Research On Small Target Detection In Remote Sensing Image Based On Deep Learning

With the continuous maturity of remote sensing technology,the acquisition of remote sensing images has become easy,and the emergence of a large number of remote sensing data has gradually expanded its application fields,whether in military aspects such as national defense or in civil aspects such as urban traffic planning and agricultural detection.Therefore,it is great significance to detect important targets in remote sensing images.However,the traditional target detection method is not good at accuracy,efficiency or versatility when it faces remote sensing images.In recent years,the performance of deep learning algorithm in the field of target detection has far exceeded the traditional detection algorithm,but in the target detection of remote sensing images,further improvement is needed to ensure the detection results.Based on this background,this paper studies the target detection algorithm of remote sensing image.The main work and innovations are as follows:1.The training of network model needs a lot of data,but the current open source data set can not meet the training requirements of this experiment.Therefore,this paper takes the DOTA data set of Wuhan University as the basis and intercepts some remote sensing images through Google Earth software,and forms the data set required by the experiment after manual annotation.In this paper,the number of samples of small targets is mainly increased in the data set,and the annotation format is transformed into the format required by this algorithm.Finally,the data set is divided into training set,verification set and test set.2.The target detection algorithm based on deep learning is improved.Through the research on yolov5 algorithm,it is found that is easy to miss detection in the detection of small targets in remote sensing images.In view of the shortcomings of yolov5 algorithm,the following optimization is made: adding a module combining spatial attention and channel attention mechanism,the meaningful part of the features extracted by the algorithm can be strengthened,and then the feature expression ability of anchor points can be enhanced by using the feature enhancement of boundary limit points,so as to optimize the target positioning part.Finally,through the comparative analysis of the detection results with multiple algorithms,the superiority of the improved algorithm is proved.3.Based on the improved yolov5 algorithm,this paper designs a remote sensing image target detection system,which can automatically complete the target detection and return the results after the user uploads the picture.After testing,it is proved that the system function is basically realized.