RS-SSKD: Self-Supervision Equipped with Knowledge Distillation for Few-Shot Remote Sensing Scene Classification

Abstract

While growing instruments generate more and more airborne or satellite images, the bottleneck in remote sensing (RS) scene classification has shifted from data limits toward a lack of ground truth samples. There are still many challenges when we are facing unknown environments, especially those with insufficient training data. Few-shot classification offers a different picture under the umbrella of meta-learning: digging rich knowledge from a few data are possible. In this work, we propose a method named RS-SSKD for few-shot RS scene classification from a perspective of generating powerful representation for the downstream meta-learner. Firstly, we propose a novel two-branch network that takes three pairs of original-transformed images as inputs and incorporates Class Activation Maps (CAMs) to drive the network mining, the most relevant category-specific region. This strategy ensures that the network generates discriminative embeddings. Secondly, we set a round of self-knowledge distillation to prevent overfitting and boost the performance. Our experiments show that the proposed method surpasses current state-of-the-art approaches on two challenging RS scene datasets: NWPU-RESISC45 and RSD46-WHU. Finally, we conduct various ablation experiments to investigate the effect of each component of the proposed method and analyze the training time of state-of-the-art methods and ours.

Enhancing Few-Shot Learning in Lightweight Models via Dual-Faceted Knowledge Distillation

“…Recent studies have shown that self-supervised learning effectively improves the generalization performance of transfer-learning models in FSC tasks [ 34 ]. Based on this, in this paper, we use a linear combination of cross-entropy and self-supervised loss function to pre-train the teacher network model and select rotation prediction as the self-supervised learning task.…”

Surface Defect Segmentation Algorithm of Steel Plate Based on Geometric Median Filter Pruning

“…Knowledge distillation from the heavy network as the teacher to the light network as the student can be used as a way to improve the performance of the student network. Zhang et al (2021) proposed a novel two-branch network that took three pairs of original transformed images as input and incorporated a class activation graph to drive the network to mine the most relevant class-specific regions. This strategy ensured that the network generated differentiated embeddings and a round of self-knowledge distillation was set up to prevent overfitting and improve performance.…”

A Hierarchical Prototypical Network for Few-Shot Remote Sensing Scene Classification

“…Following recent FSRSSC studies [10,[22][23][24], we utilize ResNet-12 as a backbone for feature extraction. We also adopt the pre-training strategy as suggested in [22] to better initialize the meta-learner feature extractor.…”