Remote Sensing Image Retrieval Based On Convolutional Neural Network And Fuzzy Classification

Convolutional Neural Networks(CNN)have made great achievements in many fields such as image classification and target recognition.Some remote sensing image retrieval methods have used CNN features to retrieve images,but these methods only use the feature extraction ability of CNN,and ignore the powerful classification ability of CNN.At the same time,most of the current remote sensing image retrieval techniques find the required images by traversing and matching the entire image database.Such image retrieval methods have problems such as excessive search range,long retrieval time and low retrieval efficiency.Therefore,how to help users efficiently retrieve images that are similar or even the same as the query image categories from a large number of remote sensing images has become an urgent problem in the field of remote sensing image retrieval.Aiming at the problems in remote sensing image retrieval methods,this paper proposes two methods for remote sensing image retrieval research:(1)This paper proposes a remote sensing image retrieval method based on weighted distance,which assigns different weights to the search image according to the category probability of the query image,and combines the weight and the feature distances to improve retrieval accuracy.(2)Based on the first method,this paper proposes a fuzzy classification method,which classifies the classification confidence of the query image into three categories: high confidence,medium confidence and low confidence.Different search types of retrieval image are determined by different categories.The fuzzy classification method reducing the search range and the retrieval time,this method can greatly reduce the time complexity under the premise of ensuring the retrieval precision.The main innovations of this paper are as follows:1.The remote sensing image retrieval method based on weight distance makes full use of CNN's feature extraction ability and classification ability.It is a simple and effective remote sensing image retrieval method.2.In this paper,the method for determining the search range by using the classification confidence of the query image is proposed.By this method,the search range can be greatly reduced and the speed can be improved.