The increasing deployment of digital healthcare systems has led to the continuous transmission of highly sensitive patient data, raising urgent concerns about data leakage in high-noise, high-loss, and dynamically changing. Existing privacy-preservation techniques often struggle to provide robustness, low overhead, and real-time responsiveness under high jitter and packet loss, limiting their effectiveness in rapid detection and accurate tracing of leaks. To address these challenges, we propose a Reinforcement Learning-Driven Joint Watermarking Framework (RLDJ-W). First, it utilizes a reinforcement learning strategy to adaptively modulate the watermark embedding interval, ensuring both invisibility and enhancing the watermark’s survivability in harsh channels. Then, it leverages Bi-LSTM to capture and model multi-granularity time-series features of network flows, thereby dynamically evaluating the invisibility of the watermark flows. Finally, a high-performance decoding network based on MLP is designed to achieve efficient and accurate watermark information extraction. Experimental results demonstrate that the watermarking capacity of RLDJ-W achieves 2.25 bit/s, requiring only an average of 5.88 packets per bit of watermark. It also maintains over 85% detection accuracy even under 100ms delay jitter and 40% packet loss, consistently outperforming state-of-the-art baselines.

Edge computing (EC), as a computing paradigm that mitigates cloud load and reduces task latency, has attracted widespread attention from both academia and industry. Current research on EC primarily focuses on edge task offloading problems, while effectively matching tasks with microservices after offloading is also crucial for reliable task processing. Therefore, considering the differentiated hardware resource requirements of various task types, this paper designs a heterogeneous computing model that enables precise matching between tasks and edge server (ES) computational capabilities. To ensure ESs proactively deploy effective microservices and maintain trustworthy operations, we introduce an incentive mechanism and an ES discriminator algorithm. Considering diverse demands in EC scenarios, where ESs pursue higher incentive returns while reducing energy consumption, and the system aims to minimize latency and maintain reliability under limited incentive budgets, we construct an interconnected satisfaction model between ESs and the system. Based on this, we propose a bilateral consensus service placement (BCSP) algorithm that balances incentive consensus between ESs and the system, achieving rational microservice deployment and optimized task processing efficiency. Experimental results show that compared with existing algorithms, the proposed BCSP algorithm better accommodates multi-party requirements and enhances both efficiency and reliability in microservice placement.

The exponential growth of sensitive patient information and diagnostic records in digital healthcare systems has increased the complexity of data protection, while frequent medical data breaches severely compromise system security and reliability. Existing privacy protection techniques often lack robustness and real-time capabilities in high-noise, high-packet-loss, and dynamic network environments, limiting their effectiveness in detecting healthcare data leaks. To address these challenges, we propose a Swarm Intelligence-Based Network Watermarking (SIBW) method for real-time privacy data leakage detection in digital healthcare systems. SIBW integrates fountain codes with outer error correction codes and employs a Multi-Phase Synergistic Swarm Optimization Algorithm (MPSSOA) to dynamically optimize encoding parameters, significantly enhancing the robustness and interference resistance of watermark detection. Additionally, a reliable synchronization sequence and lightweight embedding mechanism are designed to ensure adaptability to complex, dynamic networks. Experimental results demonstrate that SIBW achieves over 90% detection accuracy under high latency jitter and packet loss conditions, surpassing existing methods in both robustness and efficiency. With a compact design of only 3.7 MB, SIBW is particularly suited for rapid deployment in resource-constrained digital healthcare systems.

Deep learning has significantly advanced medical image processing, yet the inherent inclusion of personally identifiable information (PII) within medical images-such as facial features, distinctive anatomical structures, rare lesions, or specific textural patterns-poses a critical risk to patient privacy during data transmission. To mitigate this risk, we introduce the Medical Semantic Diffusion Model (MSDM), a novel framework designed to synthesize medical images guided by semantic information, synthesis images with the same distribution as the original data, which effectively removes the PPI of the original data to ensure robust privacy protection. Unlike conventional techniques that combine semantic and noisy images for denoising, MSDM integrates Adaptive Batch Normalization (AdaBN) to encode semantic information into high-dimensional latent space, embedding it directly within the denoising neural network. This approach enhances image quality and semantic accuracy while ensuring that the synthetic and original images belong to the same distribution. In addition, to further accelerate synthesis and reduce dependency on manually crafted semantic masks, we propose the Spread Algorithm, which automatically generates these masks. Extensive experiments conducted on the BraTS 2021, MSD Lung, DSB18, and FIVES datasets confirm the efficacy of MSDM, yielding state-of-the-art results across several performance metrics. Augmenting datasets with MSDM-generated images in nnUNet segmentation experiments led to Dice scores of 0.6243, 0.9531, 0.9406, and 0.9562 underscoring its potential for enhancing both image quality and privacy-preserving data augmentation.

