Road Object Detection And Recognition Based On Deep Learning In Remote Sensing Image

With the continuous innovation of remote sensing technology,a lot of high-resolution and rich content information remote sensing images have emerged.Remote sensing image detection and recognition has very important practical application value in both civil and military fields,and has gradually become the research focus of many scholars.Based on the research background of remote sensing image road object detection and recognition,aiming at the problems of low detection accuracy and slow detection speed in the current remote sensing image road object detection and recognition algorithm,this paper makes relevant research and improvement.The main work is as follows:The image preprocessing algorithm based on remote sensing image is analyzed.The image enhancement methods such as image smoothing,image sharpening and histogram equalization are mainly analyzed;the image denoising methods such as mean filtering,median filtering and Gaussian filtering are analyzed in image smoothing;the Sobel algorithm and high pass filtering algorithm in differential method are mainly analyzed in image sharpening;Finally,the effect of histogram equalization algorithm on remote sensing image processing is analyzed.the relevant algorithms are tested and analyzed.Traditional methods are used to extract and classify remote sensing images.In the feature extraction algorithm,the SURF algorithm,the LBP algorithm and the HOG algorithm are analyzed.In the feature classification algorithm,the SVM algorithm is mainly analyzed.Finally,the road objects in the remote sensing image are detected and recognized by using the combination of HOG feature algorithm and SVM classifier.This paper focuses on the related algorithms of deep learning in the detection and recognition of road objects.This paper mainly analyzes and studies the R-CNN series algorithm,SSD algorithm and YOLO series algorithm,and focuses on the Faster R-CNN algorithm based on the regional proposal and YOLOv3 algorithm using the idea of direct regression.In addition,according to the characteristics of remote sensing image data set,the algorithm of YOLOv3 is improved in three aspects: first,by adding multi-scale feature fusion,the algorithm can extract the features of the shallower layer information,thus making the image feature information obtained by the algorithm richer;second,by increasing the reliability of bbox location information Calculation and optimization of IOU loss,improvement of loss function;Thirdly,k-means clustering is applied to the remote sensing image data set to optimize the anchor value of the data set.At last,it compares and analyzes with other deep learning algorithms in detection accuracy,detection speed,detection effect of small targets,and verifies the effectiveness of the improved algorithm.According to the needs of the actual project,a remote sensing image road object detection system based on deep learning is designed.The system can achieve the task of road object detection and recognition in remote sensing image.