect target detection tasks, a LMC-YOLOv8n-based method for GPR defect target recognition is proposed. Firstly, the SimAM attention mechanism is incorporated into the feature fusion network. Then, the RFAConv (Receptive Field Attention Convolution) module was incorporated. A lightweight LSCSBD detection head was also designed. These improvements effectively enhance the efficiency of feature fusion. They enable more complete extraction of small target details from feature maps. Additionally, the model's ability to learn global contextual information is strengthened. Finally, model accuracy is further improved through CWD (Channel-Wise Distillation) knowledge distillation. Experimental results demonstrate that the LMC-YOLOv8n network model achieves improvements of 3.8% in mAP50(%) and 4.6% in mAP50: 95(%) compared to the original network. IT ENHANCES DETECTION ACCURACY WITHOUT ADDING EXTRA PARAMETERS. IT IS WELL-SUITED FOR IDENTIFYING DEFECTS LIKE GROUND VOIDS AND CRACKS. MOREOVER, THE MODEL MEETS BOTH ACCURACY AND REAL-TIME PROCESSING REQUIREMENTS. WHILE IMPROVING TARGET RECOGNITION PERFORMANCE, IT ALSO REDUCES COMPUTATIONAL LOAD AND PARAMETER COUNT. THESE ADVANTAGES INDICATE PROMISING APPLICATION PROSPECTS.