The exponential growth of multi-modal data in the real world poses significant challenges to efficient retrieval, and traditional single-modal methods are no longer suitable for the growth of multi-modal data. To address this issue, hashing retrieval methods play an important role in cross-modal retrieval tasks when referring to a large amount of multi-modal data. However, effectively embedding multi-modal data into a common low-dimensional Hamming space remains challenging. A critical issue is that feature redundancies in existing methods lead to suboptimal hash codes, severely degrading retrieval performance; yet, selecting optimal features remains an open problem in deep cross-modal hashing. In this paper, we propose an end-to-end approach, named Robust Supervised Deep Discrete Hashing (RSDDH), which can accomplish feature learning and hashing learning simultaneously. RSDDH has a hybrid deep architecture consisting of a convolutional neural network and a multilayer perceptron adaptively learning modality-specific representations. Moreover, it utilizes a non-redundant feature selection strategy to select optimal features for generating discriminative hash codes. Furthermore, it employs a direct discrete hashing scheme (SVDDH) to solve the binary constraint optimization problem without relaxation, fully preserving the intrinsic properties of hash codes. Additionally, RSDDH employs inter-modal and intra-modal consistency preservation strategies to reduce the gap between modalities and improve the discriminability of learned Hamming space. Extensive experiments on four benchmark datasets demonstrate that RSDDH significantly outperforms state-of-the-art cross-modal hashing methods.

Detecting industrial defects presents formidable challenges, as anomalies often appear as tiny, low-contrast objects on complex surfaces with varied orientations. This work introduces two lightweight attention modules that strengthen feature discrimination while incurring minimal computational overhead. Our lightweight anisotropic spatial-channel attention (LASCA) module captures directional texture cues using bidirectional depthwise-separable convolutions and fuses spatial-channel information via a soft-gated mechanism. We further propose the Adaptive Energy-Guided Interaction Stream (AEGIS Stream), a novel feature enhancement block designed for high-precision small-object detection. Unlike conventional attention mechanisms, the AEGIS Stream employs a parameter-free energy-guided slicing strategy to capture fine-grained local saliency, coupled with a global interaction stream that aggregates cross-spatial context. Integrating this module into the C2f structure forms the AEGIS-C2f module, which enables robust feature fusion in complex industrial backgrounds. On the GC10-DET benchmark, our model reaches 68.9% mAP ₅₀ , a +1.7 percentage-point gain over YOLOv8n, with only 3.01 M parameters and 8.1 GFLOPs. Substantial improvements are also recorded on NEU-DET (+5.3 pp mAP ₅₀ ), DeepPCB (+0.3 pp), and VOC2007 (+0.6 pp). Ablation studies confirm that LASCA and AEGIS deliver complementary benefits. The resulting framework supports real-time, high-precision defect inspection with negligible added cost.