The real-time processing of large-scale, heterogeneous tasks—including CPU-only, general-purpose GPU, and specialized GPU tasks—poses significant challenges in Internet of Things (IoT) systems, driven by severe GPU resource fragmentation, inefficient CPU and memory resource utilization on edge servers. These issues often compromise system processing performance and server stability. To address these issues, we formulate a multi-stage mixed-integer nonlinear programming (MINLP) model, to jointly optimize GPU fragmentation rate and system processing capability. We then introduce a novel deviation-based Lyapunov optimization framework that explicitly maintains memory utilization around a predefined optimal threshold, effectively balancing resource usage and system stability. Finally, to achieve real-time decision-making for massive tasks in dynamic systems with randomly arriving tasks, we propose the MA-LHTO algorithm, a multi-agent deep reinforcement learning approach that incorporates a multi-head architecture, entropy-based exploration, and a parameter reset mechanism. Experimental results confirm that our algorithm significantly improves resource utilization, and exhibits good performance under various working conditions.

Medical image segmentation is a crucial technology for advancing smart healthcare, yet its performance is hampered by limited model generalization due to scarce annotated data. Current solutions face significant challenges: transfer learning from natural images struggles to adapt effectively to medical image feature distributions owing to domain gaps, while mainstream semi-supervised co-training frameworks (e.g., Cross Pseudo Supervision, CPS) suffer from premature consensus among sub-networks, diminishing the utility of unlabeled data. To address this, we propose ReConCPS, an enhanced co-training framework integrating feature reconstruction and a dual perturbation mechanism. First, to mitigate transfer learning difficulties, we introduce an unsupervised feature reconstruction task as an auxiliary branch in the pre-trained encoder, leveraging unlabeled data to guide the learning of more discriminative medical image features. Second, targeting CPS's premature convergence, we devise a task perturbation strategy: the main network concurrently performs segmentation and reconstruction, while the auxiliary network focuses solely on segmentation, thereby establishing behavioral discrepancy through objective divergence. Finally, we implement a dual perturbation mechanism-applying feature-space perturbations to enhance robustness and employing heterogeneous encoder architectures (structural perturbation) for the two sub-networks-collectively promoting diverse feature learning. Experiments on public medical image datasets demonstrate the superiority of ReConCPS in addressing annotation scarcity and enhancing model generalization.

Electrocardiogram (ECG) signals play a critical role in the clinical diagnosis of arrhythmias. However, their non-stationary nature, multi-lead configuration, and multi-label annotations pose significant challenges for feature extraction and discriminative modeling. Existing approaches often rely on singlescale features or neglect the semantic dependencies among labels, limiting their ability to capture the intricate time-frequency patterns and label correlations inherent in ECG data. To address these issues, we propose a novel multi-label ECG classification framework, MTA-Net, which enhances the modeling of complex pathological features from both time-frequency and semantic perspectives. Specifically, MTA-Net leverages the Discrete Wavelet Packet Transform to extract high-resolution, multi-scale timefrequency representations, significantly improving the sensitivity to localized rhythm abnormalities. A label embedding module is introduced to incorporate label information into the feature interaction process, establishing explicit semantic connections between the samples and labels. Furthermore, a dual-branch masked encoder is designed to separately model structural and semantic representations of ECG signals. Finally, a labelguided decoder integrates these structural and semantic cues to produce accurate multi-label predictions. Extensive experiments conducted on two public multi-label ECG datasets, PTB-XL and CPSC2018, demonstrate that MTA-Net outperforms existing state-of-the-art methods in terms of AUC, F1-score, and accuracy, validating its effectiveness in handling complex multi-label ECG classification tasks.

