Research And Implementation Of Generic Object Detection Algorithm Based On Deep Learning

Generic object detection has important research value in the fields of autonomous driving,robot navigation,video analysis,and image understanding,and has attracted much research attention in recent years.With the rapid development of deep learning,especially the deep neural networks,generic object detection has made remarkable progress.The detection accuracy of generic object detection based on deep learning has been greatly improved,which makes it possible to apply autonomous driving and robot navigation in real life.This article focuses on generic object detection based on deep learning,which aims to improve the detection accuracy of Faster R-CNN.Faster R-CNN is the mainstream framework for generic object detection methods.However,a single Faster R-CNN framework no longer has advantages compared with the latest detection models.So based on Faster R-CNN,this article proposes various improvements to ameliorate its detection results.This article mainly focuses on feature extraction,normalization methods,and anchor sizes to improve object detection results.The deep neural network is used for feature extraction.Due to large learning capacity and strong expressive power of the deep neural network,semantic,high-level,robust features can be learned.Meanwhile,feature pyramids architecture is adopted to address multi-scale problems.The feature pyramids architecture is used to produce a multi-scale feature representation,which enables the model to detect objects across a wide range of scales.Group normalization is used to address small batches.Group normalization divides the channels of input features into several groups and computes the mean and variance within each group for normalization.Therefore,its computation is independent of batch sizes,thus avoiding the problem of small batches.Moreover,k-means clustering algorithm is used finally to determine anchor sizes of the network on the purpose of making the network do bounding-box regression more easily.By improving Faster R-CNN in the above aspects,our model achieves 1.6 points higher Average Precision(AP)than Faster R-CNN on MS COCO2017.