Abstract
Recently, methods based on Faster region-based convolutional neural network (R-CNN) have been popular in multi-class object detection in remote sensing images due to their outstanding detection performance. The methods generally propose candidate region of interests (ROIs) through a region propose network (RPN), and the regions with high enough intersection-over-union (IoU) values against ground truth are treated as positive samples for training. In this paper, we find that the detection result of such methods is sensitive to the adaption of different IoU thresholds. Specially, detection performance of small objects is poor when choosing a normal higher threshold, while a lower threshold will result in poor location accuracy caused by a large quantity of false positives. To address the above issues, we propose a novel IoU-Adaptive Deformable R-CNN framework for multi-class object detection. Specially, by analyzing the different roles that IoU can play in different parts of the network, we propose an IoU-guided detection framework to reduce the loss of small object information during training. Besides, the IoU-based weighted loss is designed, which can learn the IoU information of positive ROIs to improve the detection accuracy effectively. Finally, the class aspect ratio constrained non-maximum suppression (CARC-NMS) is proposed, which further improves the precision of the results. Extensive experiments validate the effectiveness of our approach and we achieve state-of-the-art detection performance on the DOTA dataset.
Highlights
Multi-class object detection is one of the main tasks in automatic analysis of remote sensing (RS) images, it is indispensable in many applications such as urban management, traffic monitoring, search and rescue missions, military uses [1,2] and so on
To evaluate the proposed IoU-Adaptive Deformable region-based convolutional neural network (R-convolutional neural network (CNN)) quantitatively, we provide ablation experiments on the validation set of the DOTA dataset we compare our method to the ones mentioned in Xia et al [3] work for horizontal bounding boxes (HBB) prediction task as well as Seyed Majid Azimi et al [23] based on the test set whose ground-truth labels are undisclosed
The average precision (AP) values of small objects like ships, small vehicles and large vehicles increase more than other objects, which illustrate the favorable performance of our methods for small object detection
Summary
Multi-class object detection is one of the main tasks in automatic analysis of remote sensing (RS) images, it is indispensable in many applications such as urban management, traffic monitoring, search and rescue missions, military uses [1,2] and so on. With the rapid development of RS technology, a large number of high quality satellite and aerial images can be obtained more In this case, more complex diversity changes in scale, direction and shape make automatic multi-class object detection in remote sensing images more challenging, which has attracted more and more attention at the same time. In order to obtain more accurate detection results, the region-based detection algorithm has become the mainstream method for solving multi-class object detection problems [5,6,7,8] All these algorithms are modified for different problems based on the Faster R-CNN [9] detection framework, in which proposals generated by the RPN are sent to the R-CNN network for classification and regression. We can not achieve a satisfactory detection performance when we use the Faster R-CNN directly for multi-class detection in remote sensing images
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have