KAN-based networks, while offering improved interpretability compared to traditional models used in medical image segmentation, often struggle with limited adaptability to diverse imaging environments, making them less ideal for such tasks. To address this issue, we propose a Fourier Kolmogorov–Arnold–based (FKAN) Group Multi–scale Fusion Network, termed FKAN–GMFNet, which incorporates an FKAN layer into a labeled intermediate representation, introducing an Enhanced–FKAN block. Furthermore, we develop the Attention Group Multi–scale Aggregation (ATGMA) module, which leverages attention mechanisms and grouping strategies to effectively fuse feature masks with both high– and low–scale feature information, thereby achieving a comprehensive multi-scale feature representation. Extensive experiments demonstrate that the FKAN GMFNet significantly outperforms seven state–of–the–art methods in both Dice and IoU scores, where the Dice and IoU scores for the IAS–L dataset are 88.82% and 80.09%, respectively. Code is available at https://github.com/zx123868/FKAN-GMFNet.

Object detection technology plays an essential role in Unmanned Aerial Vehicle (UAV) sea search and rescue missions, which aim to quickly locate crucial targets such as trapped people and ships at sea in the complex marine environment. However, due to the restricted view angle of the UAV and the specificity of the working environment, the remote sensing images captured in the marine environment are characterized by small targets and considerable interference on the sea surface, which brings significant challenges to the UAV sea rescue mission. To address this issue, we propose an object detection model based on the Transformer architecture in this paper. The model takes FCDS-DETR as the baseline and introduces a two-dimensional Gaussian probability density distribution as an additional attention mechanism to speed up the model’s location of suspicious targets in the image to improve the detection accuracy of the model for targets. At the same time, the denoising training method is introduced in the model’s training process to stabilize the bipartite graph matching and promote the convergence of the model. On the SDS ODv2 and AFO object detection datasets, our model achieves an average precision of 51% and 54.5%, respectively, which is an improvement of 5.7% and 4.4% compared to the baseline model’s performance on these datasets.

With higher peak data rates, enhanced reliability, improved energy efficiency, and reduced radio latency, 6G enables cooperative perception in autonomous vehicle clusters. Existing research mainly focuses on establishing communication-connected and structurally stable clusters, while overlooking how members collaborate in perception. To address this gap, we propose a collaborative perception-based autonomous vehicle cluster modeling method for expressways, leveraging the capabilities of 6G networks. This method facilitates collaborative perception within vehicle clusters through the exchange of sensory information among member vehicles. First, we introduce a perception interaction mechanism among vehicles as a foundation for constructing clusters. We then present a primary vehicle selection method and analyze the cluster’s perception gain, collaborative efficiency, and collaborative reliability. Based on this, we develop a vehicle cluster model and solve it using a genetic algorithm. We present a vehicle cluster formation method that brings together vehicles achieving Pareto optimal solutions through specific interaction rules, thereby forming a cluster. The simulation results demonstrate that the proposed method outperforms existing methods in terms of perception gain (PG), collaborative efficiency (CE), and collaborative reliability (CR).

As smart cities evolve, rising computational demands strain infrastructures. Offloading tasks to edge cloud data centers offers potential but faces challenges like high latency, energy use, and data leakage, especially in dense urban areas. This paper presents a low-latency, energy-efficient digital twin (DT) architecture tailored for smart cities, integrating edge computing (EC) and multiple s (IRS) to enhance communication. Dynamic voltage and frequency scaling (DVFS) technology is considered for user devices to reduce energy consumption. To mitigate the risk of user privacy leakage during task offloading, we address sensitive user location data that may be exposed by proposing a perturbed sliding task queue (PSTQ) algorithm based on differential privacy (DP), and demonstrate the effectiveness of the algorithm. To optimize task processing time and energy efficiency, we decompose the complex problem using block coordinate descent and propose an intelligent scheduling for energy sustainability (ISES) algorithm based on Karush-Kuhn-Tucker conditions and deep reinforcement learning (DRL). Experimental results demonstrate that our proposed architecture and algorithms achieve over 90% improvement in key optimization objectives, alleviating the computational pressure on existing devices while significantly enhancing task processing efficiency and energy sustainability.

