Fine-grained leaf image retrieval (FGLIR) is an emerging content-based image retrieval (CBIR) task that distinguishes cultivars of the same plant species. It remains highly challenging due to subtle inter-cultivar variations and the absence of supervision. Existing training-free methods generally treat convolutional features as local descriptors for aggregation, but they often overlook spatial correlations and structural cues. To address this gap, we extend Local Binary Patterns (LBP) from the raw image domain to the deep feature domain for the first time, enabling explicit modeling of spatial relationships. Specifically, each leaf image is partitioned into R-Angle patches to preserve invariant traits, which are processed by pre-trained CNNs to produce deep feature maps. We then propose the Local Deep-Binary Pattern Transform (LDbPT) and Local Deep Contrast-Pattern Transform (LDcPT) to enhance spatial and contrast encoding, together with anR-Angle Grouped Average Pooling strategy to embed shape information into feature aggregation. Furthermore, we hybridize CNNs with Vision Transformers (ViTs) within this framework, effectively combining local structural extraction with global contextual reasoning. Extensive experiments on two public FGLIR benchmarks with comprehensive evaluations show that our method achieves state-of-the-art retrieval accuracy while demonstrating robustness, efficiency, and strong generalization, confirming the effectiveness of the proposed training-free hybrid representation.

Hybrid precoding can support high data rate with low cost for millimeter wave (mmWave) multi-antenna systems. To achieve near-optimal performance with low computing latency and enable end-to-end learning, deep learning has been introduced for optimizing hybrid precoding. Most research efforts focus on learning hybrid precoding under static channels. In mobile communications, however, the channel aging effect incurs severe performance degradation of multi-antenna systems. In this paper, we resort to a proactive optimization method to learn the analog and digital precoders for multiple users in time-varying mmWave channels with implicit channel prediction. We consider the practical frame structure used in prevalent cellular systems, and propose a method to learn the hybrid precoding polices for multiple downlink subframes in parallel. Simulation results demonstrate that the proposed method performs closely to the hybrid precoding that assuming perfect future channel information and outperforms existing methods.

Physical unclonable functions (PUFs) are promising solutions for low-cost device authentication; hence, ignoring the security of PUFs is becoming increasingly difficult. Generally, strong PUFs are vulnerable to classical machine learning (ML) attacks; however, classical ML attacks do not perform well on strong PUFs with complex structures. Side-channel analysis (SCA) hybrid attacks provide efficient approaches to modeling XOR APUF. However, owing to the inadequate exploitation of all available data, recent SCA hybrid attacks may fail on novel PUF designs, such as MPUF and iPUF. Thus, herein, we introduce a method that combines challenge-response pairs with side-channel information to construct challenge-synthetic-feature pairs (CSPs) via feature cross, thereby making it possible to model strong PUFs through multiclass classification. We propose multiclass classification-based SCA hybrid attacks to model strong PUFs with complex structures. When provided with CSPs, the proposed hybrid attacks use a feed-forward neural network with a softmax activation function to build combined models of PUFs. The combined models predict class labels for given challenges and then reveal responses through simple mappings from these labels. Experimental results show that the proposed attacks could model 16-XOR APUF, (128,5)-MPUF, (8,8)-iPUF, and (2,16)-iPUF with accuracies exceeding 94%. Compared with state-of-the-art modeling techniques, the proposed attack has advantages in terms of modeling accuracy, time cost, and the size of required training data.