Research On Object Detection Of High Resolution Images Based On Convolution Neural Network

To solve the low object detection accuracy due to the large-scale remote sensing image with small object spatial rang,multi-scale target,this paper developed a object detection framework(Deep-YOLOv3)suitable for high resolution images.Firstly,it divided the large-scale remote sensing image into 416*416 image slices.Secondly,it extracted the features by the convolutional neural network,and obtained 5 scales feature map.Thirdly,divided the image slice into the 5 scale,and then used the grid cell including ground true to predicting the object.Finally,it used the NMS to suppress the bounding boxes in order to the final predicted results.This paper is solved the low detection accuracy due to the complexity of the actual scene,the variable shape and structures of the object in order to improve the object detection accuracy of high-resolution remote sensing image.And it used the multi-scale detection strategy to solve multi-scale object detection.The experimental results show that based on the original algorithm the Deep-YOLOv3 model effectively improves the object detection accuracy of remote sensing image,and it provides an effective solution for object detection of high resolution remote sensing images.There are 22 figures,5 tables and 49 references in this paper...