The Research And Implementation Of Airspace Image Object Detection Based On Deep Learning

Recently,satellite photography,aerial photography and remote sensing technology have matured.We can acquire high-definition and stable ground-based visual images with tools such as satellites,aircraft and drones.However,due to the difficulty of detecting object in the air-space image and the information in the airspace image has great digging value,so the high-definition airspace image object detection technology has a very high research value.In computer vision,the technology of object detection in the image has been studied for many years and has achieved remarkable results.Recently,with the rapid development of deep learning and the continuous improvement of convolutional neural networks,the development of computer vision has reached a new peak.At the same time,as one of the important applications in the field of computer vision,Object detection has always been the focus of research.Convolutional neural network has also made great progress in object detection.Firstly,we investigate the field of high-definition airspace image object detection and related technologies.Then,aiming at the task of air-space image object detection,object detection task is done using the three popular network frameworks:R-CNN,Fast R-CNN and Faster R-CNN,and achieve a better effect of detection and identification,and the exper-imental results and data as a standard compared with the subsequent ex-periments.Aiming at a series of difficulties in airspace object detection,such as too wide variation of object angle,high resolution of most images,too high detection speed and low resolution of image,change of shape and image background.In this paper,we use a full convolution network(R-FCN)to construct a spatial image object detection framework and op-timize it to overcome the difficulties in the object detection of airspace.Finally,based on the full convolutional network,this paper realizes the most advanced high-precision detection in multi-category object de-tection tasks by adapting and combining the latest innovative technolo-gies,minimizing the computational cost and obtaining PVANet,then we construct airspace image object detection framework based on PVANet.And we optimize the PVANet in terms of "universal" based on the Uber-Net.Experiments show that the optimized and improved PVANet is better than the full convolutional network framework in the detection and recognition of airspace image object detection tasks,the effect and speed have been further improved.