Object Detection With Region-based Convolutional Neural Network

Deep convolution neural network has been widely applied to various fields of computer vision in recent years and has greatly improved the performance of computer vision. The performance of object detection is limited by the feature representation of the object before the presence of the depth convolution neural network, and the actual performance is poor.Deep convolution neural network is applied to object detection because of its powerful feature extraction ability.This study is devoted to the study of deep convolution neural network and its application in object detection. Firstly, we focus on the structure of deep convolution neural network, over-fitting problem and gradient vanishing problem. We describe the principle of residual learning to solve the problem that the convergence speed of deep convolution network is too slow. In this paper, we focus on the Faster-RCNN object detection algorithm, including feature extraction and region proposal generation.The performance of Faster-RCNN is analyzed to solve the defects of small-scale target detection.Aiming at the problem of small-scale target detection, this paper proposes a Multi-Level RCNN model. The image is processed by a single deep convolution neural network, and the multi-scale feature of network is aggregated. It is compressed into a unified feature space by max-pooling, bilinear interpolation and convolution, which is defined as the Multi-Level feature. The multi-level feature is used to extract the region proposal and identify and locate the object. The experimental results show that the multi-scale feature improves the object detection performance and effectively solves the problem of small-scale object detection.