RECEPTIVE FIELD-EXPANDED AND CONTEXT-AWARE SINGLE SHOT DETECTOR

Single-shot detector (SSD) ignores the context from the proposal boxes and is not accurate enough for small object detection. In this paper, a new object detection method, called RFCSSD, is proposed. The RFCSSD method enhances the context information by extending the receptive field of the SSD target detector multi-scale feature map and fuses the depth semantic abstraction to improve the accuracy of small object detection. The experimental results show that the RFCSSD can significantly improve the weakness of SSD and achieve more accurate detection performance. The RFCSSD has better mAP than the existing algorithms.