To address the issues of low detection accuracy and misclassification of similar components in existing micro-motor armature surface defect detection methods, this study proposes an improved YOLOv11n-based approach for detecting surface defects in micro-motor armatures by integrating deep learning techniques. First, by adopting the concepts of efficient partial convolution and residual connections, we designed a partial multi-scale feature aggregation module named C3K2-multi scale partial feature aggregation (C3K2-MSPFA). This significantly enhances the detection capability for objects at different scales, thereby improving the model’s detection accuracy. Second, we introduce omni-dimensional dynamic convolution (ODConv) and adaptive downsampling (ADown) to design a lightweight omni-dimensional adaptive downsampling (OD-ADown) module, reducing the parameter count and computational load of the C3K2-MSPFA module. Finally, to address the weak generalization and slow convergence issues of complete-IoU loss (CIoU) in detection tasks, we employ distance-IoU (DIoU) loss to enhance model accuracy and accelerate bounding box regression speed. Experiments were conducted on a self-built dataset, and the results showed that the improved model achieved an average accuracy of 94.2%, a recall rate of 90.9%, an accuracy rate of 95.9%, 2.15×106 parameters, and a model size of 4.5 MB. Compared with the original YOLOv11n network model, the accuracy, recall, and average precision have been improved by 1.3%, 4.6%, and 2.7%, respectively. Compared with the original model, the number of parameters and model size were reduced by 16.67% and 15.09%, respectively. It can meet the deployment requirements of mobile and embedded devices, and provide certain effective technical support for the development of surface defect detection of armatures in micro and special electric machines.