【摘要】 针对车辆型号繁多、部分型号间差异较小带来车辆分类困难的问题,构建一种基于改进的Mask R-CNN细粒度车辆型号识别算法。改进后的算法采用聚合残差-特征金字塔网络(ResNeXt-FPN)提取特征图;调整了区域建议网络(RPN)中锚(Anchor)的尺寸大小;用Soft-NMS代替了非极大值抑制算法(NMS),以提高检测精度;去除掩码分支,节省了预测时间。为了验证算法改进的效果,将其与最新的目标检测算法进行对比。实验结果证明,改进的算法提高了车辆识别的准确率,比原始算法准确率提升了2%。更多还原

【Abstract】 This model aims to solve the classification difficulty caused by the wide variety of car models and little di-fferentiation between some models, an improved fine-grained car recognition algorithm based on Mask R-CNN was propose-d. The improved algorithm uses aggregated residual-feature pyramid networks(ResNeXt-FPN) to extract feature maps; the anchor ratio in the region proposal network(RPN) is adjusted; the Soft-NMS algorithm is used to replace the non-maximum value suppression(NMS) algorithm in order to improve the detection accuracy; removing the mask branch. In order to verif-y the effectiveness of the improved algorithm, it was compared with the state-of-the-art object detection methods. The exper-imental results show that the improved algorithm improves the accuracy of car recognition, the performance improvement is about 2%.更多还原