The rapid advancement of deep learning has significantly accelerated progress in target detection. However, the detection of small targets remains challenging due to their susceptibility to size variations. In this paper, we address these challenges by leveraging the latest version of the You Only Look Once (YOLOv7) model. Our approach enhances the YOLOv7 model to improve feature preservation and minimize feature loss during network processing. We introduced the Spatial Pyramid Pooling and Cross-Stage Partial Channel (SPPCSPC) module, which combines the feature separation and merging ideas. To mitigate missed detections in small target scenarios and reduce noise impact, we incorporated the Coordinate Attention for Efficient Mobile Network Design (CA) module strategically. Additionally, we introduced a dynamic convolutional module to address misdetection and leakage issues stemming from significant target size variations, enhancing network robustness. An experimental validation was conducted on the FloW-Img sub-dataset provided by Okahublot. The results demonstrated that our enhanced YOLOv7 model outperforms the original network, exhibiting significant improvement in leakage reduction, with a mean Average Precision (mAP) of 81.1%. This represents a 5.2 percentage point enhancement over the baseline YOLOv7 model. In addition, the new model also has some advantages over the latest small-target-detection algorithms such as FCOS and VFNet in some respects.
Crowding and occlusion pose significant challenges for pedestrian detection, which can easily lead to missed and false detections for small-scale and occluded pedestrian objects in dense pedestrian scenarios. To enhance dense pedestrian detection accuracy, we propose the Residual Transformer YOLO (RT-YOLO) algorithm in this paper. The RT-YOLO algorithm enhances the multi-scale fusion strategy based on YOLOv7 and introduces a dedicated detection layer for small-scale occluded targets. It also integrates Resnet and Transformer structures to improve the small-scale feature layer and detection head, enhancing feature extraction capabilities. Additionally, the RT-YOLO algorithm incorporates the Normalization-based Attention Module (NAM) into the backbone and neck networks to identify the region of interest. The experiments demonstrate that on the CrowdHuman and WiderPerson datasets, at IOU (Intersection over Union) = 0.5, the overall improvement in is 3.8% and 3.4%. In the IOU range from 0.5 to 1, the improvement in : 95 is 5.1% and 4%. RT-YOLO achieves an FPS of 67, maintaining real-time performance. On the VOC2007 dataset, has been enhanced by 5.1%, indicating higher effectiveness and robustness.
1 December 2025 in Measurement Science and Technology
In response to the challenges of insufficient feature information, strong background interference and limited hardware resources faced by object detection in UAV infrared images, we propose a lightweight and efficient UAV infrared target detection network based on multi-scale attention (LEANet). First, a perception-enhanced lightweight convolution module based on C3K2 is designed to boost the backbone network’s capability in feature extraction and reduce the model’s parameter count. Then, an efficient multi-scale attention module is to capture long-range dependencies between targets and contextual information, thus enhancing focus on potential targets while suppressing background noise in complex scenes. Next, an efficient lightweight feature pyramid network is designed, which not only incorporates a dynamic upsampling method to tackle the blurring and detail loss in infrared small targets caused by traditional upsampling methods but also introduces group sparse convolution to improve model computational efficiency and a lightweight feature extraction module to diminish the model’s parameters. Finally, the Shape-IoU loss function is utilized to boost the model’s adaptability to diverse target shapes and sizes, thereby boosting the detection accuracy of diverse targets observed via UAV infrared sensors. Experimental results on the public HIT-UAV dataset and the self-built X-UAV dataset show that LEANet achieves the optimality mAP50 with minimal increases in computational effort and inference time compared to the lightweight models in the Yolo series. Notably, LEANet reduces model parameters by 13.6% relative to YOLOV5n, the smallest model in the YOLO series.
Multimodal pedestrian detection has garnered significant attention due to its potential applications in complex scenarios. The complementarity characteristics between infrared and visible modalities can enhance detection performance. However, the design of cross-modal fusion mechanisms and the in-depth exploration of inter-modal complementarity still pose challenges. To address this, we propose TCCDNet, a novel network integrating cross-modal complementarity. Specifically, the efficient multi-scale attention C2f (EMAC) is designed for the backbone, which combines the C2f structure with an efficient multi-scale attention mechanism to achieve feature weighting and fusion, thereby enhancing the model’s feature extraction capacity. Subsequently, the cross-modal complementarity (CMC) module is proposed, which enhances feature discriminability and object localization accuracy through a synergistic mechanism combining channel attention and spatial attention. Additionally, a deep semantic fusion module (DSFM) based on a cross-attention mechanism is incorporated to achieve deep semantic feature fusion. The experimental results demonstrate that TCCDNet achieves a MR−2 of 7.87% on the KAIST dataset, representing a 3.83% reduction compared to YOLOv8. For the other two multimodal pedestrian detection datasets, TCCDNet attains mAP50 scores of 83.8% for FLIR ADAS and 97.3% for LLVIP, outperforming the baseline by 3.6% and 1.9% respectively. These results fully validate the effectiveness and advancement of the proposed method.