Abstract:With the rapid development of remote sensing technology and the growing demand for applications, the classical deep learning-based object detection model is bottlenecked in processing incremental data, especially in the increasing classes of detected objects. It requires models to sequentially learn new classes of objects based on the current model, while preserving old categories-related knowledge. Existing class-incremental detection methods achieve this goal mainly by constraining the optimization trajectory in the feature of output space. However, these works neglect the case where the previously learned background is a new category to learn, resulting in performance degradation in the new category because of the conflict between remaining the background-related knowledge or updating the background-related knowledge. This paper proposes a novel class-incremental detection method incorporated with the teacher-student architecture and the selective distillation (SDCID) strategy. Specifically, it is the asymmetry architecture, i.e., the teacher network temporarily stores historical knowledge of previously learned objects, and the student network integrates historical knowledge from the teacher network with the newly learned object-related knowledge, respectively. This asymmetry architecture reveals the significance of the distinct representation of history knowledge and new knowledge in incremental detection. Furthermore, SDCID selectively masks the shared subobject of new images to learn and previously learned background, while learning new categories of images and then transfers the classification results of the student model to the background class following the judgment model of the teacher model. This manner avoids interferences between the background category-related knowledge from a teacher model and the learning of other new classes of objects. In addition, we proposed a new incremental learning evaluation metric, C-SP, to comprehensively evaluate the incremental learning stability and plasticity performance. We verified the proposed method on two object detection datasets of remote sensing images, i.e., DIOR and DOTA. The experience results in accuracy and C-SP suggest that the proposed method surpasses existing class-incremental detection methods. We further analyzed the influence of the mask component in our method and the hyper-parameters sensitive to our method.

Incremental Classification of Remote Sensing Images with Feature Pyramid and Class Hierarchy

Incremental Learning of Remote Sensing Target Classification with Class Hierarchy

Incremental Detection of Remote Sensing Objects with Feature Pyramid and Knowledge Distillation

Semantic Guided Level-Category Hybrid Prediction Network for Hierarchical Image Classification.

Fuzzy Rough Sets-Based Incremental Feature Selection for Hierarchical Classification

A Class-Incremental Detection Method of Remote Sensing Images Based on Selective Distillation

Uncertainty-Aware Hierarchical Refinement for Incremental Implicitly-Refined Classification

Class-Incremental Recognition of Objects in Remote Sensing Images with Dynamic Hybrid Exemplar Selection

Distillation-Constrained Prototype Representation Network for Hyperspectral Image Incremental Classification

InfRS: Incremental Few-Shot Object Detection in Remote Sensing Images

Incremental Scene Classification Using Dual Knowledge Distillation and Classifier Discrepancy on Natural and Remote Sensing Images

Superpixel-Based Long-Range Dependent Network for High-Resolution Remote-Sensing Image Classification

Incremental feature selection for large-scale hierarchical classification with the arrival of new samples

Boosting Hyperspectral Image Classification with Dual Hierarchical Learning

Class-Incremental Learning for Remote Sensing Images Based on Knowledge Distillation

Hybrid Rotation Self-Supervision and Feature Space Normalization for Class Incremental Learning

Hierarchical Deep Features Progressive Aggregation for Remote Sensing Images Scene Classification

Incremental Learning With Open-Set Recognition for Remote Sensing Image Scene Classification

Learning Multiscale and Deep Representations for Classifying Remotely Sensed Imagery

Class-Incremental Novel Category Discovery in Remote Sensing Image Scene Classification via Contrastive Learning