Research And Implementation Of Remote Sensing Object Detection Based On Deep Learning

Object detection is an important task in the field of computer vision and image processing.It is widely used in the fields of identity recognition,quantity statistics,location determination and so on.Due to the development of deep learning and the growth of computing resources,neural network technology has replaced most of the traditional image algorithms and made great progress in the field of object detection.Remote sensing image refers to the generation of images through long-distance detection.The study of remote sensing object detection is of far-reaching significance for resource assessment,environmental exploration and other tasks.However,the difference between remote sensing image and ordinary image leads to the network model used in the field of object detection is often difficult to be directly applied to remote sensing object detection.This paper mainly studies and improves the algorithm of object detection in the field of remote sensing,and applies it to the fields with similar characteristics of remote sensing images.(1)This paper proposes a new remote sensing object detection model EFM-Net,which includes multi branch feature extraction module to obtain more semantic information through different branches.The background filtering module is used to weaken the background information and enhance the network’s attention to objects.The occlusion improvement module generates masks to form occlusion,and enhances the features through the boundary eigenvalues,so as to improve the robustness of the network to occluded objects.Network model is tested on the the data set DOTA v1.0,NWPU VHR-10,UCAS-AOD.In DOTA v1.0,the detection accuracy of the two bounding boxes has reached 75.48%and 76.27%,which is improved by 4.26%and 3.34%respectively compared with the basic model.The detection accuracy on NWPU VHR-10 and UCAS-AOD reached 92.10%and 96.45%respectively.(2)This paper proposes an improved cloud removal algorithm,which combines the self attention algorithm with the atmospheric scattering physical model,and integrates the cloud removal module into the object detection network.We are in the data set DOTA v1.0,three cloud effects with different concentrations are generated through the cloud generation algorithm,and on this basis,the ablation experiment and comparison experiment are completed.Under the heavy cloud effect,the detection accuracy of the network model on the two boundary boxes is 41.86%and 44.22%,which is 2.78%and 2.36%higher than that of the improved AOD net,respectively.(3)Aiming at the practical application of the algorithm model,we introduce the specific application project of intelligent mine products.Then,the functions of coal mine brain and the application process of algorithm model and business model are described as a whole.The model can judge the specific task according to the output of the camera,output the alarm information and record it on the output log to remind the staff.The algorithm proposed in this paper has better detection effect,so as to reduce the occurrence of false positives.