Object Detection And Segmentation Based On Deep Learning In Complex Background

Object detection and segmentation in complex background is one of the core issues in the field of computer vision,and its main task is to identify and locate target objects in the image.Especially in the field of aerospace,it is of great strategic significance in the actual combat process for the accurate identification and positioning of military targets.Under complex background,the object is easily affected by insufficient illumination,occlusion,background interference and other factors,making it difficult to detect and has low robustness.Therefore,this paper designs a Convolutional Neural Network(CNN)structure in the view of deep learning to detect and semantically segment the targets,which is based on the classical target detection and segmentation algorithms proposed at present and combined with the actual situation in complex background.In this paper,has done the following works:firstly,establish an Internet Image Database that satisfies the complex background conditions.And according to the classical object detection method—the Faster R-CNN principle,the paper improves the region generation network,extracts a more detailed local feature for the image objects in complex background,and designs a new location selection network structure,which improves the detection accuracy of location.Then,on the basis of the target detection network,two different methods are adopted to semantically segment the target in complex background:full convolutional neural network and mask segmentation,which achieves end-to-end,pixel-to-pixel training and learns the deeper features of the image.Through experimental verification,this method achieves higher detection accuracy on the Internet image database.Tank model image is used to test the single target,multi-target,different degrees of occlusion and camouflage in complex background and two different segmentation results are shown,which concludes that this method has certain robustness and practicality.