Few-shot class-incremental learning (FSCIL) presents a greater challenge compared with few-shot task-incremental learning (FSTIL) due to the need to classify all previous classes without prior knowledge of the session identifier (session-ID). To address this, we propose Bamboo, a novel framework for FSCIL that introduces a cascading inference mechanism to explicitly infer the session-ID for each sample. This mechanism is enabled by a novel, session-specific equiangular tight frame prototype (ETF-P) classifier. By adaptively fusing session-agnostic and session-specific semantics, the ETF-P classifier reliably determines if a sample belongs to its associated session, which is the core decision required at each step of the cascade. Considering the incremental nature of the learning process, which resembles the continuous growth of bamboo, we treat the base session classifier as the foundational bamboo node and progressively add new session classifiers as additional nodes on top. During the testing phase, each sample flows sequentially through the bamboo nodes, from top to bottom, to determine its session-ID and to be classified accordingly. Overall, the Bamboo framework is capable of perceiving session-ID without prior knowledge and classifying each sample within the correct session, leading to state-of-the-art performance on multiple benchmark datasets.

Few-shot class-incremental learning (FSCIL) requires learning new knowledge without forgetting old knowledge. Forward compatibility can reserve space for novel classes while maintaining base class knowledge in incremental learning. Better forward compatibility is crucial for effectively mastering all knowledge, especially when dealing with a few unknown new classes. In this article, we propose the better forward compatibility pretraining (BFCP) to further enhance forward compatibility in FSCIL. We adopt a two-stage training for the backbone network in the base session. First, we train the backbone network at the image-level to enhance its feature extraction capability, enabling the model to extract valuable information from unknown class images. Second, we fine-tune the backbone network at the feature-level with fake prototypes and instances to achieve clustering base classes and reserve space for unknown new classes. For all incremental new sessions, we freeze the backbone network and employ prototype rectification without further training to refine the prototypes of the novel classes. We conduct extensive experiments with different input scales, including federated cross-domain pretraining and cross-domain class-incremental experiments. BFCP efficiently handles both novel and base classes of each incremental session and significantly outperforms state-of-the-art methods, achieving an average accuracy of 63.47% on the CIFAR100 dataset.

Task-incremental learning methods that adopt knowledge distillation face two significant challenges: confidence bias and knowledge loss. These challenges make it difficult to effectively balance the stability and plasticity of the network in the incremental learning process. In this article, we propose double confidence calibration focused distillation (DCCFD) to address these challenges. We introduce intratask and intertask confidence calibration (ECC) modules that can mitigate network overconfidence during incremental learning and reduce the degree of feature representation bias. We also propose a focused distillation (FD) module that can alleviate the problem of knowledge loss during the task increment process, improving model stability without reducing plasticity. Experimental results on the CIFAR-100, TinyImageNet, and CORE-50 datasets demonstrate the effectiveness of our method, with performance that matches or exceeds the state of the art. Furthermore, our method can be used as a plug-and-play module to consistently improve class-incremental learning methods.

Infrared dim small target (IDST) detection is of great significance in security surveillance and disaster relief. However, the complex background interference and tiny targets in infrared images keep it still a long-term challenge. Existing deep learning models stack network layers to expand the model fitting capability, but this operation also increases redundant features which reduce model speed and accuracy. Meanwhile, small targets are more susceptible to positional bias, with this dramatically reducing the model’s localization accuracy. In this article, we propose a guided attention and joint loss (GA-JL) network for infrared small target detection. More specifically, the method visualizes the feature maps at each resolution through a two-branch detection head (TDH) module, filters out the features that are strongly related to the task, and cuts out the redundant features. On this basis, the guided attention (GA) module guides the prediction layer features using the features that are associated closely with the task and combines spatial and channel bidirectional attention to make the prediction layer feature maps embedded with effective messages. Finally, through the joint loss (JL) module, the target position regression is performed with multiangle metrics for enhancing the target detection accuracy. Experimental results of our method on the SIATD, SIRST, and IRSTD_1k datasets reveal that it is capable of accurately identifying IDSTs, remarkably reduces the false alarm rate, and outperforms other methods.

Infrared Dim Small Target (IDST) detection holds significant importance in early target warning and ground monitoring. However, IDST detection remains a long-standing challenge due to the low signal-to-noise ratio and low contrast. Feature fusion is an effective approach for feature enrichment and improving poor performance in IDST detection. Existing feature fusion methods tend to overlook the importance of focusing on both multi-layer and single-layer features, which prevents target features from being fully exploited, resulting in suboptimal outcomes for IDST detection. In this paper, we present a Multi-Angle Pyramid Feature Fusion Network (MAPFF), which selects fusion objects from multiple perspectives and then fuses them. Namely, multilayer features and single-layer features - two perspectives of the fusion object - are selected and fused separately. The MAPFF network consists of two primary modules: a Cross-Layer Complementary Feature (CLCF) module and an Atrous Spatial Pyramid Pooling with Attention (AttnASPP) module. To effectively fuse semantic and geometric detail information, the CLCF module adaptively combines different layer features as complementary features, while the original layer features serve as the main features. Concurrently, through channel shuffle, the complementary and main features achieve substantial information exchange. The AttnASPP module employs parallel atrous convolutions with multiple dilation rates to obtain multi-scale information and incorporates an attention mechanism to emphasize effective features. Experimental results on the SIATD, SIRST and IRSTD_1k datasets demonstrate that our method can precisely identify IDSTs, significantly reduce the false alarm rate, and outperform other methods.

Federated Continual Learning (FCL) approaches exist two major problems of the probability bias and the imbalance in parameter variations. These two problems lead to catastrophic forgetting of the network in the FCL process . Therefore, this paper proposes a novel FCL framework, Federated Probability Memory Recall (FedPMR), to mitigate the probability bias problem and the imbalance in parameter variations. Firstly, for the probability bias problem, this paper designs the Probability Distribution Alignment (PDA) module, which consolidates the memory of old probability experience. Specifically, PDA maintains a replay buffer and uses the probability memory stored in the buffer to correct the offset probabilities of the previous tasks during the two-stage training. Secondly, to alleviate the imbalance in parameter variations, this paper designs the Parameter Consistency Constraint (PCC) module, which constrains the magnitude of neural weight changes for previous tasks. Concretely, PCC applies a set of adaptive weights to subsets of the regularization term that constrains parameter changes, forcing the current model to be sufficiently close to the past model in parameter space distance. Experiments with various levels of task similitude across clients demonstrate that our technique establishes the new state-of-the-art performance when compared to previous FCL approaches.

In infrared scenes, humans can easily observe objects in the scene with their eyes, even dim ones. To make the robot have the same visual ability, this paper proposes a pyramid-feature fusion target detection network, called RLPGB-Net, which combines reinforcement learning with aerial targets in the infrared scene. It makes use of the powerful decision-making ability of reinforcement learning to give corresponding weights to the extracted features and highlight the significant features of infrared dim small targets. In reinforcement learning, we use priori strategy guidance and long-term training methods to train weight-regulating agents. To eliminate the local influence on the detection results, such as bright interference points similar to the target, and to solve the problem of dim target detection effectively, the global context boundary attention module is introduced to eliminate the disadvantage of local comparison by using the global characteristics of different dimensions. At the same time, it can prevent the edge information of the refined target from being submerged in the background. Experimental results on SAITD and SIRST data sets show the effectiveness of the proposed method.