Remote Sensing Image Retrieval Based On Convolutional Neural Network And Ant Colony Optimization Algorithm

With the rapid growth of remote sensing images,how to accurately and effectively retrieve images from large image databases has become a requirement for the management and utilization of remote sensing images.CBIR is a good method to solve the above problems and has become an important research direction of remote sensing applications.The aim of image retrieval is to find the relevant images or the same class with the query image.Remote sensing image retrieval is usually based on the visual information of images itself,which only considers the relationship between two images and ignores the internal relationship between different images.At the same time,the ability of feature extraction of CNN has been used to retrieve images in lots of works;however,the powerful classification ability of CNN is ignored by most researchers.To solve these problems,this paper proposes a method for remote sensing image retrieval that base on convolutional neural networks and Ant Colony Algorithm.Firstly,we establish the semantic network of retrieval images by the positive feedback of ACO,and use a pheromone matrix to save the image semantic relationship degree.Secondly,the semantic network of the retrieval images is improved through continuous iteration.Finally,in the process of the final image retrieval,we make full use of the semantic relationship of the retrieval images to improve the retrieval results.Besides,in the process of image retrieval,we use the classification probability of CNN and image-to-class distance to calculate the similarity between images for further improving retrieval performance.Experiments on two publicly available datasets of remote sensing,PatternNet and UCM_LandUse,are carried to verify the validation and promising of the proposed method.The results show that the proposed method is robust,and can improve the precision of remote sensing image retrieval.The main innovation of this method is as follows:(1)This paper describes the semantic relationship between images in the retrieval dataset by ACO and makes full use of the semantic relationship to improve the performance of remote sensing image retrieval.(2)The classification ability of CNN,the category information contained in the sample with labels,and the inherent category information of the query image are fully utilized to improve the remote sensing image retrieval performance.