An Improved SSD Object Detection Algorithm Based on Attention Mechanism and Feature Fusion

Abstract The Single Shot MultiBox Detector (SSD) is a well-known object detection method, but its of small objects not effective. This paper makes modifications to the SSD method address insufficient semantic information in low-level feature maps, thus enhancing detectability for objects. First, Feature Pyramid Network (FPN) incorporated into so that shallow map, which primarily utilized detecting objects, contains more addition rich location information. Second, Convolutional Block Attention Module (CBAM) introduced reinforce network’s capability learn key features and improve missed detections. experimental data indicate this algorithm achieves 78.1% mAP PASCAL VOC2007test, 3.9% improvement compared with conventional SSD, also has great Fast R-CNN Faster R-CNN. As well as,this better meets real-time requirements.