Cancer diagnosis with incomplete multi-omics data is inevitable since there are widely missing arbitrary omics in real-world applications. Currently, although much progress has been made in handling missing data, existing methods for cancer diagnosis still struggle to obtain credible predictions due to the relatively high uncertainty of missing omics data. In this paper, we propose a trusted and attention-based multi-omics integrated framework for cancer diagnosis using incomplete multi-omics data (TAIMONET). For the raw omics data, the framework first builds an attention-based shared encoder to pick out important features and align different omics data. Next, it utilizes generative adversarial learning to impute the missing omics features based on the extracted available omics features. Meanwhile, different reconstructed complete omics data are weighted and fused according to the attention mechanism. Moreover, it uses true-class-probability to calibrate the classification results and improve the reliability of the results. Extensive experiments on four real-world multimodal medical datasets are conducted. Compared to state-of-the-art methods, the superior performance and trustworthiness of our proposed model are clearly validated.

In the rapidly advancing landscape of intelligent medical technology, high-precision and widely applicable medical image segmentation techniques are pivotal in advancing personalized treatment plans and enhancing patient experiences. To address inherent obstacles, such as complex structures and features that are difficult to capture in segmentation tasks, we first propose a CMH module that merges convolutional neural networks (CNNs) with state space models (e.g., Mamba). This module hierarchically integrates CNN’s local priors into the Mamba layer, adeptly capturing specific details of target objects and broader contextual information from different perspectives. Subsequently, we introduce an information fusion module based on Mamba (MIH), aimed at dynamically fusing feature information from different paths and levels to enhance feature expression and discrimination. Building upon the abovementioned modules, we develop an unexplored segmentation model, CMHNet, capable of flexibly capturing feature maps of varying scales and angles from medical images. In the experiments, we present CMHNet’s distinguished performance in segmentation tasks through massive experiments on four medical datasets.

In medical imaging, segmentation of tissues and organs helps doctors accurately determine disease subtypes and propose treatment plans. Few-shot learning, which can be trained with a small amount of annotated data and has better generalization, has achieved excellent results in medical image segmentation, where large-scale annotated data is scarce. However, most current few-shot segmentation models rely on the foreground part of support features, establishing spatial information connections with query features through the foreground part and ignoring the differences in foreground classes between query and support images. Therefore, we propose a novel few-shot segmentation model based on cross-attention: aligned patch calibration attention network (APCANet), which filters out pixels detrimental to segmentation based on cross-attention between support features and query features. Specifically, our model is patch-based, setting the query patches as Query and the aligned support patches as Key and Value for cross-attention. Then, in the cross-attention matrix, by distinguishing the tokens in the patches, we set the weights of non-compliant tokens in the attention matrix to negative infinity, thereby calibrating the harmful pixels in the support features. Additionally, we generate a prior mask for the query image and concatenate it with the query features. Subsequently, we divide the support and query features into patches and align foreground and background patches based on similarity. These operations can better establish spatial correlation between support and query features. The experimental results on the Abd-CT, Card-MRI, and Abd-MRI datasets demonstrate the effectiveness of our method.Code is released at: https://github.com/HengTaoDing/APCANet

Echocardiography is essential for evaluating cardiac anatomy and function during early recognition and screening for congenital heart disease (CHD), a widespread and complex congenital malformation. However, fetal CHD recognition still faces many difficulties due to instinctive fetal movements, artifacts in ultrasound images, and distinctive fetal cardiac structures. These factors hinder capturing robust and discriminative representations from ultrasound images, resulting in CHD's low prenatal detection rate. Hence, we propose a multi-scale gated axial-transformer network (MSGATNet) to capture fetal four-chamber semantic information. Then, we propose a SPReCHD: four-chamber semantic parsing network for recognizing fetal CHD in the clinical treatment of the medical metaverse, integrating MSGATNet to segment and locate four-chamber arbitrary contours, further capturing distinguished representations for the fetal heart. Comprehensive experiments indicate that our SPReCHD is sufficient in recognizing fetal CHD, achieving a precision of 95.92%, a recall of 94%, an accuracy of 95%, and a$F_{1}$score of 94.95% on the test set, dramatically improving the fetal CHD's prenatal detection rate.

There exists growing evidence that circRNAs are concerned with many complex diseases physiological processes and pathogenesis and may serve as critical therapeutic targets. Identifying disease-associated circRNAs through biological experiments is time-consuming, and designing an intelligent, precise calculation model is essential. Recently, many models based on graph technology have been proposed to predict circRNA-disease association. However, most existing methods only capture the neighborhood topology of the association network and ignore the complex semantic information. Therefore, we propose a Dual-view Edge and Topology Hybrid Attention model for predicting CircRNA-Disease Associations (DETHACDA), effectively capturing the neighborhood topology and various semantics of circRNA and disease nodes in a heterogeneous network. The 5-fold cross-validation experiments on circRNADisease indicate that the proposed DETHACDA achieves the area under receiver operating characteristic curve of 0.9882, better than four state-of-the-art calculation methods.

