YOLOv8-Plus: A Small Object Detection Model Based on Fine Feature Capture and Enhanced Attention Convolution Fusion

Small object detection holds significant value in various practical applications. However, due to their limited pixel coverage, weak feature information, and susceptibility to background noise, YOLOv8 faces challenges in detecting small objects, including low recognition accuracy and missed detections. To address these issues, we propose an improved small object detection model, YOLOv8-Plus. First, to tackle the difficulty in detecting subtle features of small objects in the YOLOv8 model, we add a dedicated output layer, TDLayer, in addition to the original three output layers. This new layer generates larger feature maps, allowing for better differentiation of fine details in small objects. Second, to improve feature processing, we design the C2FDSC module, which adaptively adjusts detection strategies based on the shape and characteristics of small objects, ensuring fine details are captured. Finally, to mitigate the impact of background noise, we introduce the EACF module, which combines the advantages of CNNs and attention mechanisms to effectively reduce noise interference, improving both accuracy and robustness in small object detection. Experimental results on the VisDrone2019 dataset show that the improved YOLOv8-Plus model achieves a 6.7% and 4.7% increase in mAP50, respectively, compared to the baseline model. YOLOv8-Plus outperforms other state-of-the-art models, demonstrating superior performance in small object detection tasks in complex scenarios.