Object Detection In Remote Sensing Image Based On Deep Learning

Remote sensing image detection plays an important role in both military and civilian fields,and can be applied to many fields such as marine monitoring,urban planning,and weather monitoring in my country.In recent years,deep neural networks have been successfully applied in the field of computational vision,and have demonstrated superior performance in the fields of target detection,segmentation,tracking,and recognition.With the rapid development of my country’s aerospace technology and remote sensing technology,the quantity and quality of remote sensing image data have been improved,providing strong support for the establishment of deep learning-based target detection methods.The work of this paper is based on the research of remote sensing image target detection algorithm based on deep learning,and conducts indepth research on the characteristics of large image size,small target size and complex background information of remote sensing images.The contributions of this article are summarized as follows:(1)Aiming at the problem of high resolution of remote sensing images and small size of detected targets,a remote sensing image target detection method based on multi-scale convolution fusion of holes is proposed.First,use the principle of hole convolution to design a parallel multi-scale hole convolution module with shared parameters,and add it to the residual network to adapt to different scale targets,to further obtain richer features,while maintaining the last one of the residual network The resolution of the stage output feature map remains unchanged.Secondly,strengthen the exchange of information at the front and back layers through the fusion of deep and shallow features.Finally,the prediction network consists of two stages of classification and regression cascade,and the features are matched before the second prediction to solve the problem of misalignment of the secondary prediction features.Experimental results show that the method can effectively improve the detection accuracy of small targets by introducing a multi-scale cavity convolution module.(2)Aiming at the problem of complicated remote sensing image background information and interference to target detection,a polar coordinate target detection algorithm based on a cross-channel supervised spatial attention mechanism is proposed.Firstly,the algorithm constructs an attention mechanism from the two dimensions of channel and space,so as to improve the information of the target area and weaken the background interference.In order to guide the training of attention module,this paper constructs loss function information to supervise the generation of attention map.Secondly,in view of the complex regression problem in the representation method of the rotating frame in the Cartesian coordinate system,this article refers to the target detection method based on the polar coordinate system,and completes the target detection task by predicting the target pole,polar diameter and polar angle.The experimental results show that by introducing the attention mechanism,the network can effectively weaken the interference background information,thereby achieving superior detection performance.