基于改进Faster R-CNN的红外舰船目标检测算法

针对Faster R-CNN算法中对于红外舰船目标特征提取不充分、容易出现重复检测的问题,提出了一种基于改进Faster R-CNN的红外舰船目标检测算法.首先通过在主干网络VGG-16中依次引出三段卷积后的3个特征图,将其进行特征拼接形成多尺度特征图,得到具有更丰富语义信息的特征向量;其次基于数据集进行Anchor的...

Infrared Ship Target Detection Algorithm Based on Improved Faster R-CNN

To solve the problem of insufficient feature extraction and repeated detection of infrared ship targets by the Faster R-CNN algorithm, a ship target detection algorithm based on an improved Faster R-CNN is proposed. First, three feature graphs are drawn from the backbone network, VGG-16, after a three-segment convolution, and the features are spliced to form a multi-scale feature graph to obtain a feature vector with richer semantic information; second, the Anchor is improved based on the dataset, and the number and size of the Anchor boxes are reset; finally, the loss function of the improved Faster R-CNN is optimized to improve the feature extraction ability of the target. An analysis of the experimental results on the test dataset demonstrates that the average accuracy of the improved detection algorithm was 83.98%, which is 3.95% higher than that of the original Faster RCNN.