Zhen Xu

Dynamic wavelet optimization and hierarchical reinforcement learning for fault diagnosis in wind turbines under variable operating conditions

EFEM-YOLO: an efficient feature extraction network for surface defect detection of photovoltaic cell

Stable operation of photovoltaic (PV) cells is crucial for reliable and efficient electricity generation in power systems. However, their poor defect characterization, large size variation, and high background noise lead to low automatic identification accuracy. In this paper, a YOLO-based network with an efficient feature extraction network, termed EFEM-YOLO, is proposed for accurate defect detection in PV cells. First, a multi-scale dilated convolution feature pyramid module is proposed. By constructing an adaptive feature pyramid via parallel convolutional paths, it enhances the representation of multi-scale defects. Second, the C2f-dynamic gated activation network is introduced. By incorporating a dynamic gated nonlinear activation mechanism and a cross-stage dual-branch feature aggregation strategy, the model’s adaptability to multi-scale defects and recognition accuracy are improved. Finally, a novel SmartShapeIoU loss function is proposed. This mitigates localization bias caused by object scale variations in high-noise environments, thereby improving bounding box regression accuracy. Experimental results demonstrate that the proposed method outperforms state-of-the-art methods in detection accuracy. Additional dataset experiments validate the superior detection accuracy and generalization of the proposed method for small object detection.

YOLO-TVP: Real-time open-vocabulary object detection with Semantic-Target Soft Cross-Entropy and text-visual prompts

Open-vocabulary object detection (OVD) models generally leverage vision-language pre-trained models to recognize novel categories via arbitrary text prompts. Nevertheless, their performance is restricted by two core limitations: semantic discontinuity arising from hard binary supervision in contrastive learning, and the inflexibility of single prompts to convey complex detection intents in practical scenarios. To tackle these issues, this paper proposes YOLO-TVP (Text–Visual Prompt), an efficient OVD framework built on the YOLO architecture, with two key designs. First, a Semantic-Target Soft Cross-Entropy (ST-SoftCE) loss is introduced. It constructs semantic target distributions from inter-class similarities in the shared text embedding space for open-vocabulary inference and supervises the detector's classification branch. This design embeds semantic relevance into supervision and enhances fine-grained discrimination among semantically similar categories. Second, a unified class-prompt embedding interface is developed to support both textual and visual prompts. Text prompts are projected into the shared prompt space, while visual prompts are formed by learnable weighted fusion of CLIP semantic priors and backbone visual features, eliminating the need for multi-modal prompt concatenation during inference. Experiments validate the effectiveness: on Flickr30k image-text retrieval with ResNet-50, ST-SoftCE improves Recall@1 by 9.56% over standard cross-entropy. For open-vocabulary detection trained on COCO+Flickr30k and evaluated on LVIS, ST-SoftCE delivers a 13.15% relative mAP 50 improvement to YOLO-World. With scratch-trained detector weights and frozen vision-language encoders used only for prompt-side priors, YOLO-TVP achieves competitive prompt-conditioned detection while preserving real-time efficiency, verifying its advances in semantic alignment and prompt-adaptive detection.