【摘要】 针对Faster R-CNN目标检测算法中小目标检测精度不高和定位不准确的问题,提出一种基于多层特征融合的小目标检测方法。运用多层特征融合的方式丰富特征图的信息,提升小目标检测的精度,在目标候选区域时对锚框进行新的设定,丰富锚框的比例与大小,进一步加强目标候选区域的生成,提升小目标检测精度和增强目标的定位效果。在测试数据集PASCAL VOC 2007进行验证,验证结果表明,与Faster R-CNN相比,检测速度没有受到明显的影响,目标总体检测精度提升了2.2%,其中小目标检测精度提升更为显著。更多还原

【Abstract】 To solve the problem that the accuracy of small object detection is not high and the positioning is not accurate in Faster R-CNN object detection algorithm,a small target detection method based on multi-feature fusion was proposed.The multifeature fusion method was used to enrich the information of the feature map,and the accuracy of the small object detection was improved.The anchor frame was newly set in the object candidate region,the proportion and size of the anchor frame were enriched and the generation of the object candidate region was further enhanced,which was beneficial to the improvement of the small object detection accuracy.The positioning of the object was enhanced.In the test data set PASCAL VOC 2007,the verification results show that compared with Faster R-CNN,the detection speed is not significantly affected,the overall object detection accuracy is improved by 2.2%,and the detection accuracy of small objects is significantly improved.更多还原