【摘要】 传统的图像识别方法不能有效地检测出电力部件的具体位置,同时在干扰物较多的场景下识别准确率较低。针对上述问题,提出来一种改进的Faster R-CNN的电力部件识别算法,使用深度卷积网络自动从图像中提取最适合电力部件特征。Faster R-CNN方法,利用"Hot Anchors"代替均匀采样的锚点来避免大量额外的计算,提高检测效率。最后,21检测框架被修改成4类电力部件检测。实验结果表明:改进的Faster R-CNN的电力部件识别算法,在检测效率和准确率方面分别提升了16.1%和3.8%。更多还原

【Abstract】 In the conventional image recognition method, the specific position of the electrical devices cannot be effectively detected, and the recognition accuracy is low in the scene with many interferents. In view of the problems above, an improved electrical devices algorithm for Fast R-CNN is proposed, which uses a deep convolution network to automatically extract features that are most suitable for electrical devices from images. Based on the Faster R-CNN framework, a great amount of extra calculation is avoided by using "Hot Anchors" instead of uniformly sampled anchor points, thus improving the detection efficiency. Finally, the Faster R-CNN framework for a 21 classification problem was modified to be a framework for electrical devices detection. The experimental results show that the improved Fast R-CNN electrical devices detection algorithm has an improvement in detection efficiency and accuracy, which is 16.1% and 3.8% respectively.更多还原