Adverse conditions such as intense illumination and inclement weather pose challenges to object detection tasks. Current methodologies aim to enhance detector performance by improving image quality through various means. In low-visibility agricultural scene object detection, improving detection accuracy is insufficient with image quality enhancement alone. Existing detectors designed for low-visibility conditions do not exhibit strong performance in agricultural scenes. In this paper, we introduce a novel benchmark called Low-visibility Crop Detection (LVCD), extending the low-visibility task to agricultural scenes. Furthermore, we propose a foundational framework, Point cloud Guided Segmentation Network (PGSNet), which learns additional point cloud cues to compensate for the missing details of targets caused by low-visibility in images, extending the model’s representational capacity to two modalities. To facilitate research, we gather a multi-modal dataset, LVScene4K, comprising images and corresponding point clouds of various crop types. In order to address the challenges posed by low-visibility, we specifically designed corresponding modules. The encoder component encodes features at multiple scales and constructs multi-scale receptive fields to enable the model to simultaneously extract more detailed features from both images and point clouds. The decoder analyzes both local and global details as well as texture features of the target, while adaptively capturing characteristics under various adverse conditions to improve the model’s discriminative performance under low-visibility conditions. In the end, the model applies iterative refinement strategy for progressively optimizing the detection outcomes. Comprehensive experiments conducted on the LVScene4K demonstrate the effectiveness and robustness of PGSNet in the LVCD task.

In recent years, adversarial attacks on video recognition models have attracted increasing attention. However, most existing strategies are extensions of image-based methods, where adversarial perturbations are computed independently and embedded into individual frames. This independent per-frame perturbation process wastes computational resources and leads to excessive query consumption. To address this problem, we introduce Frequency Domain Distributed Perturbations (FDP), a straightforward yet effective black-box video attack method using temporal correlations between video frames. Specifically, FDP first converts the input video into the frequency domain and calculates globally coordinated adversarial perturbations in the spectral space. By conducting global optimization in the frequency domain, FDP improves the effectiveness of each query, significantly decreasing the total number of queries needed. The resulting perturbations are temporally distributed across frames to preserve the spatiotemporal structure. Furthermore, we introduce a frequency-sensitive mask to identify the spectral regions most critical to the model's predictions. By applying perturbations only to these key frequency bands, FDP further reduces the perturbation search space and improves query efficiency. Extensive experiments demonstrate that our method significantly reduces query consumption while achieving higher attack success rates than state-of-the-art approaches.