With the evolution of 6G technology, mobile edge computing is rapidly advancing as a crucial application scenario. This research presents an innovative method for vehicle-edge task offloading decision-making, leveraging real-time data inputs such as channel conditions, image entropy, and detector confidence levels. We propose a collaborative task processing framework for vehicle-edge computing that effectively combines lightweight and heavyweight models to cater to varying demands, ensuring efficient task execution. Additionally, the study introduces a custom-designed reinforcement learning algorithm aimed explicitly at optimizing offloading scheduling. This algorithm boosts decision-making accuracy and efficiency and features a comprehensive reward system to achieve a balanced trade-off between detection performance and latency. The frameworks efficacy is thoroughly evaluated in complex driving scenarios using the SODA10M dataset. Our results indicate the frameworks capability to achieve convergence, enhance precision, ensure stability, and maintain a lightweight operation, emphasizing its suitability for real-world implementation. This work provides practical and efficient strategies for intelligent driving task scheduling to meet the requirements of contemporary dynamic environments.

Numerous biological studies have shown that considering disease-associated micro RNAs (miRNAs) as potential biomarkers or therapeutic targets offers new avenues for the diagnosis of complex diseases. Computational methods have gradually been introduced to reveal disease-related miRNAs. Considering that previous models have not fused sufficiently diverse similarities, that their inappropriate fusion methods may lead to poor quality of the comprehensive similarity network and that their results are often limited by insufficiently known associations, we propose a computational model called Generative Adversarial Matrix Completion Network based on Multi-source Data Fusion (GAMCNMDF) for miRNA-disease association prediction. We create a diverse network connecting miRNAs and diseases, which is then represented using a matrix. The main task of GAMCNMDF is to complete the matrix and obtain the predicted results. The main innovations of GAMCNMDF are reflected in two aspects: GAMCNMDF integrates diverse data sources and employs a nonlinear fusion approach to update the similarity networks of miRNAs and diseases. Also, some additional information is provided to GAMCNMDF in the form of a 'hint' so that GAMCNMDF can work successfully even when complete data are not available. Compared with other methods, the outcomes of 10-fold cross-validation on two distinct databases validate the superior performance of GAMCNMDF with statistically significant results. It is worth mentioning that we apply GAMCNMDF in the identification of underlying small molecule-related miRNAs, yielding outstanding performance results in this specific domain. In addition, two case studies about two important neoplasms show that GAMCNMDF is a promising prediction method.

Four-chamber (FC) views are the primary ultrasound(US) images that cardiologists diagnose whether the fetus has congenital heart disease (CHD) in prenatal diagnosis and screening. FC views intuitively depict the developmental morphology of the fetal heart. Early diagnosis of fetal CHD has always been the focus and difficulty of prenatal screening. Furthermore, deep learning technology has achieved great success in medical image analysis. Hence, applying deep learning technology in the early screening of fetal CHD helps improve diagnostic accuracy. However, the lack of large-scale and high-quality fetal FC views brings incredible difficulties to deep learning models or cardiologists. Hence, we propose a Pseudo-Siamese Feature Fusion Generative Adversarial Network (PSFFGAN), synthesizing high-quality fetal FC views using FC sketch images. In addition, we propose a novel Triplet Generative Adversarial Loss Function (TGALF), which optimizes PSFFGAN to fully extract the cardiac anatomical structure information provided by FC sketch images to synthesize the corresponding fetal FC views with speckle noises, artifacts, and other ultrasonic characteristics. The experimental results show that the fetal FC views synthesized by our proposed PSFFGAN have the best objective evaluation values: SSIM of 0.4627, MS-SSIM of 0.6224, and FID of 83.92, respectively. More importantly, two professional cardiologists evaluate healthy FC views and CHD FC views synthesized by our PSFFGAN, giving a subjective score that the average qualified rate is 82% and 79%, respectively, which further proves the effectiveness of the PSFFGAN.

