摘要: Aluminum alloy stamped parts are widely used in highprecision industrial fields such as aerospace and automotive industries, where timely crack detection is crucial to ensure their performance and safety. Traditional crack detection methods mainly rely on manually designed feature extraction algorithms. While certain success has been achieved in simple scenarios, there are still limitations in detection accuracy and robustness when dealing with complex backgrounds and significant variations in crack morphology. This paper proposes a crack detection method called CGCYOLO, which integrates Channel Aware Fusion (CAF), GSSPPF, and Cross Scale Path Aggregation Network (CSPAN) to enhance the model’s feature extraction and detection capabilities. Experimental results show that CGCYOLO demonstrates higher accuracy and stronger robustness in crack detection tasks for aluminum alloy stamped parts, indicating its broad application potential.