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.

Improve Online Boosting Algorithm From Self-Learning Cascade Classifier

Improving Part based Object Detection by Unsupervised, Online Boosting

LACBoost and FisherBoost: Optimally Building Cascade Classifiers

Incremental Learning of Boosted Face Detector

Optimally Training a Cascade Classifier

Hybrid Cascade of Active/lazy Boosting

Training Effective Node Classifiers for Cascade Classification

Z-Adaboost: Boosting 2-thresholded weak classifiers for object detection

Efficiently Learning a Detection Cascade with Sparse Eigenvectors

A Revised AdaBoost Algorithm——AD AdaBoost

On the Design of Cascades of Boosted Ensembles for Face Detection

On-line Fall Detection Via a Boosted Cascade of Hybrid Features.

Real-time face detection with self-adaptive cost-sensitive AdaBoost

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

Training a Multi-Exit Cascade with Linear Asymmetric Classification for Efficient Object Detection

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

Incremental Training of a Detector Using Online Sparse Eigendecomposition

Asymmetric Totally-Corrective Boosting for Real-Time Object Detection

Online coordinate boosting

Robust visual tracking via online semi-supervised co-boosting

Enhancing classification performance through multi-source online transfer learning algorithm with oversampling