YOLOv5-CBAM: A Small Object Detection Model Based on YOLOv5 and CBAM

At present, object detection is a crucial technology, which plays an indispensable role in the field of object recognition, such as garbage detection conducted by unmanned boats in cleaning rivers and ship detection in traveling. Nowadays, YOLO algorithm is widely used and object detection requires the system to process external environment information in real time. With a fast processing speed, YOLO can process video streams at a very high frame rate. Thanks to its high-speed detection capabilities, it has become an ideal candidate algorithm in autonomous driving. However, the current YOLO technology is lack of detection accuracy for small objects, which may lead to potential risks and uncertainties in scenarios that require high-precision detection. Thus, this paper proposes a YOLOv5-CBAM small object recognition model, which adds two layers of CBAM to the backbone network of the original YOLOv5 model. Channel attention mechanism and spatial attention mechanism are introduced to promote the detection accuracy of small objects. In this paper, through experiments on WSODD dataset and compared with YOLOv5, it is confirmed that the YOLOv5-CBAM model proposed in this paper achieves better results in small object detection, play an indispensable part in the obstacle detection.