Abstract: In modern industrial defect detection, deep learning methods have been widely applied but are often limited by the small-sample problem. Addressing this issue can improve detection accuracy and efficiency, reduce technical barrier, and enable small and medium-sized enterprises to lower costs and improve productivity. This paper first defines the small-sample problem and discusses its implications, then reviews the current research, analyzes the advantages and limitations of representative methods and their applicable scenarios, and compares costs and feasibility to offer practical guidance. Finally, future research directions and potential solutions are discussed, aiming to provide theoretical and practical references for addressing the small-sample problem in defect detection and promoting further development and application in this field.