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.

A Class Incremental Learning Method with Weight Re-Initialization and Multiple Distillation for Image Classification

Class Incremental Learning with Deep Contrastive Learning and Attention Distillation

Maintaining Discrimination and Fairness in Class Incremental Learning

Class Incremental Learning Via Multi-hinge Distillation

Incremental Land Cover Classification Via Label Strategy and Adaptive Weights

Class incremental learning of remote sensing images based on class similarity distillation

Memory Efficient Class-Incremental Learning for Image Classification

Class-Incremental Learning: Survey and Performance Evaluation on Image Classification

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

Class Incremental Learning with Multi-Teacher Distillation

Discrimination Correction and Balance for Class-Incremental Learning

Large Scale Incremental Learning

Discriminative Distillation to Reduce Class Confusion in Continual Learning

An integrated classification model for incremental learning

Enhancing Class-Incremental Learning for Image Classification Via Bidirectional Transport and Selective Momentum

Class-Incremental Learning: A Survey

Class-Incremental Learning using Diffusion Model for Distillation and Replay

Reducing Catastrophic Forgetting in Online Class Incremental Learning Using Self-Distillation

Rethinking Class-Incremental Learning from a Dynamic Imbalanced Learning Perspective

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

Reformulating Classification As Image-Class Matching for Class Incremental Learning