Due to the broad usage and widespread popularity of drones, the demand for a more accurate object detection algorithm for images captured by drone platforms has become increasingly urgent. This article addresses this issue by first analyzing the unique characteristics of datasets related to drones. We then select the widely used YOLOv7 algorithm as the foundation and conduct a comprehensive analysis of its limitations, proposing a targeted solution. In order to enhance the network’s ability to extract features from small objects, we introduce non-strided convolution modules and integrate modules that utilize attention mechanism principles into the baseline network. Additionally, we improve the semantic information expression for small targets by optimizing the feature fusion process in the network. During training, we adopt the latest Lion optimizer and MPDIoU loss to further boost the overall performance of the network. The improved network achieves impressive results, with mAP₅₀ scores of 56.8% and 94.6% on the VisDrone2019 and NWPU VHR-10 datasets, respectively, particularly in detecting small objects.

This paper addresses the challenge of small object detection in remote sensing image recognition by proposing an improved YOLOv8-based lightweight attention cross-scale feature fusion model named LACF-YOLO. Prior to the backbone network outputting feature maps, this model introduces a lightweight attention module, Triplet Attention, and replaces the Concatenation with Fusion (C2f) with a more convenient and higher-performing dilated inverted convolution layer to acquire richer contextual information during the feature extraction phase. Additionally, it employs convolutional blocks composed of partial convolution and pointwise convolution as the main body of the cross-scale feature fusion network to integrate feature information from different levels. The model also utilizes the faster-converging Focal EIOU loss function to enhance accuracy and efficiency. Experimental results on the DOTA and VisDrone2019 datasets demonstrate the effectiveness of the improved model. Compared to the original YOLOv8 model, LACF-YOLO achieves a 2.9% increase in mAP and a 4.6% increase in mAP_S on the DOTA dataset and a 3.5% increase in mAP and a 3.8% increase in mAP_S on the VisDrone2019 dataset, with a 34.9% reduction in the number of parameters and a 26.2% decrease in floating-point operations. The model exhibits superior performance in aerial object detection.

In this paper, an object detection and recognition method based on improved YOLOv5 is proposed for application on unmanned aerial vehicle (UAV) aerial images. Firstly, we improved the traditional Gabor function to obtain Gabor convolutional kernels with better edge enhancement properties. We used eight Gabor convolutional kernels to enhance the object edges from eight directions, and the enhanced image has obvious edge features, thus providing the best object area for subsequent deep feature extraction work. Secondly, we added a coordinate attention (CA) mechanism to the backbone of YOLOv5. The plug-and-play lightweight CA mechanism considers information of both the spatial location and channel of features and can accurately capture the long-range dependencies of positions. CA is like the eyes of YOLOv5, making it easier for the network to find the region of interest (ROI). Once again, we replaced the Path Aggregation Network (PANet) with a Bidirectional Feature Pyramid Network (BiFPN) at the neck of YOLOv5. BiFPN performs weighting operations on different input feature layers, which helps to balance the contribution of each layer. In addition, BiFPN adds horizontally connected feature branches across nodes on a bidirectional feature fusion structure to fuse more in-depth feature information. Finally, we trained the overall improved YOLOv5 model on our integrated dataset LSDUVD and compared it with other models on multiple datasets. The results show that our method has the best convergence effect and mAP value, which demonstrates that our method has unique advantages in processing detection tasks of UAV aerial images.

Object detection in low-light conditions is increasingly relevant across various applications, presenting a challenge for improving accuracy. This study employs the popular YOLOv7 framework and examines low-light image characteristics, implementing performance enhancement strategies tailored to these conditions. We integrate an agile hybrid convolutional module to enhance edge information extraction, improving detailed discernment in low-light scenes. Convolutional attention and deformable convolutional modules are added to extract rich semantic information. Cross-layer connection structures are established to reinforce critical information, enhancing feature representation. We use brightness-adjusted data augmentation and a novel bounding box loss function to improve detection performance. Evaluations on the ExDark dataset show that our method achieved an mAP₅₀ of 80.1% and an mAP_50:95 of 52.3%, improving by 8.6% and 11.5% over the baseline model, respectively. These results validate the effectiveness of our approach for low-light object detection.

The objects in UAV aerial images have multiple scales, dense distribution, and occlusion, posing considerable challenges for object detection. In order to address this problem, this paper proposes a real-time multi-scale object detection method based on an improved YOLOv7 model (ATS-YOLOv7) for UAV aerial images. First, this paper introduces a feature pyramid network, AF-FPN, which is composed of an adaptive attention module (AAM) and a feature enhancement module (FEM). AF-FPN reduces the loss of deep feature information due to the reduction of feature channels in the convolution process through the AAM and FEM, strengthens the feature perception ability, and improves the detection speed and accuracy for multi-scale objects. Second, we add a prediction head based on a transformer encoder block on the basis of the three-head structure of YOLOv7, improving the ability of the model to capture global information and feature expression, thus achieving efficient detection of objects with tiny scales and dense occlusion. Moreover, as the location loss function of YOLOv7, CIoU (complete intersection over union), cannot facilitate the regression of the prediction box angle to the ground truth box—resulting in a slow convergence rate during model training—this paper proposes a loss function with angle regression, SIoU (soft intersection over union), in order to accelerate the convergence rate during model training. Finally, a series of comparative experiments are carried out on the DIOR dataset. The results indicate that ATS-YOLOv7 has the best detection accuracy (mAP of 87%) and meets the real-time requirements of image processing (detection speed of 94.2 FPS).