Accurate Object Localization in Remote Sensing Images Based on Convolutional Neural Networks.

Abstract

In this paper, we focus on tackling the problem of automatic accurate localization of detected objects in high resolution remote sensing images. The two major problems for object localization in remote sensing images caused by the complex context information such images contain are achieving generalizability of the features used to describe objects and achieving accurate object locations. To address these challenges, we propose a new object localization framework, which can be divided into three processes: region proposal, classification, and accurate object localization process. First, a region proposal method is used to generate candidate regions with the aim of detecting all objects of interest within these images. Then, generic image features from a local image corresponding to each region proposal are extracted by a combination model of 2-D reduction convolutional neural networks (CNNs). Finally, to improve the location accuracy, we propose an unsupervised score based bounding box regression (USB-BBR) algorithm, combined with a nonmaximum suppression algorithm to optimize the bounding boxes of regions that detected as objects.

Existing System

Rotation Invariant Parts Based Model, Contour-Based Spatial Model, Spatial Sparse Coding Bag-Of-Words Model.

Proposed System

In this paper, we focus on accurate localization of detected objects rather than simple object detection. The feature extraction problem for object detection in remote sensing images using convolutional neural network (CNN) models is tackled. In this paper, we proposed an object localization framework based on CNN in remote sensing images. The framework uses the CNN models to extract object features and obtain classification results. In the first stage, we used a selective search method to generate the major part of the candidate object regions. In the second stage, we designed a dimension reduction model using trained models to initialize the network weights and then use it to extract features and classify the objects to different categories. We also tested a retrained model and a fine-tuned model. In the third stage, we proposed a new USB-BBR algorithm, as part of the accurate object localization process, to obtain better detection localization precision, and we used NMS to decrease the number of overlapped regions. The addition of the USB-BBR method can help to obtain an optimal bounding box for each group of classified regions.

Architecture

BLOCK DIAGRAM