Target Detection In Remote Sensing Images Based On Faster R-CNN

With the rapid development of satellite remote sensing image(RSI)technology,how to quickly extract the required information from RSI and accurately detect specific targets has become a hotspot of current research.Traditional target detection methods usually per-form feature extraction manually to train the classifier.Extracting representative features and distinguishing features of targets is a key factor to improve detection accuracy.Faster Region-Convolutional Neural Networks,Faster R-CNN,is a deep learning model with large data processing capabilities.It can automatically learn features from images by establishing a layered structure similar to the human brain.In recent years,Faster R-CNN has been suc-cessfully applied to object detection,speech recognition and image classification due to its good learning capacity.Therefore,this thesis focuses on target detection in remote sensing images based on Faster R-CNN,and the main contents are as follows.This thesis presents a target detection algorithm in remote sensing images based on Faster R-CNN,which is constructed using VGG16 convolutional neural network and Caffe deep learning framework.First,the algorithm selects GaoFen-2 optical remote sensing im-ages with a resolution of 1 meter,and uses support vector machine to preprocess the images,dividing the larger detection area into smaller regions of interest(ROI)that may contain tar-gets.The algorithm then applies an R-CNN-based target detection algorithm to ROI images.To improve the detection result of small and gathering targets,we adopt an effective target detection framework,Faster-RCNN,and improve the structure of its original convolutional neural network,VGG16,by using multiresolution convolutional features and performing ROI pooling on a larger feature map in a region proposal network.Finally,experiments were designed to use R-CNN and Faster R-CNN to detect targets in remote sensing images.The experimental results show that the deep learning method based on Faster R-CNN can identify selected targets more quickly and accurately.This thesis also conducts comparative experiments on remote sensing images of other targets such as oil tank,playground,and overpass objects.In the same experimental environment,good experimental results were obtained,the target detection rate was high,and the detection time of each picture was less than 0.2 seconds,which fully verified the effectiveness and reliability of the algorithm.In summary,this thesis presents a remote sensing image target detection algorithm based on Faster R-CNN.Experiments show that this algorithm can achieve rapid and accu-rate detection of aircraft and other selected objects.The algorithm is of promotion signifi-cance in image target detection application and is of certain reference significance for target detection research based on other deep learning models.