Automated image detection is a crucial research direction in the field of computer vision, which is widely applied in various industrial production processes. The first step in the detection is to target the detection objectives, involving region segmentation, target localization, feature extraction and target classification. The deep learning methods have been widely applied in the field of image detection, whereas the traditional deep learning methods rely heavily on large scale data for network model training, which poses significant challenges for applications in the domains with limited data availability. The few-shot learning model, emerged in recent years, brings an approach that can utilize less data for available model training. It exhibits the excellent adaptability and generalization capabilities, with the advantages of fast training speed and reduced overfitting risk. In this study, we designed a few-shot learning network model targeting image classification. The few image data samples are pose-corrected by the transferred parameters from classes with sufficient samples, and the corrected image data are transformed into vectors and then passed into the network. Utilizing metric-based methods for similarity estimation among samples, which serve as classification features, facilitates the enhancement of the current network model’s generalization capability. This approach effectively reduces the risk of model overfitting, thereby improving classification accuracy and generalization performance. The outcomes of this study provide a feasible method for solving image classification tasks in practical image detection scenarios, serving as a supplement to the field of few sample learning algorithms.