MicroRNAs (miRNAs) influence several biological processes involved in human disease. Biological experiments for verifying the association between miRNA and disease are always costly in terms of both money and time. Although numerous biological experiments have identified multi-types of associations between miRNAs and diseases, existing computational methods are unable to sufficiently mine the knowledge in these associations to predict unknown associations. In this study, we innovatively propose a heterogeneous graph attention network model based on meta-subgraphs (MSHGANMDA) to predict the potential miRNA-disease associations. Firstly, we define five types of meta-subgraph from the known miRNA-disease associations. Then, we use meta-subgraph attention and meta-subgraph semantic attention to extract features of miRNA-disease pairs within and between these five meta-subgraphs, respectively. Finally, we apply a fully-connected layer (FCL) to predict the scores of unknown miRNA-disease associations and cross-entropy loss to train our model end-to-end. To evaluate the effectiveness of MSHGANMDA, we apply five-fold cross-validation to calculate the mean values of evaluation metrics Accuracy, Precision, Recall, and F1-score as 0.8595, 0.8601, 0.8596, and 0.8595, respectively. Experiments show that our model, which primarily utilizes multi-types of miRNA-disease association data, gets the greatest ROC-AUC value of 0.934 when compared to other state-of-the-art approaches. Furthermore, through case studies, we further confirm the effectiveness of MSHGANMDA in predicting unknown diseases.

At present, solid-state fermentation (SSF) is mainly controlled by artificial experience, and the product quality and yield are not stable. Therefore, predicting the quality and yield of SSF is of great significance for improving the utility of SSF. In this article, we propose a deep learning utility prediction (DLUP) scheme for the SSF in the Industrial Internet of Things, including parameters collection and utility prediction of the SSF process. Furthermore, we propose a novel edge-rewritable Petri net to model the parameters collection and utility prediction of the SSF process and further verify their soundness. More importantly, DLUP combines the generating ability of least squares generative adversarial network with the predicting ability of fully connected neural network to realize the utility prediction (usually use the alcohol concentration) of SSF. Experiments show that the proposed method predicts the alcohol concentration more accurately than the other joint prediction methods. In addition, the method in our article provides evidences for setting the ratio of raw materials and proper temperature through numerical analysis.

Cloud manufacturing (CMfg) platform consists of the cloud services, manufacturing technology, and the Internet of Things, which provides solutions for large-scale personalized customization through the service model. However, the service flexibility and resource allocation of CMfg are two factors that restrict the production time and cost of CMfg. A CMfg service model based on rewritable Petri nets (RPNs) is established, where the reconfiguration process of personalized customization is described by the rewritable rules of RPN. On this basis, the performance of the reconfiguration of the personalized customization service process is analyzed (this model analysis method can analyze the soundness of the reconfiguration process). In addition, we establish the resource allocation strategy of CMfg based on nondominated sorting genetic algorithm to obtain the best personalized customization scheme in terms of time and cost. The results of simulation and comparison experiments show that the method proposed in this article can obtain the optimal solution for both production time and cost.

Fetal congenital heart disease (CHD) is the most common type of fatal congenital malformation. Fetal four-chamber (FC) view is a significant and easily accessible ultrasound (US) image among fetal echocardiography images. Automatic detection of four fetal heart chambers considerably contributes to the early diagnosis of fetal CHD. Furthermore, robust and discriminative features are essential for detecting crucial visualizing medical images, especially fetal FC views. However, it is an incredibly challenging task due to several key factors, such as numerous speckles in US images, the fetal four chambers with small size and unfixed positions, and category confusion caused by the similarity of cardiac chambers. These factors hinder the process of capturing robust and discriminative features, hence destroying the fetal four chambers’ precise detection. Therefore, we propose an intelligent feature learning detection system (FLDS) for FC views to detect the four chambers. A multistage residual hybrid attention module (MRHAM) presented in this paper is incorporated in the FLDS for learning powerful and robust features, helping FLDS accurately locate the four chambers in the fetal FC views. Extensive experiments demonstrate that our proposed FLDS outperforms the current state-of-the-art, including the precision of 0.919, the recall of 0.971, the$F_{1}$score of 0.944, the mAP of 0.953, and the frames per second (FPS) of 43. In addition, our proposed FLDS is also validated on other visualizing nature images such as the PASCAL VOC dataset, achieving a higher mAP of 0.878 while input size is 608 × 608.