Maritime small object detection is critical for UAV-based sea surveillance but remains challenging due to the small size of targets and interference from sea reflections and waves. This paper proposes SeaLSOD-YOLO, a lightweight detection algorithm based on YOLOv11, designed to improve small object detection accuracy while maintaining real-time performance. The method incorporates four key modules: Shallow Multi-scale Output Reconstruction, which fuses shallow and mid-level features to preserve fine-grained details; SPPF-FD, which combines spatial pyramid pooling with frequency-domain adaptive convolution to enhance sensitivity to high-frequency textures and suppress sea-surface interference; attention-based feature fusion, which emphasizes small object features through channel and spatial attention; and dynamic multi-scale sampling, which optimizes feature representation across different scales. Experiments on the SeaDroneSee dataset demonstrate that, compared with YOLOv11s, the proposed method improves precision from 75.6% to 81.9%, recall from 62.6% to 73.5%, and mAP@0.5 from 67.9% to 77.0%. The mAP@0.5:0.95 also increases from 41.1% to 44.9%. The model achieves an inference speed of 256 FPS. Although the parameter size increases from 18.2 MB to 30.8 MB, the method maintains a favorable balance between detection accuracy and computational efficiency. Comparative evaluation further shows superior performance in detecting small maritime objects such as buoys and lifeboats. These results indicate that SeaLSOD-YOLO effectively balances accuracy, efficiency, and real-time capability in complex maritime environments. Future work will focus on further optimization of attention mechanisms and upsampling strategies to enhance the detection of extremely small targets.

In industrial vision, the lack of defect samples is one of the key constraints of depth vision quality inspection. This paper mainly studies defect detection under a small training set, trying to reduce the dependence of the model on defect samples by using normal samples. Therefore, we propose a Knowledge-Embedding Relational Network. We propose a Knowledge-Embedding Relational Network (KRN): firstly, unsupervised clustering and convolution features are used to model the knowledge of normal samples; at the same time, based on CNN feature extraction assisted by image segmentation, the conv feature is obtained from the backbone network; then, we build the relationship between knowledge and prediction samples through covariance, embed the knowledge, further mine the correlation using gram operation, normalize the power of the high-order features obtained by covariance, and finally send them to the prediction network. Our KRN has three attractive characteristics: (I) Knowledge Modeling uses the unsupervised clustering algorithm to statistically model the standard samples so as to reduce the dependence of the model on defect data. (II) Covariance-based Knowledge Embedding and the Gram Operation capture the second-order statistics of knowledge features and predicted image features to deeply mine the robust correlation. (III) Power Normalizing suppresses the burstiness of covariance module learning and the complexity of the feature space. KRN outperformed several advanced baselines in small training sets on the DAGM 2007, KSDD, and Steel datasets.