Typical Object Detection Of Remote Sensing Images Based On Deep Learning

With the continuous development of remote sensing technology and computer vision technology,remote sensing image target detection plays an important role in many fields such as military and civilian use.Traditional target detection algorithms need to manually design features that are often interfered by many factors,affecting the performance of the target detection algorithm.As scholars deepen their research on deep learning,deep learning has become a powerful tool for research and exploration in the field of target detection.In the field of target detection,in the context of big data applications and the rapid improvement of computer hardware performance,deep learning has obvious advantages over traditional visual detection algorithms.However,the more complex the design of the deep learning model,the more parameters involved and the greater the amount of calculations,which results in the algorithm being unable to run in real time.Therefore,the target detection technology based on deep learning still has many difficulties in people's daily life applications.In view of the complex remote sensing image target and the advantages of the target detection algorithm based on deep learning,this paper improves the detection algorithm of aircraft targets based on the Bag-of-words model and proposes a multi-scale end-to-end deep learning target detection algorithm.The specific work content is as follows:(1)For massive remote sensing image data,this paper sets up a typical target sample database of multi-source remote sensing images for remote sensing image aircraft targets,and manually labels the sample data set as the standard sample training test set for this experiment.(2)In order to solve the complex changes in target morphology of remotely sensed images and the obvious background interference.According to targets in remote sensing images,features are extracted to construct a bag-of-words model.Based on that bag-of-words model,an aircraft targets detection algorithm of Hough voting is proposed.The accuracy of this algorithm has reached more than 70%.(3)For the current mainstream two-step deep learning target detection framework,this paper combines the positioning algorithm with the deep learning framework and adds a multi-scale feature extraction pyramid after the last layer of the convolutional layer of the network model.Based on which a multi-scale end-to-end deep learning target detection framework is designed.The basic network uses an improved VGG-Net network model.The objective loss function based on multi-scale end-to-end deep learning target detection is improved.The Focal loss function replaces the cross-entropy loss function,and the positive and negative sample proportions are balanced,which improves the accuracy of target detection.In the sample data set produced by this paper,the detection accuracy of the algorithm reached 89%,the missed detection rate and false detection rate were controlled at a relatively low level,and the operating speed was greatly improved.Experiments show that the deep learning target detection algorithm can effectively improve the target detection performance.