Abstract:Deep-learning-based object detectors have substantially improved state-of-the-art object detection in remote sensing images in terms of precision and degree of automation. Nevertheless, the large variation of the object scales makes it difficult to achieve high-quality detection across multiresolution remote sensing images, where the quality is defined by the Intersection over Union (IoU) threshold used in training. In addition, the imbalance between the positive and negative samples across multiresolution images worsens the detection precision. Recently, it was found that a Cascade region-based convolutional neural network (R-CNN) can potentially achieve a higher quality of detection by introducing a cascaded three-stage structure using progressively improved IoU thresholds. However, the performance of Cascade R-CNN degraded when the fourth stage was added. We investigated the cause and found that the mismatch between the ROI features and the classifier could be responsible for the degradation of performance. Herein, we propose a Cascade R-CNN++ structure to address this issue and extend the three-stage architecture to multiple stages for general use. Specifically, for cascaded classification, we propose a new ensemble strategy for the classifier and region of interest (RoI) features to improve classification accuracy at inference. In localization, we modified the loss function of the bounding box regressor to obtain higher sensitivity around zero. Experiments on the DOTA dataset demonstrated that Cascade R-CNN++ outperforms Cascade R-CNN in terms of precision and detection quality. We conducted further analysis on multiresolution remote sensing images to verify model transferability across different object scales.

Learning Cascaded Shared-Boost Classifiers for Part-Based Object Detection

Object detection via fusion of global classifier and part-based classifier

Adaptive Object Detection by Implicit Sub-Class Sharing Features

Multiple Instance Ffeature for Robust Part-Based Object Detection

Multi-view multi-pose object detection using splited cascade model

Object detection by combined model based on cascaded adaboost

Boosting 2-Thresholded Weak Classifiers over Scattered Rectangle Features for Object Detection.

Cascade Object Detection with Complementary Features and Algorithms

Cascaded One-Vs-rest Detection Network for Fine-Grained Recognition Without Part Annotations

Object detection via structural feature selection and shape model

Cascading Classifier with Discriminative Multi-Features for a Specific 3D Object Real-Time Detection.

Combining fast extracted edge descriptors and feature sharing for rapid object detection

Multi-Scale Shared Features for Cascade Object Detection

Improving Part based Object Detection by Unsupervised, Online Boosting

Cascade object detection with deformable part models

Object Detection with Discriminatively Trained Part Based Models

High Quality Object Detection for Multiresolution Remote Sensing Imagery Using Cascaded Multi-Stage Detectors.

Discriminative Hough-Voting For Object Detection With Parts

Cascade one-vs-rest detection network for fine-grained recognition without part annotations

Discriminative Semantic Parts Learning For Object Detection

Contour Based Object Detection Using Part Bundles