An Improved Faster R-CNN for Object Detection

Among various target detection algorithms, Faster R-CNN is an algorithm with excellent performance both in detection accuracy and in detection speed at present. However, it still has some shortcomings such as too many negative samples. To address the problem of Faster R-CNN, two strategies, hard negative sample mining and alternating training, are introduced. Hard negative sample mining is used to obtain hard negative sample which retrain the model for improving the trained model, and the alternating training make RPN and Fast R-CNN in Faster R-CNN share convolutional layers, rather than learn two independent networks. The simulation result show that the proposed algorithm has great advantages in terms of detection accuracy.