基于改进Faster RCNN的钢线圈头部小目标检测算法

Improved faster RCNN based steel coil head detection algorithm for small objects

针对钢线圈剪切自动化中数据难以收集、钢线圈头部像素面积小及位置不确定等问题,提出一种改进Faster RCNN钢线圈头部小目标检测算法。对于目标像素面积占比小,区别特征不明显的问题,通过加入FPN对融合特征进行检测,同时,在网络中加入PAM并行注意力模块,提高特征图质量和区域建议网络生成的预选框质量。实验表明,改进后的网络在VOC2007数据集上mAP比原始Faster RCNN提高了约5%;在钢线圈数据集上mAP比原始Faster RCNN提高了约4%,实验表明改进算法具有一定的有效性。

Aiming at the problems of the difficult data collection,small pixel area and uncertain position of steel coil head in steel coil cutting automation,an improved Faster RCNN algorithm for small objects detection of steel coil head was proposed.For the problem that the target pixel area accounts for a small proportion and the distinguishing features are not obvious,the fusion features were detected by adding FPN.At the same time,PAM parallel attention module was added to the network to improve the quality of the feature map and the quality of the preselected box generated by the region proposal network.The experiment shows that the detection accuracy of the improved network is improved by about 5%on VOC2007 dataset;the detection accuracy is improved by about 4%on the steel coil dataset.Experimental results show that the improved algorithm is effective.