To address the challenge of balancing multiscale object recognition accuracy and computational efficiency for real-time detection tasks on unmanned aerial vehicle (UAV) edge-computing platforms, we propose an edge-attention synergy network (EASNet). First, a spatial pyramid information architecture is constructed to enable collaboration perception of local texture features and global semantic information through a cross-layer feature interaction mechanism. Second, a channel-adaptive dynamic weighting module is designed, leveraging a differential gating strategy to resolve semantic conflicts in multiresolution feature fusion. Finally, we propose a hybrid metric loss with learnable normalization. It dynamically adjusts Wasserstein scale sensitivity and incorporates fixed-weight IoU constraints, boosting small-target localization in complex backgrounds. The experimental results demonstrate that our method achieves mAP50 scores of 95.6%, 93.4%, 78.9%, 58.5%, and 46.5% on the RSOD, NWPU VHR-10, DIOR, VEDAI, and VisDrone datasets, respectively, with improvements ranging from 4.2% to 8.3% over the YOLOv10s baseline. Furthermore, through embedded deployment validation, we successfully implemented the model on a Jetson Nano edge-computing platform, attaining 34 FPS for real-time vehicle and pedestrian detection in campus monitoring scenarios.

Quantitative Phase Imaging (QPI) has been shown to complement established fluorescence microscopy as well as objective measurements of morphology and dynamics for cellular tissue studies. However, due to its inherent weak-signal measurements, the coherence of the laser light source, the roughness of the object under test or the complex scattering environment, QPI exhibits various types of complex noise, with Poisson-Gaussian noise and scattering noise being the main noise sources. In recent years, significant advancements have been made in the field of deep learning-based denoising algorithms, which have shown considerable efficacy in the denoising of individual noisy data. However, these algorithms have been observed to be less effective when confronted with other types of noisy data, and lack a unified model that can simultaneously remove complex noise from QPI. The present study proposes a QPI denoising approach based on the denoising diffusion probability model (DDPM). The denoising process of DDPM is comprised of two constituent parts: the forward process, which gradually adds standard Gaussian a priori noise to the original image until the image is completely random; and the reverse diffusion chain, which gradually recovers an undisturbed ‘clean’ image by inference from a given a priori noise, thus eliminating various types of complex noise in QPI. A comparative analysis was conducted between the conventional denoising approach based on BM3D and the deep network denoising algorithm with U-Net as the backbone, and the proposed method was evaluated through experimental validation using simulated Gaussian noise, scattering noise and fluorescence microscopy dataset (FMD). The experimental results demonstrate the superior denoising, detail restoration and generalization performance of the proposed method, signifying its significant potential for practical applications.

Saliency object detection has been widely used in computer vision tasks such as image understanding, semantic segmentation, and target tracking by mimicking the human visual perceptual system to find the most visually appealing object. The U2Net model has shown good performance in salient object detection (SOD) because of its unique U‐shaped residual structure and the U‐shaped structural backbone incorporating feature information of different scales. However, in the U‐shaped structure, the global semantic information computed from the topmost layer may be gradually interfered by the large amount of local information dilution in the top‐down path, and the U‐shaped residual structure has insufficient attention to the features in the salient target region of the image and will pass redundant features to the next stage. To address these two shortcomings in the U2Net model, this paper proposes improvements in two aspects: to address the situation that the global semantic information is diluted by local semantic information and the residual U‐block (RSU) module pays insufficient attention to the salient regions and redundant features. An attentional gating mechanism is added to filter redundant features in the U‐structure backbone. A channel attention (CA) mechanism is introduced to capture important features in the RSU module. The experimental results prove that the method proposed in this paper has higher accuracy compared to the U2Net model.

The detection of small infrared targets is still a challenging task and efficient and accurate detection plays a key role in modern infrared search and tracking military applications. However, small infrared targets are difficult to detect due to their weak brightness, small size and lack of shape, structure, texture, and other information elements. In this paper, we propose a target detection method. First, to address the problem that the proximity of targets to high‐brightness clutter leads to missed detection of candidate targets, a Gaussian differential filtering preprocessed image is used to suppress high‐brightness clutter. Second, a density‐peaked global search method is used to determine the location of candidate targets in the preprocessed image. We then use local contrast to the candidate target points to enhance the gradient features and suppress background clutter. The Facet model is used to compute multidirectional gradient features at each point. A new efficient surrounding symmetric region partitioning scheme is constructed to capture the gradient characteristics of targets of different sizes in eight directions, followed by weighting the candidate target gradient characteristics using the standard deviation of the symmetric region difference. Finally, an adaptive threshold segmentation method is used to extract small targets. Experimental results show that the method proposed in this paper has better detection accuracy and robustness compared with other detection methods.

Distributed fiber optic vibration sensing system has been widely used in safety monitoring with distinct advantages, and the feature extraction and classification methods of fiber optic signals directly determine the real-time performance and reliability of the monitoring system. The existing research feature extraction methods are unidimensional and time consuming, which cannot balance the goals of high accuracy and low time consumption for safety monitoring systems. In this paper, we propose an efficient recognition framework for fusing signal time-frequency features, called TFF-CNN, based on Gramian Angular Difference Fields(GADF) and FFT co-generation matrix(FFTT) to manually extract two-dimensional time-frequency image features of fiber optic signals, taking the significant advantages of CNN in image processing and combining a two-channel model and a fusion module that simulates human decision-making behavior. The accuracy of TFF-CNN is 99.30% and the detection response time is only 0.6s. Compared with other methods in this field, TFF-CNN has the advantages of low false alarm rate and short time consumption, which is more suitable for deployment in security monitoring field with distributed fiber optic sensing system. Index Terms: Distributed optical fiber vibration sensing system(DVS), Two-dimensional multi-feature, CNN, intrusion detection, real-time monitoring