基于改进的Faster RCNN的Logo检测分析

Logo detection analysis based on improved faster RCNN

针对因抽象特征提取困难、速度慢而导致的Logo检测耗时长、准确率低的问题,提出一种改进的Faster RCNN的Logo检测方法。采用分水岭算法对图像进行预处理,采用多尺度锚框生成方法生成锚框,用非极大值抑制方法对重叠的候选框进行筛选。实验结果表明,与传统Faster RCNN方法相比,该方法提高了Logo检测的准确性和鲁棒性。

Aiming at the problem that the detection of the abstract features is difficult and the speed is slow,the detection time is low and the accuracy is low.An improved Faster Rcnn logo detection method is proposed.The watershed algorithm is used to preprocess the image,the multi-scale anchor frame generation method is used to generate the anchor frame,and the non-maximum value suppression method is used to filter the overlapping candidate frames.The experimental results show that compared with the traditional FasterRCNN method,this method improves the accuracy and robustness of Logo detection.