Attention-Infused Network for Detecting Small Marine Debris Objects in Aerial Images

Small object detection is a critical task in computer vision, with applications ranging from vehicle and pedestrian monitoring to environmental protection. One particularly urgent application is the detection of marine debris, which is vital for mitigating ecological damage and ensuring navigational safety. Automating this process through aerial imagery and computer vision can substantially reduce manual effort. The FloW-Img dataset introduced by Cheng and colleagues provides annotated aerial images of floating debris in inland waters, offering a valuable benchmark for this task. However, accurately detecting small-scale debris remains a significant challenge. To address this challenge, we propose a novel detection framework that integrates attention-based mechanisms—including the Simple Attention Module (SimAM) with slicing operations and the Convolutional Block Attention Module (CBAM)—into a multi-scale feature extraction backbone. We also incorporate dynamic and separable convolution layers to enhance representational power while maintaining computational efficiency. Experimental results on the benchmark dataset show that our method achieves the highest precision score (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 8 6 0}$</tex>) among all evaluated models while maintaining a strong recall rate (0.837). The model also delivers a competitive mean average precision (mAP) score at 0.5 Intersection over Union (IoU) threshold (0.867), surpassing several state-of-the-art object detectors in precision and matching the best in detection accuracy. These findings highlight the effectiveness of our attention-enhanced approach for detecting small marine debris objects in aerial imagery.