Research and Implementation of Improved Faster R-CNN Target Detection Algorithm

Target detection is becoming more and more widely used in the field of computer vision. Aiming at the problem that different feature extraction networks lead to different accuracy of detection models, this paper proposes an improved algorithm based on the Faster R-CNN algorithm, and improves the feature extraction part of the experiment. This paper chooses to implement the algorithm on the Pascal VOC20017 data set. First, the LeNet-5 network is used to complete the feature extraction in the improved Faster R-CNN algorithm, and then the VGG16 network originally used by the algorithm is replaced, and then the feature is compared at the RPN network layer. The judgment of the image is normalized on the ROI pooling layer, and finally the candidate area is classified and identified, and the target detection is finally realized, whose accuracy of the single object recognition in the picture has reached more than 95%.